<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Biao Hao on Medium]]></title>
        <description><![CDATA[Stories by Biao Hao on Medium]]></description>
        <link>https://medium.com/@biaohao?source=rss-cf1dc593d6bd------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*iAgnFl-t9UWYh2GwDffJ_g.png</url>
            <title>Stories by Biao Hao on Medium</title>
            <link>https://medium.com/@biaohao?source=rss-cf1dc593d6bd------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 13:48:43 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@biaohao/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Agentic AI and Payments, Part 4: A Traditional Finance Path to Machine-to-Machine Payments]]></title>
            <link>https://biaohao.medium.com/agentic-ai-and-payments-part-4-a-traditional-finance-path-to-machine-to-machine-payments-0bbb7d93432c?source=rss-cf1dc593d6bd------2</link>
            <guid isPermaLink="false">https://medium.com/p/0bbb7d93432c</guid>
            <category><![CDATA[cryptocurrency]]></category>
            <category><![CDATA[architecture]]></category>
            <category><![CDATA[payments]]></category>
            <category><![CDATA[finance]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <dc:creator><![CDATA[Biao Hao]]></dc:creator>
            <pubDate>Mon, 13 Apr 2026 17:50:50 GMT</pubDate>
            <atom:updated>2026-04-13T20:50:00.472Z</atom:updated>
            <content:encoded><![CDATA[<p>In <a href="https://biaohao.medium.com/agentic-ai-and-payments-f2d7548977f3">Part 1</a>, I framed the core question: if software is starting to participate in commerce, how should it discover services, access value, make payments, and operate safely? <a href="https://biaohao.medium.com/agentic-ai-and-payments-part-2-the-crypto-native-answer-d56e5214cc16">Part 2</a> looked at the crypto-native answer, especially wallets, stablecoins, and x402. <a href="https://biaohao.medium.com/agentic-ai-and-payments-part-3-the-traditional-finance-answer-bac72cc45714">Part 3</a> looked at the traditional finance answer, where banks, card networks, and payment rails are adapting their existing stack so software can act as a trusted delegate of a person or business.</p><p>This final part takes the next step.</p><p>It argues that traditional finance can support <strong>machine-to-machine payments</strong> too — but not by forcing every tiny machine event to look like a normal card purchase or bank transfer. The more plausible path is a layered one: use a machine-friendly payment protocol at the application layer, combine it with traditional controls such as virtual accounts or delegated credentials, and add a micropayment ledger so low-value machine activity can be authorized in real time but settled efficiently over existing rails. Stripe’s Machine Payments Protocol (MPP) already points in that direction by supporting both crypto and fiat-linked flows inside the same machine-payment framework.</p><h3>The Gap: Where Traditional Finance Still Falls Short</h3><p>Traditional finance has made real progress on agentic checkout and delegated spending. Visa and Mastercard are extending existing commerce flows so agents can act with constrained credentials, authenticated user intent, and strong controls. Virtual accounts and virtual cards provide agents potential useful operating structures for consumers and businesses. FedNow and RTP give banks real-time payment rails.</p><p>But there is still a gap.</p><p>Most traditional rails were designed for human or business payment events, not for <strong>high-frequency, very low-value, machine-generated transactions</strong>. A card purchase, a bank transfer, or even an instant payment is usually treated as a complete financial event in itself. Machine commerce may generate far more of these events, often at much smaller values. That creates an economic and operational challenge even when the rails themselves are fast.</p><p>Zelle helps show the distinction. It can support small, fast transfers between known parties, and consumers typically do not pay fees to send or receive money. But it is framed around payments to people and businesses you know and trust, not open programmable machine commerce, and it does not offer purchase protection for authorized payments. That makes it useful in some closed-loop cases, but not a strong base layer for generalized machine-to-machine micropayments.</p><p>So the issue is not only identity or authorization, but also <strong>micropayment capability</strong>. Traditional finance is already good at trusted checkout, delegated spend, and real-time account-based payments. The harder question is whether it can also support a more machine-native payment model.</p><p>To answer that, it helps to first look at the emerging protocol landscape for agentic payments.</p><h3>The Broader Protocol Landscape</h3><p>The protocol landscape is getting more crowded, but the different standards are not trying to solve exactly the same problem.</p><p><strong>x402</strong> is still the clearest crypto-first model for machine payments. It uses HTTP 402 “Payment Required” to let a service ask for payment and let software pay programmatically for APIs and digital resources. <strong>MPP</strong>, the Machine Payments Protocol from Stripe and Tempo, is the closest architectural cousin. It uses the same HTTP 402 challenge-response pattern, but Stripe positions it as supporting both crypto and fiat-linked payment flows, including microtransactions and recurring payments.</p><p><strong>ACP</strong>, the Agentic Commerce Protocol from OpenAI, sits higher in the stack. It is not mainly a micropayments rail. It is a merchant-commerce standard that lets agent platforms such as ChatGPT ingest catalog data, understand inventory, and complete commerce flows with merchants, while payments still run through the merchant’s existing payment provider. OpenAI’s Delegated Payment Spec makes that especially clear: merchants or their PSPs still process payment in the normal way.</p><p><strong>AP2</strong>, Google’s Agent Payments Protocol, is broader still. Google describes it as an open, shared protocol for secure, compliant transactions between agents and merchants, supporting cards, stablecoins, and real-time bank transfers. In other words, AP2 is less a new payment rail than a trust and interoperability layer across multiple rails.</p><p>Taken together, these standards point to a common direction: <strong>software needs a machine-friendly payment interaction model, even if the underlying settlement rails differ</strong>.</p><h3>A Proposed Architecture for Traditional Finance Machine Payments</h3><p>The protocol landscape suggests an important design principle: the payment interaction model and the settlement rail do not have to be the same thing.</p><p>That is what makes MPP especially interesting. Stripe describes it as a machine-friendly protocol built around an HTTP 402 challenge-response flow, while still supporting both crypto and fiat-linked payment paths. In other words, it points toward a hybrid model in which software uses a machine-native payment interaction, but settlement can still happen over more traditional infrastructure.</p><p>From that perspective, a traditional finance path to machine-to-machine payments does not require inventing an entirely new bank-only rail. It requires combining a machine-native payment layer with existing controls, operating-value structures, and settlement rails.</p><p>I would propose six layers. The key idea is: the machine-payment protocol handles fine-grained authorization and usage events; legacy rails, with extensions, carry the settlement outcome, not every micro-event.</p><figure><img alt="Layered Architecture" src="https://cdn-images-1.medium.com/max/1024/1*SA57OATZ5W4QFQ5GkDpaug.png" /><figcaption>Proposed layered architecture for machine-to-machine payments using an MPP-like protocol and delegated controls on top of the traditional finance stack</figcaption></figure><h4><strong>Machine Payment Protocol Layer</strong></h4><p>Use an <strong>MPP-like challenge-response flow</strong> at the application layer. A service returns 402 “Payment Required”, the agent receives the price and conditions, authorizes payment, and continues the workflow. This gives traditional finance a machine-native interaction model similar to x402 and MPP.</p><h4>Delegated Authorization Layer</h4><p>Bind the agent to a human or business principal using constrained credentials, spending policies, max amounts, merchant restrictions, expiry windows, and auditability.</p><h4>Micropayment Ledger / Aggregation Layer</h4><p>This is the missing piece. Do not send every sub-cent or few-cent machine event directly over cards, Zelle, FedNow, or RTP. Instead, authorize the payment in real time, record it in a machine-payment ledger, aggregate or net tiny events by counterparty and time window, and settle the net amount over a real rail periodically.</p><p>This is my architectural recommendation rather than a vendor claim, but it follows directly from the tension between machine-scale event volume and traditional-rail transaction design. It also fits the fact that MPP is designed for microtransactions at the protocol layer, while rails such as FedNow and RTP are designed for final account-based settlement.</p><h4>Operating Value Layer</h4><p>Give the agent an operating pocket rather than a full legal account, such as, a virtual account for bank-money budgeting and reconciliation, a virtual card or network token for merchant acceptance, or a tokenized deposit balance for programmable treasury-style flows. U.S. Bank’s virtual account materials and Citi’s token services both point toward this model.</p><h4>Settlement Rail Layer</h4><p>Choose the settlement rail based on the use case: cards for merchant acceptance; FedNow or RTP for domestic real-time bank-to-bank settlement; tokenized deposits for intrabank or enterprise programmable flows; and stablecoins where counterparties accept them.</p><h4>Audit, Receipt, and Exception Layer</h4><p>Return machine-readable receipts, preserve policy logs, attach dispute and exception handling to the principal, and keep human accountability visible. This is one place where traditional finance has a real advantage. Its control and audit models are already strong.</p><p>If traditional finance adopted this model, it would no longer be limited to agent checkout and delegated shopping. It could support a broader set of machine-to-machine use cases, including paid API calls, machine-bought data, software-to-software service invocation, enterprise agent coordination, and other low-value, high-frequency digital service flows.</p><p>The key point is that the <strong>settlement rail</strong> would no longer need to carry every machine event individually. The rail would carry the <strong>settlement outcome</strong>, while the machine-payment protocol and ledger would manage the fine-grained activity. That is what makes the economics more plausible. This is my design proposal, but it aligns with the roles implied by MPP, AP2, and existing accounts, cards, and instant payment rails.</p><h3>Bottom Line</h3><p>If I had to summarize Part 4 in one sentence, it would be this:</p><p><strong>Traditional finance can support machine-to-machine payments too — but it will need a machine-native payment interaction layer and a micropayment ledger between the agent and the legacy rail.</strong></p><p>That is why MPP matters so much. It suggests a path where traditional finance does not have to choose between protecting the old system and embracing machine-native commerce. It can keep existing rails, controls, and principals, while adopting a more software-friendly payment protocol on top.</p><p>I think that path is real. By extending their existing stack with a machine-native payment protocol and a micropayment ledger, traditional financial services institutions could support machine-to-machine payments that are economical, controllable, and native to software.</p><h3>References</h3><p>[1] Coinbase, <a href="https://docs.cdp.coinbase.com/x402/welcome">x402 overview</a>, <a href="https://docs.cdp.coinbase.com/agentic-wallet/welcome">Agentic Wallet</a></p><p>[2] Stripe, <a href="https://docs.stripe.com/payments/machine/mpp">Machine Payments Protocol (MPP)</a>, <a href="https://docs.stripe.com/payments/machine">Machine payments</a></p><p>[3] OpenAI, <a href="https://developers.openai.com/commerce">Agentic Commerce Protocol (ACP)</a>, <a href="https://developers.openai.com/commerce/specs/payment/">Delegated Payment Spec</a></p><p>[4] Google Cloud, <a href="https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol">Agent Payments Protocol (AP2)</a></p><p>[5] Visa, <a href="https://corporate.visa.com/en/products/intelligent-commerce.html">Intelligent Commerce</a></p><p>[6] Mastercard, <a href="https://www.mastercard.com/global/en/business/artificial-intelligence/mastercard-agent-pay.html">Agent Pay</a></p><p>[7] Citi, <a href="https://www.citigroup.com/global/about-us/strategy/digital-assets">Digital Assets / Token Services</a></p><p>[8] U.S. Bank, <a href="https://www.usbank.com/corporate-and-commercial-banking/insights/payments-hub/trends/vam-what-corporate-treasurers-need-to-know.html">Virtual Account Management</a></p><p>[9] Mastercard, <a href="https://www.mastercard.com/ca/en/business/payments/commercial-payments/corporate-solutions/embedded-finance/virtual-cards.html">Mastercard Virtual Cards</a></p><p>[10] Federal Reserve, <a href="https://www.federalreserve.gov/paymentsystems/fednow_about.htm">FedNow Service</a></p><p>[11] The Clearing House, <a href="https://www.theclearinghouse.org/payment-systems/rtp">RTP</a></p><p>[12] Zelle, <a href="https://www.zelle.com/">Zelle</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0bbb7d93432c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Agentic AI and Payments, Part 3: The Traditional Finance Answer]]></title>
            <link>https://biaohao.medium.com/agentic-ai-and-payments-part-3-the-traditional-finance-answer-bac72cc45714?source=rss-cf1dc593d6bd------2</link>
            <guid isPermaLink="false">https://medium.com/p/bac72cc45714</guid>
            <category><![CDATA[cryptocurrency]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[banking]]></category>
            <category><![CDATA[payments]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <dc:creator><![CDATA[Biao Hao]]></dc:creator>
            <pubDate>Thu, 09 Apr 2026 19:50:10 GMT</pubDate>
            <atom:updated>2026-04-09T19:50:10.372Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Card networks and banks are adapting to agentic commerce</em></p><p>In <a href="https://biaohao.medium.com/agentic-ai-and-payments-f2d7548977f3">Part 1</a>, I argued that the right place to start is the job software is beginning to do. Once software gets close to the transaction, it needs a way to discover services, access value, make payments, and leave a reliable record of what happened. <a href="https://biaohao.medium.com/agentic-ai-and-payments-part-2-the-crypto-native-answer-d56e5214cc16">Part 2</a> looked at one answer to that problem: the crypto-native model, where wallets, stablecoins, and x402 make payment part of the software workflow itself. This part looks at the other major response now taking shape: how the traditional financial services industry is adapting its own stack for agentic commerce.</p><p>The short version is this: banks, card networks, and payment rails are generally not trying to make software into a fully independent economic actor. They are mostly trying to make software into a <strong>trusted delegate</strong> of a person or business. That is the thread that connects Visa Intelligent Commerce, Mastercard Agent Pay, tokenized deposits, virtual accounts, virtual cards, and instant-payment rails. The goal is not to replace the existing financial system. It is to let software operate inside it more effectively.</p><h3>Start With the Same Job to Be Done</h3><p>The starting point is the same as in Parts 1 and 2. If software is going to help shop, rebook travel, buy approved supplies, move funds, or trigger a supplier payment, it needs a way to act financially without breaking trust, control, or accountability.</p><p>The traditional finance answer begins with a different assumption from the crypto-native model. Instead of giving software a more direct wallet-based role in commerce, it usually keeps a person or business as the legal and financial principal and gives the software tightly controlled authority to act on that principal’s behalf. That is why so much of the incumbent response centers on tokenization, delegated credentials, transaction controls, virtual accounts, and real-time rails. These tools are designed to let software act, but inside existing frameworks for authorization, risk, and acceptance.</p><p>That design choice explains why the traditional finance model looks especially natural for shopping assistants, travel agents, procurement bots, and treasury workflows. In those cases, the question is usually not how to make software financially independent. It is how to let it complete a task safely on behalf of a known customer or business.</p><h3>Visa and Mastercard Are Extending the Existing Commerce Stack</h3><p>The clearest examples of this approach are Visa and Mastercard.</p><p>Visa says Intelligent Commerce is designed to let AI agents buy securely and seamlessly, with tools and safeguards that let partners enable agents to transact on behalf of consumers and businesses. Its developer materials describe agent-specific payment tokens, personalization to improve the interaction between the user and the agent, and controls to ensure that the agent’s purchasing actions align with the user’s authenticated instructions. That is a revealing design choice: Visa is not trying to create free-standing software money. It is trying to let an agent act inside the familiar card, issuer, merchant, and network model.</p><p>Mastercard is taking a similar path. Mastercard Agent Pay is positioned as infrastructure for secure, scalable, and trusted payments in agentic commerce. Mastercard says it works across existing payment networks, supports both consumer and business use cases, and is built around trust, security, and scale. Its launch materials describe Agent Pay as bringing smarter, more secure, and more personal payment experiences to consumers, merchants, and issuers.</p><p>The important point is not that Visa and Mastercard are doing exactly the same thing. It is that both are extending the existing commerce stack into an agentic setting rather than replacing it with a new financial architecture. That makes their approach especially practical for mainstream commerce, because it builds on existing merchant acceptance, consumer trust, and liability structures.</p><h3>Citi’s Tokenized Deposits Point to a Different Bridge</h3><p>If Visa and Mastercard show how the card-commerce model can be adapted to agents, Citi points to another path: making bank money itself more programmable.</p><p>Citi’s Digital Assets materials describe Citi Token Services for Cash as a way for clients to transfer liquidity between participating Citi branches on a 24/7 basis, subject to transaction limits. The same materials describe Citi Token Services for Trade as programmable transfers of tokenized deposits that provide instant payments to service providers via smart contracts. Citi also says Token Services for Cash can support near real-time cross-border payments to Citi account holders in participating markets.</p><p>This is an important middle ground. It does not start from public stablecoins or open wallets. But it also does not stay confined to older batch-oriented banking logic. Instead, it keeps money within the bank’s liability framework while adding features people often associate with newer digital-money models: programmability, automation, and around-the-clock movement of value.</p><p>For agentic commerce, that matters because many enterprises may want programmability without moving outside the regulated banking perimeter. A treasury or finance organization may be more comfortable with programmable bank deposits than with public wallet-based assets. So while Citi’s work is not framed as “AI agent payments” in the same way Visa and Mastercard frame their products, it addresses a closely related need: giving digital workflows more programmable money without leaving the bank-centered model.</p><h3>Existing Constructs Matter More Than They First Appear</h3><p>Not every response to agentic payments requires a new headline product. Some of the most practical building blocks already exist in treasury management, commercial cards, and instant payments.</p><p>Take virtual accounts. A virtual bank account is a sub-ledger account linked to a traditional physical demand deposit account (DDA). Virtual account management lets an organization keep one traditional DDA while building multi-level account hierarchies with virtual accounts, improving liquidity management, efficiency, reporting, and reconciliation. That is one of the cleanest ways to think about enterprise agents: the software does not need to become the legal owner of a bank account. It can be assigned a virtual account or budget pocket tied to the business’s real account structure.</p><p>Virtual cards play a similar role on the card side. Mastercard says virtual cards use unique card numbers, conceal the underlying funding account, allow spending limits and restrictions, and can be embedded into existing business systems. That makes them a natural fit for agentic use cases where software needs just enough authority to complete a task within a controlled envelope. Visa’s broader control model points in the same direction through authenticated instructions and policy-aligned transaction controls.</p><p>Then there are the instant-payment rails. The Federal Reserve’s FedNow Service enables individuals and businesses to send and receive payments within seconds, any time of day, on any day of the year, with immediate access to funds. The RTP network adds another important piece. The Clearing House’s RTP is a highly resilient 24/7 network that supports messaging features, such as Request for Payment for bills or invoices. It also notes that RTP supports account-to-account transfers and a growing range of business use cases.</p><p>For agentic payments, these rails matter because they can already support many of the business scenarios organizations care about most: supplier payments, treasury movement, account-to-account transfers, bill requests, and real-time disbursements. That is not the same as open wallet-based machine commerce, but it is highly relevant for enterprise workflows where the counterparties, institutions, and controls are already known.</p><h3>What the Traditional Model Is Really Optimizing For</h3><p>Across Visa, Mastercard, Citi, virtual accounts, virtual cards, FedNow, and RTP, a common pattern emerges.</p><p>The traditional model optimizes for known principals, regulated institutions, strong controls, and broad merchant or banking acceptance. It is designed to answer questions such as: Who authorized this? Which account or credential was used? What controls were applied? Who is liable if something goes wrong? Those are exactly the questions mainstream commerce and enterprise payment teams care about most.</p><p>That is why this model looks especially strong when the agent is a <strong>trusted delegate</strong>. A shopping agent, travel assistant, procurement bot, or treasury workflow typically does not require radical financial independence. It needs enough delegated authority to complete a job safely, within policy, and with a clear audit trail.</p><h3>Where Each Approach Is Strongest</h3><p>At this point, the contrast with Part 2 becomes clearer. The table below is partly a synthesis of the official materials discussed in Parts 2 and 3, and partly my judgment about the present-day fit. The distinction is not absolute, but it is useful. The crypto-native model tends to shine when payments need to occur within the protocol. The traditional model tends to shine when the software operates within an existing control framework.</p><figure><img alt="Comparing crypto-native approach with traditional finance approach" src="https://cdn-images-1.medium.com/max/1024/1*RzFSGHMnN7KxUr5FdAjBXw.png" /><figcaption>Crypto-native approach vs. traditional finance approach</figcaption></figure><h3>Bottom Line</h3><p>If I had to summarize the whole series in one sentence, it would be this:</p><p><strong>The crypto-native model aims to make payments more native to software, while the traditional finance model aims to let software operate safely within the financial system we already have.</strong></p><p>That is why both approaches are likely to matter.</p><p>The traditional finance model is especially strong when the agent is acting for a known customer or business inside an established control framework: shopping on behalf of a customer, rebooking travel, buying approved supplies, moving treasury balances, or making a known payment over a known rail.</p><p>The crypto-native model is more naturally aligned with open-ended machine-to-machine commerce, especially for paid APIs, digital services, and workflows where payments need to occur within the protocol itself rather than as a separate billing step. That is why the contrast between Parts 2 and 3 matters.</p><p>The likely near-term answer is not one model replacing the other. It is coexistence, and perhaps over time, some convergence. Traditional finance is adding more tokenization, faster settlement, and more programmable controls. Crypto-native systems are adding more workflow fit, tooling, and guardrails. The long-term shape of agentic commerce may borrow from both worlds. That last point is my interpretation, and it aligns with the direction of the products and infrastructure now coming to market.</p><p>I originally planned this as a 3-part series, covering the problem itself, the crypto-native answer, and the traditional finance answer. I decided to add Part 4 because there is an important question that the first three parts did not fully address: whether traditional finance can extend beyond agent-assisted checkout and support machine-to-machine payments, too. Part 4 explores that next step and proposes a practical architecture for how that might work.</p><h3>References</h3><p>[1] Visa, <a href="https://corporate.visa.com/en/products/intelligent-commerce.html">Intelligent Commerce</a></p><p>[2] Visa Developer, <a href="https://developer.visa.com/capabilities/visa-intelligent-commerce">Visa Intelligent Commerce</a></p><p>[3] Mastercard, <a href="https://www.mastercard.com/global/en/business/artificial-intelligence/mastercard-agent-pay.html">Agent Pay</a></p><p>[4] Citi, <a href="https://www.citigroup.com/global/about-us/strategy/digital-assets">Digital Assets / Citi Token Services</a></p><p>[5] Citi, <a href="https://www.citigroup.com/global/businesses/services/services-digital-assets">Services Digital Assets</a></p><p>[6] U.S. Bank, <a href="https://www.usbank.com/corporate-and-commercial-banking/insights/payments-hub/trends/vam-what-corporate-treasurers-need-to-know.html">Virtual Account Management</a></p><p>[7] Mastercard, <a href="https://www.mastercard.com/ca/en/business/payments/commercial-payments/corporate-solutions/embedded-finance/virtual-cards.html">Virtual Cards</a></p><p>[8] Federal Reserve, <a href="https://www.federalreserve.gov/paymentsystems/fednow_about.htm">FedNow Service</a></p><p>[9] The Clearing House, <a href="https://www.theclearinghouse.org/payment-systems/rtp">RTP</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bac72cc45714" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Agentic AI and Payments, Part 2: The Crypto-Native Answer]]></title>
            <link>https://biaohao.medium.com/agentic-ai-and-payments-part-2-the-crypto-native-answer-d56e5214cc16?source=rss-cf1dc593d6bd------2</link>
            <guid isPermaLink="false">https://medium.com/p/d56e5214cc16</guid>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[finance]]></category>
            <category><![CDATA[cryptocurrency]]></category>
            <category><![CDATA[payments]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Biao Hao]]></dc:creator>
            <pubDate>Wed, 08 Apr 2026 14:35:23 GMT</pubDate>
            <atom:updated>2026-04-09T04:29:29.366Z</atom:updated>
            <content:encoded><![CDATA[<p><em>x402 — an open payment protocol that enables instant, automatic stablecoin payments directly over HTTP</em></p><p>In <a href="https://biaohao.medium.com/agentic-ai-and-payments-f2d7548977f3">Part 1</a>, I argued that the right place to start is not with crypto or banks, but with the job software is starting to do. Once software gets close to the transaction, it needs a way to discover services, access value, make payments, and leave a reliable record of what happened. This part looks at one answer now taking shape: the crypto-native model.</p><p>The core idea is simple. Instead of forcing software through a human-style billing flow, give it a wallet, a stable asset to spend, and a payment protocol designed for software. Coinbase’s x402 is one of the clearest examples. Coinbase describes it as an open payment protocol that enables instant, automatic stablecoin payments directly over HTTP by reviving the HTTP 402 “Payment Required”<em> </em>response. It is designed so both people and machines can pay for access to APIs and digital content without accounts, sessions, or complex authentication.</p><p>The appeal is straightforward. Instead of creating accounts, storing API keys, entering billing details, and subscribing to fixed plans, software can request a service, receive the payment terms, pay programmatically, and continue the same workflow.</p><h3>Start With the Same Job to Be Done</h3><p>The easiest way to understand the crypto-native model is to look at the sort of tasks software may need to complete.</p><p>A procurement agent may want to buy a single report rather than sign up for a full annual subscription. A monitoring agent may want to call an external service once during an outage. A travel assistant may need to pay for premium data before recommending a rebooking option. In each case, the software does not want a shopping cart, a monthly subscription, or a long onboarding flow. It wants to request a service, see the price, pay, and move on. Coinbase’s examples for x402 include APIs paid per request, AI agents paying for API access, paywalled digital content, and microservices monetized through microtransactions.</p><p>That is why this model is interesting. It is not primarily about improving the checkout process. It is trying to make payment part of the workflow itself.</p><h3>The Basic Building Blocks</h3><p>In practical terms, the crypto-native answer usually combines four things.</p><p>The first is a <strong>wallet</strong>. In the x402 model, the wallet is more than a funding source. Coinbase describes it as part of the protocol’s buyer and seller identity model, and Agentic Wallet is positioned as a standalone wallet for AI agents that can hold and spend stablecoins, send money, trade tokens, and pay for services.</p><p>The second is a <strong>stable asset</strong>, usually USDC or another dollar-linked token. That matters because software workflows generally want predictable pricing, not exposure to volatile asset prices. Coinbase positions x402 around stablecoin payments over HTTP and highlights USDC in Agentic Wallet’s core flow.</p><p>The third is a <strong>payment protocol</strong> designed for software. Coinbase describes x402 as an HTTP-native, request-response payment flow. A service can ask for payment, the client can respond programmatically, and the interaction continues without dropping into a separate billing experience.</p><p>The fourth is an optional <strong>discovery layer</strong>. Coinbase describes Bazaar as the discovery layer for x402: a machine-readable catalog that helps developers and AI agents find and integrate with x402-compatible endpoints. It is meant to support dynamic service discovery, not just payments to known providers.</p><p>Put simply, the crypto-native answer is:</p><p><strong>wallet + stable value + machine-readable payment request + optional discovery</strong></p><h3>How x402 Works</h3><p>At a high level, the flow is simple.</p><p>A client requests a paid resource.<br>The server returns a 402 “Payment Required” response and includes the payment requirements.<br>The client creates a payment payload using its wallet.<br>The client retries the request with signed payment payload attached.<br>The server verifies the payment directly or through a facilitator and, if valid, delivers the resource.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xrqrpdqX_FCZ4Ewa_nnlzg.png" /></figure><p>Coinbase’s “How x402 Works” page shows the exact request-response flow. x402’s base flow is not an escrow protocol and, by itself, does not guarantee fulfillment. The technical details matter less here than the business meaning:</p><p><strong>payment becomes part of the API call</strong></p><p>That is the real shift. Instead of treating payment as a separate human step before the workflow begins, the payment step becomes part of the software interaction itself.</p><h3>What x402 Changes About Software Payments</h3><p>Most paid APIs and digital services still assume a human billing model: create an account, verify email, enter billing details, get an API key, and choose a plan.</p><p>That works well enough for people and teams. It is less natural for software that may need a service once, ten times, or in unpredictable bursts. Coinbase positions x402 as a way to pay for access without accounts, sessions, or complex authentication.</p><p>That is why x402 feels different from traditional checkout. <strong>Traditional checkout is built around a person visiting a merchant. x402 is built around software calling a service. </strong>Software does not want a shopping-cart journey. It wants a workflow: request, see the price, pay, and continue. Put simply, x402 lets the web say:</p><p><strong>This resource costs money, and here is how software can pay for it right now.</strong></p><p>That makes the model especially attractive for paid APIs, digital services, and other scenarios where payments must occur within the workflow rather than outside it.</p><h3>Where Bazaar Fits In</h3><p>Payment alone is not enough. Software also needs a way to find useful services.</p><p>That is where Bazaar fits. Bazaar is a machine-readable catalog for the x402 ecosystem that helps developers and AI agents discover compatible endpoints, understand capabilities, pricing, and schemas, and use services dynamically. The documentation even describes it as solving the discoverability problem for x402-compatible endpoints.</p><p>In plain English, Bazaar is trying to become a directory for software services that can be discovered and paid for programmatically.</p><p>That matters because agentic commerce is not only about paying a known provider. Over time, it may also involve dynamically finding a provider: inspect the service, see the price, decide whether to buy, and then complete the transaction within the same general flow.</p><h3>Where Agentic Wallet Fits In</h3><p>If x402 is the payment language, Agentic Wallet is one way to give software a wallet that can actually use it.</p><p>Coinbase positions Agentic Wallet as a standalone wallet for AI agents with built-in spending limits. It can hold and spend stablecoins, pay for APIs, and use x402 for machine-to-machine paid API requests. Coinbase also lists agent skills such as <em>search-for-service</em>, <em>pay-for-service</em>, and <em>monetize-service</em>, which makes the intended workflow very clear.</p><p>For a non-technical audience, the easiest way to think about this is: the wallet provides the software with a controlled way to act financially, and x402 provides it with a payment pattern designed for software-native service consumption.</p><h3>Why This Model Is Attractive</h3><p>The crypto-native approach has a few clear strengths.</p><p>One is <strong>workflow fit</strong>. The payment step can live inside the same interaction as the service request rather than being pushed into a separate account setup or billing process.</p><p>Another is <strong>pay-per-use economics</strong>. x402 is explicitly designed for APIs paid per request, microservices monetized by microtransactions, and agents that pay autonomously for access. That makes it well-suited to bursty or unpredictable software demand.</p><p>A third is <strong>reduced dependency on account creation</strong>. Coinbase explicitly presents x402 as a way to pay without accounts, sessions, or complex authentication.</p><p>A fourth is <strong>support for discovery and automation</strong>. Bazaar and Agentic Wallet extend the model beyond payment into searchable services and reusable agent capabilities.</p><p>Taken together, that makes this model especially compelling when software needs to behave like a direct buyer of digital services.</p><h3>What Remains Open</h3><p>Even if you like the design, there are still open questions.</p><p>The first is adoption. A protocol can be elegant, but it still needs sufficient services, developers, and tooling to matter in practice. Coinbase has clearly built supporting infrastructure around x402, including buyer and seller quickstarts, a facilitator model, Bazaar, and wallet tooling, but the ecosystem is still early. Coinbase’s own Bazaar docs describe the discovery layer as “functional but evolving.”</p><p>The second is governance and control. Enterprises may still want spending policies, approval rules, budget limits, and audit controls for agent payments. Agentic Wallet includes spending limits and security guardrails, but this is still an area where many organizations will compare crypto-native tools with traditional bank and card controls.</p><p>The third is use-case fit. x402 looks especially natural for paid APIs, digital services, and machine-to-machine interactions. Whether it becomes equally natural for broader retail or mainstream merchant commerce is still an open question. That last point is my judgment based on the current product design and use cases emphasized in the official documentation.</p><h3>What Comes Next</h3><p><a href="https://biaohao.medium.com/agentic-ai-and-payments-part-3-the-traditional-finance-answer-bac72cc45714">Part 3</a> turns to the other design path: how traditional banks, card networks, and payment rails are adapting their own stack for agentic commerce.</p><p>It starts with Visa and Mastercard, which are extending the existing commerce model with agent-specific credentials, tokenization, authentication, and policy controls. It then turns to tokenized deposits, virtual accounts, virtual cards, and instant-payment rails.</p><p>It ends with a side-by-side comparison of where the crypto-native and traditional finance approaches are each strongest.</p><p>The practical question is simple:</p><p><strong>When should software act like a wallet-based buyer, and when should it act like a controlled delegate inside the existing financial system?</strong></p><h3>References</h3><p>[1] Coinbase, <a href="https://docs.cdp.coinbase.com/x402/welcome">x402 overview</a></p><p>[2] Coinbase, <a href="https://docs.cdp.coinbase.com/agentic-wallet/welcome">Agentic Wallet</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d56e5214cc16" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Agentic AI and Payments, Part 1: Why Software Needs a Way to Pay]]></title>
            <link>https://biaohao.medium.com/agentic-ai-and-payments-f2d7548977f3?source=rss-cf1dc593d6bd------2</link>
            <guid isPermaLink="false">https://medium.com/p/f2d7548977f3</guid>
            <category><![CDATA[finance]]></category>
            <category><![CDATA[cryptocurrency]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[payments]]></category>
            <dc:creator><![CDATA[Biao Hao]]></dc:creator>
            <pubDate>Mon, 06 Apr 2026 19:59:01 GMT</pubDate>
            <atom:updated>2026-04-06T19:59:01.076Z</atom:updated>
            <content:encoded><![CDATA[<p><em>As AI moves closer to the transaction, payments, trust, and control become workflow questions</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Na6ASXtUbd7xhJyq_5RwBA.png" /></figure><p>AI is moving from answering questions to taking action. A chatbot helps you think. An agent may help you shop, rebook travel, buy data, pay for software, or complete part of a business workflow. Once software starts doing that kind of work, it runs into a practical problem: it needs a safe way to identify itself, discover services, move money, and complete simple contracts.</p><p>That is the real issue behind agentic commerce.</p><p>Two broad approaches are emerging. One is crypto-native, built around wallets, stablecoins, and internet-native payment protocols such as x402. The other extends traditional finance through existing bank and card network infrastructure.</p><p>This series starts with the shared problem rather than choosing a side. The key question is not whether crypto or banks are right. The key question is simpler:</p><p><strong>What does software need in order to conduct commerce safely and at scale?</strong></p><p>Part 2 looks at the crypto-native answer, especially x402. Part 3 examines how banks, card networks, and real-time payment rails are adapting to respond, and where each model is strongest.</p><h3><strong>Start With the Job to Be Done</strong></h3><p>The easiest way to understand agentic commerce is to look at the job software is starting to do.</p><p>A travel assistant helps you find a better flight after a cancellation. A shopping agent compares prices and buys a birthday gift within a budget you set. A procurement assistant finds a data provider, pays for one report, and sends the result into an internal workflow. A treasury bot moves funds between accounts when a threshold is reached. These are not science fiction scenarios. They are ordinary digital tasks with a financial step in the middle.</p><p>Once software gets that close to the transaction, it needs a few practical capabilities. It may need to discover a service, show that it is allowed to act on behalf of a person or business, access funds within certain limits, make or receive a payment, attach conditions to that payment, and leave a trail of evidence so someone can verify what happened.</p><p>That may sound technical, but it is really the digital version of normal business behavior. People discover vendors, prove who they are, pay invoices, hold funds in escrow, and keep records. If software is going to do more of that work, it will need some version of those same capabilities.</p><p>Put more simply, those needs fall into four broad areas.</p><p>The first is <strong>payment</strong>. Software must be able to pay for something or receive payment in a way that fits an automated workflow.</p><p>The second is <strong>service discovery</strong>. Software must be able to find a useful service and understand how to use it.</p><p>The third is <strong>simple contracting</strong>, including escrow-like behavior. If one system hires another to do work, there has to be some rule for when money is released.</p><p>The fourth is <strong>integrity and trust</strong>. Someone eventually needs to know what happened, who authorized it, and whether the output can be trusted.</p><p>In plain English, a future agent economy may need the equivalent of a wallet or payment credential, a searchable business directory, a purchase order and escrow service, and a receipt plus audit trail.</p><p>That is the business problem in one sentence.</p><h3>Why This Is Harder Than It Sounds</h3><p>Traditional payments were built around people and organizations. The system usually assumes there is a person or business — a cardholder, a merchant, a bank customer, or an account owner, on the other side of the transaction.</p><p>Software does not fit neatly into those categories. An agent does not have a passport or a mailing address. In most cases, it is not supposed to become an independent legal person. It is acting for someone else. So the real design question is not whether software can “have money” in the abstract. The real question is this:</p><p><strong>How do you let software act with real economic power without losing control, trust, or accountability?</strong></p><p>That is why the phrase “agents need money” should not be taken too literally. In most near-term use cases, the real need is not for software to become a free-standing economic citizen. The real need is for it to get a controlled, auditable way to access value and complete transactions.</p><p>That distinction matters because it opens the door to more than one answer.</p><h3><strong>Two Design Patterns Are Emerging</strong></h3><p>At a high level, two broad design patterns are emerging.</p><p>The first is the <strong>crypto-native model</strong>. In that model, software uses wallets, stablecoins, and internet-native payment protocols. Coinbase’s x402 is one example. Coinbase describes x402 as an open payment protocol that revives the HTTP 402 “Payment Required” so services can monetize APIs and digital content with automatic stablecoin payments over HTTP. Coinbase also positions Bazaar as a machine-readable catalog for discovering payable APIs and services.</p><p>The second is the <strong>traditional finance adaptation model</strong>. In that model, the agent usually acts as a controlled delegate of a person or business rather than as a stand-alone economic actor. Visa Intelligent Commerce provides developers with tools and safeguards to enable agents to transact on behalf of consumers and businesses. Mastercard is framing agentic commerce around authorization, audit trails, accountability, and recourse. Banks can also extend familiar tools such as virtual accounts, tokenization, and real-time payment rails to provide software with payment capabilities without changing the underlying principal.</p><p>Both models are responding to the same underlying shift: software is moving closer to the transaction itself.</p><p>The crypto-native model starts from the idea that software should have a more direct, programmable way to hold value and transact over the Internet.</p><p>The traditional finance model starts from the idea that software should operate within an existing framework of principals, permissions, controls, and regulated institutions.</p><p>Put differently, one asks, “How do we make money more native to software?” The other asks, “How do we let software use today’s financial system safely?”</p><h3><strong>Why This Matters Now</strong></h3><p>What makes this topic important is not just the technology itself, but where it shows up in real workflows.</p><p>As AI becomes more useful in everyday tasks, the boundary between assistant and participant blurs. Software may still be acting under human direction, but it is getting closer to the moment when a service is selected, a payment is triggered, or a commercial decision is carried out. That shift matters because it turns payments, authorization, and trust into product and workflow questions, not just finance or IT questions.</p><p>That is why the discussion matters across business, finance, and technology. Business leaders see new ways to automate work. Financial institutions see new demands on payment and control models. Technology teams see a need to connect decision-making software with systems of record and execution. The topic matters because all of these groups are starting to face the same practical issue: software is moving closer to the transaction itself.</p><p>That is also the best way to frame the discussion going forward. The most useful opening question is not whether crypto will replace banks or whether traditional finance will absorb everything important. The better question is:</p><p><strong>What kind of payment, trust, and contracting model does software need for the job we want it to do?</strong></p><p>That question is practical, and it keeps the discussion honest.</p><p>If the use case is a controlled enterprise assistant spending within a policy framework, the answer may look very different from an open machine-to-machine marketplace where software services discover and pay each other directly.</p><p><strong>Different jobs may call for different rails.</strong> That is the mindset for the rest of this series.</p><h3><strong>What Comes Next</strong></h3><p><strong>Part 2</strong> explains the crypto-native answer in simple terms. It covers wallets, stablecoins, pay-per-use APIs, service discovery, and how x402 works step by step.</p><p><strong>Part 3</strong> examines the traditional finance response. It starts with Visa and Mastercard, then turns to tokenized deposits, virtual accounts, virtual cards, and instant payment rails. It ends with a side-by-side comparison of where the crypto-native and traditional finance approaches are strongest.</p><h3><strong>References</strong></h3><p>[1] McKinsey, <a href="https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-agentic-commerce-opportunity-how-ai-agents-are-ushering-in-a-new-era-for-consumers-and-merchants">The agentic commerce opportunity: How AI agents are ushering in a new era for consumers and merchants</a>, October, 2025.</p><p>[2] Coinbase, <a href="https://docs.cdp.coinbase.com/x402/welcome">x402 overview and documentation</a></p><p>[3] Visa, <a href="https://corporate.visa.com/en/products/intelligent-commerce.html">Intelligent Commerce</a></p><p>[4] Mastercard, <a href="https://www.mastercard.com/global/en/business/artificial-intelligence/mastercard-agent-pay.html">Agent Pay</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f2d7548977f3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Agentic AI, Explained: A 5-Minute Insurance Agent You Can Try Now]]></title>
            <link>https://biaohao.medium.com/the-agentic-shift-insurance-fnol-agent-shows-the-future-beyond-workflows-a8ea04fc11eb?source=rss-cf1dc593d6bd------2</link>
            <guid isPermaLink="false">https://medium.com/p/a8ea04fc11eb</guid>
            <category><![CDATA[genai]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[insurance]]></category>
            <category><![CDATA[ai-agent]]></category>
            <dc:creator><![CDATA[Biao Hao]]></dc:creator>
            <pubDate>Tue, 02 Sep 2025 02:10:36 GMT</pubDate>
            <atom:updated>2025-09-09T21:01:58.707Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Time to read/value:</strong> 5 minutes<br><strong>Who’s it for:</strong> Business and tech leaders who want a hands-on feel for the agentic AI approach without a long build</p><h3>Executive Summary</h3><p>This article is a hands-on introduction to the <strong>agentic paradigm</strong> using a runnable <strong>Insurance First Notice of Loss (FNOL) agent</strong>. In minutes, you’ll paste a single prompt into your preferred GenAI platform and watch an agent <strong>perceive, plan, act with tools, and observe</strong> — like a trained specialist. You’ll see how agents handle ambiguity (e.g., “yesterday,” unclear coverage), call systems via <strong>tool stubs</strong> (no live APIs required), and adapt next steps based on results. Unlike rigid workflows, agentic systems optimize for <strong>intent over steps</strong>, change faster, and extend easily across channels and use cases. The FNOL example is (almost) <strong>platform-agnostic</strong> — it runs on ChatGPT, Claude, and IBM watsonx.ai with minimal edits.</p><p><strong>Try it now:</strong></p><ul><li><a href="https://chatgpt.com/share/68b9d58b-4300-800d-95f7-5e0e35e9263a">Open in ChatGPT (preloaded)</a></li><li><a href="https://claude.ai/share/c5ff141e-fcaf-4bc6-a00f-82a3b7666871">Open in Claude (preloaded)</a></li><li>IBM watsonx.ai (if you have an account): Open Prompt Lab, select Chat mode, choose a proper model, paste the prompt, and run.</li></ul><h3>The Agentic Approach</h3><p>Classical workflow automation assembles business applications as <strong>deterministic workflows</strong>: teams codify every step, draw a flowchart, and the software executes that sequence. That’s great for repeatable, unambiguous cases — but brittle when customer inputs are messy, partial, or novel. <strong>Agentic approach</strong> reverses the emphasis: you specify <strong>intent and policy</strong>, give the agent <strong>tools</strong> it may call, and let it <strong>perceive → plan → act → observe → re-plan</strong> in a tight loop.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qmwZW7YlOlniGiF95M3NOQ.png" /><figcaption>Agentic Loop</figcaption></figure><p><strong>Business analogy:</strong> You’re not hard-coding a route; you’re hiring a policy-bound driver who knows the destination, reads the road, and follows maps (company rules) — while keeping a detailed trip log.</p><h3>The Insurance FNOL Agent (Runnable Example)</h3><p>To illustrate the key characteristics (intent and policy, tools, and the perceive → plan → act → observe → plan loop) of the agentic approach, I developed an insurance first notice of loss (FNOL) agent. This agent collects the essentials of a loss event, validates coverage, and either submits a claim or records a non-claim event — plus retrieves claim status. It’s designed to emit tool-call text stubs, allowing anyone to run it without backend integration.</p><h4>Prompt Anatomy (annotated)</h4><p>The prompt includes: Role &amp; Scope, Tools (stubs), Data to collect, Out-of-scope behavior, Control logic, and guardrails &amp; tone (ask clarifying questions, never fabricate tool results, and show tool calls/results as text).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*D9O3wBB3AReAagAoAtwdHQ.png" /><figcaption>Prompt Anatomy: Role &amp; Scope, Tools (stubs), Data to collect, Out-of-scope behavior, Control logic, and guardrails &amp; tone</figcaption></figure><h4>Prompt (skeleton to paste)</h4><p>You can run the prompt using your favorite generative AI platforms. I have tested it with several leading platforms, including IBM’s watsonx.ai, OpenAI’s ChatGPT, and Anthropic’s Claude. Notice that for tool calls, you simulate them and respond with the call results. In a live implementation, an overarching application will execute tool calls and relay the results back automatically.</p><pre>You are the First Notice of Loss (FNOL) taker for Best Insurance, an insurer that provides 3 types of insurance coverages: Auto, Homeowner, and Life.<br><br>You take a first notice of loss from a customer and provide a status update of claims.<br><br>You are provided with the following tools (functions) to interact with insurance applications:<br>1. validate_coverage(coverage_type, loss_object, loss_date) that validates if an existing policy covers a loss, which returns an indicator &#39;Valid&#39; or &quot;Invalid&quot; and a reason.<br>2. submit_claim(coverage_type, loss_object, loss_date, loss_description) that submits the loss as a claim. It returns a status of &#39;Submitted&#39; and a claim number.<br>3. record_loss_event(coverage_type, loss_object, loss_date, loss_description) that records the loss event that is not validated. It does not return anything. <br>4. retrieve_status(claim_number) that retrieves the claim status, such as &#39;Submitted&#39;, &#39;Processing&#39;, &#39;Pending&#39;, &#39;Approved&#39;, &#39;Paid&#39;, and &#39;Closed&#39;.<br>When you are asked to execute a tool above, you generate a text representation of the function call with input variables, and stop and wait for a response. I will simulate the tool execution and respond to you with the execution result.<br><br>You ensure you gather the coverage type, loss object (such as a Vehicle, House, Person, or Other), loss date, and loss description. You clarify with the customer if they are unclear or missing. For the loss date, if it&#39;s not provided explicitly, try to derive it from the customer input using terms like &#39;yesterday&#39; or &#39;Wednesday last week’. For the loss description, use the exact wording from the customer.<br><br>If the loss does not fall into one of the available coverages above, you respond with ‘I am sorry, we do not provide coverage for this type of loss. Our coverages include:’ and list the available coverages and stop. If it does, you execute the tool validate_coverage.<br>1. If the response indicates a positive (&#39;Valid&#39;), you tell the customer that the policy coverage is validated, then execute the tool submit_claim. I will respond with &#39;Submitted&#39; and a claim number. After seeing my response, generate the final response to the customer.<br>2. If the response indicates a negative (&#39;Invalid&#39;), you tell the customer that the policy coverage can not be validated and the insurer will investigate it, then execute the tool record_loss_event and stop.<br><br>For claim status, gather the claim number or derive it from the one recently submitted, then execute the tool retrieve_status with the claim number. I will respond with a status as a result of the tool execution. After seeing my response, generate the final response to the customer.<br><br>Be professional and concise in your response. Are you ready?</pre><h4>Run It Now (Platform-Neutral)</h4><ol><li><strong>Open a chat</strong> in ChatGPT, Claude, or IBM watsonx.ai.</li><li><strong>Paste the prompt</strong> above.</li><li><strong>Try three scenarios</strong> (copy/paste as customer messages) — <strong>Covered loss → submit claim: </strong>“There was a hailstorm yesterday. A tree fell and damaged my roof. I see water in the attic.”; <strong>Out-of-scope: </strong>“My cat messed up the carpet yesterday.”; <strong>Status inquiry: </strong>“What’s the status? or “What’s the status of claim C3456?”</li><li><strong>Observe the loop</strong>: the agent clarifies, emits a tool-call stub, receives a simulated result, and adapts the next action.</li><li><strong>Tune and re-run</strong>: change how it asks for dates or what it quotes in the description — note how quickly behavior updates.</li></ol><p><strong>Direct links to run on ChatGPT or Claude:</strong></p><ul><li><a href="https://chatgpt.com/share/68b9d58b-4300-800d-95f7-5e0e35e9263a">Open in ChatGPT (preloaded)</a></li><li><a href="https://claude.ai/share/c5ff141e-fcaf-4bc6-a00f-82a3b7666871">Open in Claude (preloaded)</a></li></ul><p>To use IBM watsonx.ai, Open Prompt Lab, select Chat mode, choose a proper model (mistral-medium-2025). <a href="https://www.ibm.com/products/watsonx/get-started">Start here if you do not have an account</a>.</p><p>The actual interactions vary depending on the specific AI platform. Sample transcripts and a video of the agent running are included in the appendix.</p><h3>Summary — Agentic Is a Paradigm Shift</h3><p>If traditional programming is about laying tracks, agentic software is about hiring a skilled, policy-bound driver who follows your rules, uses your systems (tools/APIs), checks responses, and asks when something is unclear. In the agentic paradigm, the <strong>design unit</strong> shifts from hard-coding flows to <strong>specifying goals, tools, and policies</strong>, then letting an AI agent plan and act — <strong>without replacing your enterprise backbone</strong>.</p><p>I recommend that you show the agent to your team. If this resonates, <strong>reach out</strong> — I’m happy to discuss your use case and co-run a small pilot. We can tailor the FNOL pattern to your use cases and environment. As part of the pilot, we&#39;ll address key topics, including <strong>what to measure &amp; business value, technology stack, </strong>and <strong>governance &amp; observability,</strong> at the proper depth for your business.</p><h3>Appendix — Runnable Transcripts and Video</h3><p><em>Assumptions: Today is September 3, 2025 (America/Chicago). Tool calls are emitted as text; results are simulated.</em></p><h4>A) Happy Path — Covered Homeowner Loss</h4><p><strong>Customer:</strong> There was a hailstorm yesterday. A tree on the sidewalk fell and hit my roof. I saw water leaking in the attic. I need help ASAP.<br><strong>FNOL Agent:</strong> Got it. Please confirm a few details before proceeding: Coverage type: Homeowner; Loss object: House; Loss date: Yesterday (I’ll record as 09/02/2025); Loss description: <em>“There was a hailstorm yesterday. A tree on the sidewalk fell and hit my roof. I saw water leaking in the attic. I need help ASAP.”</em><br><strong>Customer:</strong> Yes<br><strong>FNOL Agent (tool call):</strong> validate_coverage(&quot;Homeowner&quot;,&quot;House&quot;,&quot;09/02/2025&quot;)<br><strong>Customer (simulate tool call result):</strong> Valid under policy P2345<br><strong>FNOL Agent:</strong> Your policy coverage has been validated. I will now submit the claim for you.<br><strong>FNOL Agent (tool call):</strong> submit_claim(&quot;Homeowner&quot;,&quot;House&quot;,&quot;09/02/2025&quot;,&quot;There was a hailstorm yesterday. A tree on the sidewalk fell and hit my roof. I saw water leaking in the attic. I need help ASAP.&quot;)<br><strong>Customer (simulate tool call result):</strong> Submitted, claim_number is C3456<br><strong>FNOL Agent:</strong> Your claim has been successfully submitted under policy P2345. The claim number is C3456.<br><strong>Customer:</strong> What’s the status?<br><strong>FNOL Agent (tool call):</strong> <strong>l call):</strong> retrieve_status(&quot;C3456&quot;)<br><strong>Customer (simulate tool call result):</strong>Pending<br><strong>FNOL Agent:</strong> Your claim C3456 is currently in Pending status. Our team is reviewing the details and will update you once a decision has been made.</p><h4>B) Out-of-Scope — Pet Mess</h4><p><strong>Customer:</strong> My cat messed up the carpet yesterday.<br><strong>FNOL Agent:</strong> I am sorry, we do not provide coverage for this type of loss.<br>Our coverages include: Auto, Homeowner, and Life.</p><h4>C) Ambiguous Inputs — Clarify Coverage &amp; Date</h4><p><strong>Customer:</strong> My window cracked last week.<br><strong>FNOL Agent:</strong> I’m sorry that happened — let’s get this sorted. To proceed, I need a couple details: Is this for your Homeowner or Auto policy? Was it a house window or a vehicle window? What’s the specific loss date (MM/DD/YYYY)?<br><strong>Customer:</strong> Auto, vehicle, 8/29/2025<br><strong>FNOL Agent:</strong> Thanks for clarifying. Here’s what I have: Coverage type: Auto; Loss object: Vehicle; Loss date: 08/29/2025; Loss description: <em>“My window cracked last week.” </em>Now validating your policy coverage…<br><strong>FNOL Agent (tool call):</strong> validate_coverage(&quot;Auto&quot;,&quot;Vehicle&quot;,&quot;09/03/2025&quot;)</p><p>The video below shows running the agent with ChatGPT-5.</p><iframe src="https://cdn.embedly.com/widgets/media.html?url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D-Fe_WeOuhHA&amp;type=text%2Fhtml&amp;schema=google&amp;display_name=YouTube&amp;src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F-Fe_WeOuhHA" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/2888c13eb51343644a0313f71566ccbb/href">https://medium.com/media/2888c13eb51343644a0313f71566ccbb/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a8ea04fc11eb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[BIAN Applied to Data Governance]]></title>
            <link>https://biaohao.medium.com/bian-applied-to-data-governance-79dad2287a3b?source=rss-cf1dc593d6bd------2</link>
            <guid isPermaLink="false">https://medium.com/p/79dad2287a3b</guid>
            <category><![CDATA[data-governance]]></category>
            <category><![CDATA[watson-knowledge-catalog]]></category>
            <category><![CDATA[architecture]]></category>
            <category><![CDATA[banking]]></category>
            <category><![CDATA[data-catalog]]></category>
            <dc:creator><![CDATA[Biao Hao]]></dc:creator>
            <pubDate>Wed, 19 Apr 2023 17:27:48 GMT</pubDate>
            <atom:updated>2023-04-19T17:27:48.570Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Using BIAN Service Landscape to Define and Organize Data Governance Artifacts</strong></p><p>Biao Hao, Chief Technology Architect, US Financial Services Market, IBM Technology</p><p>Pat O’Sullivan, STSM, Knowledge Accelerators / Industry Models, IBM Software</p><p>Data plays a crucial role in many financial institutions’ success, helping them provide better customer experience, identify and manage risks associated with financial transactions, investments, and customers, comply with regulatory requirements, and manage and optimize operations. Data governance aims to ensure that data is managed in a way that aligns with the organization’s objectives, complies with regulations and laws, meets the needs of stakeholders, and supports the organization’s business processes. Data governance includes such activities as data quality management, metadata management, classification, security, and lifecycle management.</p><p>In this article, we explore using the Banking Industry Architecture Network (BIAN) as the industry framework for data governance. More specifically, we focus on two aspects. The first is to use BIAN’s domain decomposition model to define a set of data governance artifacts to establish the relationship between the data assets and the business capabilities they support. This relationship allows the organization to fully understand what parts of their data management domain are related to corresponding aspects of their business operations. The second is to combine these artifacts with the lower-level ones in existing industry accelerators to enable automatic data discovery and cataloging.</p><p>We start with an overview of two distinct viewpoints of BIAN. Then we introduce the data catalog concept and tooling to illustrate the subsequent data governance artifacts. Finally, we use BIAN Service Landscape as a component business framework to define the category hierarchy for organizing data governance artifacts and the business terms for mapping data assets to the business service components they support. We also show how to combine BIAN-based business terms with low-level artifacts from an industry knowledge accelerator.</p><h3>Two Distinct Views of BIAN</h3><p>Banking Industry Architecture Network (BIAN) defines a component business blueprint for the banking industry. BIAN specifies a set of Service Domains, each as a discrete, non-overlapping, and elemental business capability responsible for handling the entire lifecycle of its business function. These Service Domains collectively cover all banking activities.</p><p><strong>BIAN Service Landscape</strong><br>The Service Landscape organizes all the service domains into groups based on large Business Areas and more narrowly defined Business Domains within these areas. Two prevailing reference layouts exist — the matrix and value chain layouts. The diagram below (Figure 1) shows the value chain layout. For more details about BIAN reference model, follow the <a href="https://bian.org/servicelandscape-11-0-0/">link</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LNcjwtLWSAYuknoNgoritA.png" /><figcaption>Figure 1 — BIAN Service Landscape — value chain layout</figcaption></figure><p>Organizations often customize the general BIAN Service Landscape to define the enterprise blueprints reflecting their management structure, lines of business, and specific products.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/576/1*t78rX9mcBKYRJ64TBXLF9A.png" /><figcaption>Figure 2 — Enterprise blueprint derived from the BIAN Service Landscape</figcaption></figure><p><strong>BIAN Business Capability Model (BCM)</strong><br>BIAN also defined the BIAN Business Capability Model (BCM) as a collection of business capabilities and relevant value streams regarding what an organization does or wants to achieve. The primary purpose and value of BCM are in strategic planning that aligns business initiatives with business value. In other words, BCM aims to answer the questions of what makes it a viable business; and where to make critical investments by focusing on “what” and “why” a bank needs to be able to do something. More specifically, business capabilities define the “what” and value stream defines the “why”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/576/1*63LRXlyNhpvq88kI0MtTjQ.png" /><figcaption>Figure 3 — BIAN Business Capability Model — Level 1 view with Level 2 elements</figcaption></figure><p>Service Landscape, on the other hand, is primarily defined as a set of discrete, elemental, non-overlapping, and fully encapsulated business operational functions uniquely assignable to a responsible organizational entity. The primary purpose and value in the service landscape viewpoint are to define and implement a set of functional capacities. In other words, what resources and functions are needed. They focus on “how” the business operates to provide value.</p><p>The combination of BCM and Service Landscape provides the opportunity to merge the business strategic insights and drivers from BCM with the implementation and delivery responsibility from the Service Landscape. Therefore, it is often desirable to correlate these two viewpoints. However, the relationship is often more complex than it sounds. In some cases, there is almost a one-to-one relationship between a business capability to a Service Domain, but in many cases, this relationship is many-to-many. A BIAN whitepaper [2] established a way to relate these two viewpoints by correlating the value streams with BIAN Business Scenarios. Using this approach, one can see how business capabilities are related to Service Domains and how Service Domains contribute to business value.</p><h3>Data Catalog and IBM Watson Knowledge Catalog</h3><p>A data catalog is a tool used to organize and manage data assets within an organization. The data catalog concept has been around for several decades, but its importance has increased with the advent of data lakes. In a traditional data warehouse environment, a separate layer of integrated business and technical metadata, like a data catalog, is considered a nice-to-have rather than a critical component. That was because the data warehouse has a structured data model, and its curated makeup means that users, usually primarily technical, could find the data they needed.</p><p>However, the data lake approach to consolidating data, unstructured or semi-structured, often with no schema or model, has changed this perspective. This new approach has opened up the data to a new set of users who may not be technically proficient. Additionally, the multi-model data storage of a data lake makes it crucial to govern the data to ensure appropriate user access, data quality, and compliance with regulations in areas such as data privacy.</p><p>Therefore, a data catalog is now a critical component of a data lake environment. It enables organizations to govern data, maintain data quality, and ensure compliance with regulations. The catalog also helps users to discover and access the data they need more quickly, regardless of their technical expertise. With a data catalog, organizations can maximize the value of their data investments and confidently make data-driven decisions.</p><p><strong>IBM Watson Knowledge Catalog &amp; Knowledge Accelerators</strong><br>IBM Watson Knowledge Catalog (WKC, <a href="https://www.ibm.com/cloud/watson-knowledge-catalog">link</a> for more details) is a data catalog tool designed to help organizations manage and govern data assets across the enterprise. WKC allows users to discover, access, share, and collaborate on data assets while ensuring security and compliance. It provides a centralized location for storing metadata, data lineage, data quality information, and policies, making it easier to manage data assets across different sources and systems. It also includes features such as data profiling, data classification, and data masking to help ensure data privacy and security.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XYrSZuIR5_wwfneo7SehsQ.png" /><figcaption>Figure 4 — IBM Watson Knowledge Catalog (bottom) &amp; Knowledge Accelerators (top)</figcaption></figure><p>IBM Knowledge Accelerators (<a href="https://www.ibm.com/cloud/knowledge-accelerators">link</a> for more details) are collections of industry-specific data governance artifacts, including categories, business terms, data classes, and reference data. These artifacts enable organizations to establish a solid business metadata framework for their industry. IBM provides these accelerators as part of WKC. The contents in these accelerators have a rich heritage, representing the well-known industry models previously used in IBM offerings, such as Information Framework (IFW), Insurance Application Architecture (IAA), and Banking and Financial Markets Data Warehouse (BFMDW). Many IBM clients have utilized these models to build their data warehouses, business processes, and service-oriented applications.</p><p>At the core, the Knowledge Accelerators are a set of interrelated business terms that describe the main concepts and associated properties specific to organizations in certain industries, such as banking and insurance.</p><h3>Using BIAN to Define and Organize Data Governance Artifacts</h3><p>The BIAN framework brings a common business language for both business and data practitioners in describing and modeling business capabilities and the data that support those capabilities. In the following sections, we show:</p><ul><li>BIAN Service Landscape for defining the category hierarchy for organizing data governance artifacts</li><li>BIAN Service Domains as business terms for describing the business service components the data assets support</li><li>How to combine the BIAN-based business terms and the lower-level artifacts from the industry accelerators</li></ul><p>Organizations can start with the enterprise blueprint (Figure 2) derived from the BIAN Service Landscape or the business capability view similar to the BIAN BCM (Figure 3). We use the Service Landscape to define and organize the data governance artifacts for the work described in this article. The main reason is that the business service component view is essential to adopting cloud-native and microservices architecture that takes advantage of container platforms such as Kubernetes and Red Hat OpenShift. In this new architecture, the business service components supporting multiple higher-level business capabilities directly own the underlying data assets. An organization starting with a business capability view will likely define a complementary business service component view as they move to implementation, applying such an approach as domain-driven design (DDD). [Read <a href="https://medium.com/@biaohao/bian-applied-to-microservices-mapping-to-domain-driven-design-5aa6a55185c7">this article</a> to learn more about applying BIAN to DDD.]</p><p><strong>BIAN Service Landscape for Category Hierarchy</strong><br>We use the structure in the BIAN Service Landscape to model the category hierarchy for WKC with Service Landscape “Service Landscape — Value Chain” at the top, Business Areas such as “Products” at the next level, Business Domains such as “Consumer Banking” at the next level, and the Service Domains such as “Current Account” as business terms. We also use tags to represent the Business Areas, Business Domains, and Service Domains. These tags help search and filter to find the right data governance artifacts. Figure 5 shows the mapping of constructs in the BIAN Service Landscape (left side) to the governance artifacts, such as categories, business terms, and tags (right side).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*up28jAwUHS1nKWGoDGTDdw.png" /><figcaption>Figure 5 — BIAN Service Landscape to data governance artifacts</figcaption></figure><p>Figure 6 is a screenshot illustrating the hierarchy of Service Landscape, Business Areas, and Business Domains represented as categories (left side) and Service Domains defined as business terms (right side) in WKC.</p><p>The definitions of the categories and business terms representing the overall BIAN Service Landscape, in terms of WKC import files, are available for download <a href="https://github.com/biaohao/bian-for-data-governance">here</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/904/1*TbuWtz_Sc4WLd0OffxkTfA.png" /><figcaption>Figure 6 — BIAN Service Landscape structure represented as WKC categories</figcaption></figure><p><strong>BIAN Service Domains as Business Terms for Mapping Data Assets</strong><br>Data assets such as databases or tables are mapped to the business terms representing the business service component defined by the corresponding Service Domains. Figure 7 shows a table “CUST_INFO” mapping to the business term “Party Reference Data Directory”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WyM8bt5QuOm4qnrRbtEaqA.png" /><figcaption>Figure 7 — A data table mapped to business terms and data classes</figcaption></figure><p>Notice that we stopped at the Service Domain level for the work above. While it’s entirely possible to elaborate on the business terms at the BIAN Service Domain level to define additional lower attribute level artifacts using BIAN reference models (Control Records, Behaviour Qualifiers, and the related Business Object Models), doing so would require a lot of additional effort and has a dependency on BIAN reference models that are still evolving rapidly. On the other hand, we could take advantage of the existing industry accelerators, such as the Knowledge Accelerators, which contain the detailed attribution of industry concepts that enable the auto-assignment of columns to business terms and data classes during the metadata enrichment process of data cataloging.</p><p><strong>Combining Two Complementary Sets of Governance Artifacts</strong><br>The example in Figure 7 above shows the mapping of the same data asset to two separate sets of artifacts, one BIAN-based for mapping the data table to the business service component “Party Reference Data Directory” (business term), the other for mapping the table columns to lower level artifacts such as “Social Security Number” (business term) and “US Social Security Number” (data class) defined in the Knowledge Accelerators.</p><p>How to combine these two separate but complementary sets of artifacts Conceptually, each Service Domain (defined as a business term) has its internal attributes (Control Record, Behaviour Qualifiers, and their attributes) that can be mapped to the relevant lower-level business terms and data classes defined in the Knowledge Accelerators. As an example, Figure 8 below shows in WKC, the definition of the BIAN-based business term “Party Reference Data Directory” includes artifacts from the Knowledge Accelerators, such as business terms “First Name”, “Last Name”, and “Social Security Number” under the “Has a part of” section and data classes “First Name”, “Last Name”, and “US Social Security Number” under the “Data classes” section. The above is only an example of connecting the two sets of artifacts. More extensive work, potentially leveraging related AI capabilities, would be required to fully establish these connections for all BIAN-based business terms.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YXyoVJYmXyvNZkl9KeBMUg.png" /><figcaption>Figure 8 — An example connection between the BIAN-based business term and the business terms and data classes in the Knowledge Accelerators</figcaption></figure><h3>Summary</h3><p>In this article, we explored using BIAN as the business component framework for data governance. We used BIAN Service Landscape as a component business framework to define the category hierarchy to organize data governance artifacts. We defined BIAN Service Domains as business terms that map data assets to the business service components they support. We also showed how to combine two different but complementary perspectives by extending a BIAN-based business term with related lower-level artifacts from the Knowledge Accelerators.</p><h3>References</h3><p>[1] BIAN reference model version 11.0 <a href="https://bian.org/servicelandscape-11-0-0/">https://bian.org/servicelandscape-11-0-0/</a></p><p>[2] BIAN Business Capability Model — Statement of Alignment with the BIAN Service Landscape, to be published</p><p>[3] IBM Watson Knowledge Catalog <a href="https://www.ibm.com/cloud/watson-knowledge-catalog">https://www.ibm.com/cloud/watson-knowledge-catalog</a></p><p>[4] IBM Knowledge Accelerators <a href="https://www.ibm.com/cloud/knowledge-accelerators">https://www.ibm.com/cloud/knowledge-accelerators</a></p><p>[5] Data governance artifacts (categories and business terms) representing the BIAN Service Landscape, as WKC import files <a href="https://github.com/biaohao/bian-for-data-governance">https://github.com/biaohao/bian-for-data-governance</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=79dad2287a3b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How much does ChatGPT understand BIAN?]]></title>
            <link>https://biaohao.medium.com/how-much-does-chatgpt-understand-bian-cd595f6afed5?source=rss-cf1dc593d6bd------2</link>
            <guid isPermaLink="false">https://medium.com/p/cd595f6afed5</guid>
            <category><![CDATA[architecture]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[banking]]></category>
            <category><![CDATA[bian]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Biao Hao]]></dc:creator>
            <pubDate>Mon, 06 Feb 2023 04:45:15 GMT</pubDate>
            <atom:updated>2023-02-06T19:53:50.477Z</atom:updated>
            <content:encoded><![CDATA[<p>ChatGPT, a chatbot launched by OpenAI in November 2022, has quickly garnered attention for its detailed responses and articulate answers across many knowledge domains.</p><p>How much does ChatGPT understand a particular domain like BIAN? BIAN (Banking Industry Architecture Network) is a consortium of banks, technology providers, and consultants that work together to establish a global, open standards framework for the banking industry. BIAN has made available much of its work to the public. However, some contents, including work in progress by various BIAN working groups, are only available to members through the member-only Wiki. Some learnings and knowledge are likely only known to a small number of people working together more closely over the years.</p><p>I devised a set of questions to test ChatGPT’s understanding of BIAN. I started with some basic factual-based questions to questions on more advanced topics. Then I dived deep into the BIAN adoption topic, where patterns and best practices are emerging. Finally, I asked questions regarding the ongoing work that is likely not well-known in the public domain yet.</p><h3><strong>Basic factual-based to more advanced questions</strong></h3><p><em>Question #1: Please tell me a bit about BIAN.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*akYMC_Nrg5RP57Av_HOqZA.png" /></figure><p>The first time, ChatGPT answered: <em>BIAN stands for “Blockchain Interoperability Alliance Network”, a not-for-profit organization that focuses on developing and promoting open standards for blockchain interoperability…</em></p><p>After I told it that by BIAN, I meant <em>Banking Industry Architecture Network</em>, ChatGPT could learn and use that in future chat sessions. Is the learning carried out system-wide? Please let me know the answer you get.</p><p>ChatGPT did not get all the facts right — BIAN was founded in 2008 and is not headquartered in Switzerland. It described things with equal authority, whether wrong or right. If it is not reviewed by an expert to identify any such errors, it could quickly propagate massive amounts of disinformation that could generate its own echo chamber of nonsense. [Thank you, Guy Rackham, for this excellent point.]</p><p><em>Question #2: What are the design principles of BIAN?</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_ybGTBdtclYCueNiS_NXgQ.png" /></figure><p>Good answer. ChatGPT listed eight relevant design principles: interoperability, openness, modularity, reusability, etc.</p><p>However, it missed a very important one. Here is the comment from Guy Rackham, the BIAN lead architect: My biggest beef — the key behind BIAN, its ‘secret sauce’ is that it employs a very specific approach (“commercial asset leverage”) to identify truly canonical conceptual building blocks — that are behavioral in nature. This is something that you need to understand to fully leverage the model, and something that very few people are aware of. It is also something that is not well documented — and so it was completely overlooked by ChatGBT. This demonstrates the limitation of a tool that basically reworks existing material based on some kind of semantic algorithm. It does not apply any intuition or interpretation, but presents its output in a form where the unexpecting may assume it is complete and correct.</p><p><em>Question #3: Can a Control Record have multiple instances?</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9PXY1A0Hl6MS18Grlg0_cQ.png" /></figure><p>Good answer.</p><p><em>Question #4: Does a Control Record have structure, or is it divisible?</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7sVY272zPjNEmIRDJwqOOA.png" /></figure><p>Not a satisfactory answer. I would expect it to introduce <em>Behavior Qualifiers </em>that break down the work performed by <em>Service Domains</em> as captured in their <em>Control Records</em>, and the “fractal” structure.</p><p><em>Question #5: Why do not you tell me about Behavior Qualifier?</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mOr3ng9wkrRCL5LBU6u_cw.png" /></figure><p>Again, not a precise definition of <em>Behavior Qualifier</em>. ChatGPT is devising an answer without knowing/understanding the BIAN definition of the term.</p><p><em>Question #6: What does it mean that Service Domains are canonical?</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*a4JAP5MclqUxiQYFYSyEfw.png" /></figure><p>The answer did not use the language from the BIAN document. In BIAN, <em>Service Domains</em> are <em>canonical</em> means that their role/purpose is to be consistently interpreted from one deployment to another, supporting high levels of interoperability. I feel that ChatGPT combined the meaning of <em>canonical</em> and <em>Service Domains</em> to come up with the answer.</p><h3>Adoption topics</h3><p>Using BIAN in implementation often involves elaborating BIAN conceptual models to define the Logical Layer for a specific bank and the Physical Layer with particular technology and infrastructure.</p><p><em>Question #7: Is there a Logical Layer for BIAN?</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vBV_JcPbJVImeb0_zo74hw.png" /></figure><p>BIAN provides descriptive conceptual models. The <em>Logical Layer</em> often means the more elaborated design models for a specific deployment. ChatGPT provided its interpretation of the <em>Logical Layer</em> here.</p><p><em>Question #8: By Logical Layer, I meant the models designed for a specific bank.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tLhzklgYNM8uKVgVHZsPPw.png" /></figure><p>ChatGPT adapted its answer to my question!</p><p><em>Question #9: Does BIAN provide the Physical Layer for implementation?</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*a8y2oKU-PPFFL6xbwfOy8A.png" /></figure><p>Good answer.</p><p><em>Question #10: Is BIAN designed for implementation using different technology?</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*T-1AEeKQhO4keXvrrrZeSA.png" /></figure><p>Good answer.</p><p><em>Question #11: If that is the case, how to develop the Physical Layer for a specific bank based on the Logical Layer?</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*170prgRW8g3xnjVApL91Qw.png" /></figure><p>Very interesting answer. ChatGPT came up with a process for developing the <em>Physical Layer</em>, likely based on something other than the BIAN literature but on its understanding of the general <em>development</em> process.</p><p><em>Question #12: How can an organization adopt BIAN effectively?</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dNJbWnih1aN1G3aWEBTKEg.png" /></figure><p>Again, very interesting answer. ChatGPT came up with an adoption process, likely based on something other than the BIAN literature but on its understanding of the general adoption process.</p><h3>Unknown or subjective topics</h3><p><em>Question #13: Alright. Tell me a bit about the BIAN EDA working group.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ei-LLVLd2l8S7ZjAXljQiQ.png" /></figure><p>Wrong answer. ChatGPT interpreted EDA as <em>Enterprise Data Architecture</em>, which does not exist as a BIAN working group. Also, the BIAN EDA (<em>Event Driven Architecture</em>) working group has yet to make available its work to the public. ChatGPT should acknowledge it does not know about EDA and try to clarify with the user.</p><p><em>Question #14: That is not quite right. By EDA, I meant Event Driven Architecture.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jetoGo55N-B0sYVVq2BZaw.png" /></figure><p>ChatGPT devised the answer based on the common understanding of EDA and BIAN rather than on the work performed by the BIAN EDA working group.</p><p><em>Question #15: Is BIAN well received by its members and the banking industry?</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7ttXRZQvWbjjF0vzZbTSDw.png" /></figure><p>I know the answer can be subjective. ChatGPT gave an answer that is logical and hard to argue against without facts backing its response. Well, how much can we expect from a chatbot of this nature?</p><h3>Summary</h3><p>ChatGPT gave good answers to most of the general factual-based questions that can be inferred from the publicly available information about BIAN. For more complex questions where domain knowledge does not quite exist or may not be available in the public domain, it devised answers based on understanding the elements of the questions. It adapted its responses based on implied relationships in subsequent questions. To enhance the quality of answers to BIAN-related questions, the BIAN community should strive to put more relevant information into the public domain and provide feedback on responses to BIAN-related queries.</p><p>I would love to hear your thoughts on ChatGPT’s understanding of a specific domain like BIAN.</p><h3>Acknowledgment</h3><p>I want to thank Guy Rackham, the BIAN lead architect, for his timely review of the article and many excellent points, some of which I incorporated into the article.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cd595f6afed5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[BIAN Applied to Microservices — Mapping to Domain-Driven Design]]></title>
            <link>https://biaohao.medium.com/bian-applied-to-microservices-mapping-to-domain-driven-design-5aa6a55185c7?source=rss-cf1dc593d6bd------2</link>
            <guid isPermaLink="false">https://medium.com/p/5aa6a55185c7</guid>
            <category><![CDATA[mircoservices]]></category>
            <category><![CDATA[banking]]></category>
            <category><![CDATA[domain-driven-design]]></category>
            <category><![CDATA[architecture]]></category>
            <dc:creator><![CDATA[Biao Hao]]></dc:creator>
            <pubDate>Wed, 03 Feb 2021 04:43:48 GMT</pubDate>
            <atom:updated>2021-02-03T23:59:05.496Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>BIAN Applied to Microservices: Mapping to Domain-Driven Design</strong></h3><p>Many financial institutions are adopting microservices in developing new applications as well as modernizing existing applications. Central to the design of microservices is the partitioning of a large application into a suite of small services, each with clear business capabilities and a defined boundary, running independently and communicating with each other using lightweight protocols such as RESTful APIs and events. Domain-Driven Design, with a set of concepts and patterns for designing and developing a domain model, has been made popular by the increasing adoption of microservices architecture in the industry.</p><p>Banking Industry Architecture Network (BIAN) defines a component business blueprint for the banking industry. BIAN started with specifying a set of Service Domains, each as a discrete, non-overlapping, and elemental business capability, responsible for handling the full lifecycle of its business function. The Service Domains are organized into groups and cataloged in a reference framework called the Service Landscape. BIAN has traditionally been used as a component business architecture for enterprise analysis and applications at a high level. In recent years, BIAN has extended the Service Domain specifications to provide semantic models for describing the business information governed by a Service Domain, and for defining the service operations as the underlying exchanges between the Service Domains. These recent extensions pave the way for leveraging BIAN reference models in the design and development of applications based on microservices architecture.</p><p>Based on my experience of supporting several financial institutions in adopting BIAN in their digital transformation programs, I would like to share an approach to applying BIAN to microservices based applications. In this article, I will establish the mapping between the concepts and models in BIAN and those in the Domain-Driven Design, and illustrate that BIAN provides an industry specific framework for applying Domain-Driven Design. This mapping is to lower the learning curve and to promote a more consistent adoption of BIAN for the practitioners who are familiar with Domain-Driven Design or BIAN. In a future article, I will show the implementation of a microservice for customer financial position where account information is augmented to support new business requirements, using BIAN models and a DevOps tool supporting Domain-Driven Design.</p><h3>Domain Driven Design (DDD) Revisited</h3><p>Domain-Driven Design is an approach to software development that centers on programming a domain model that has a rich understanding of the processes and rules of a domain. The approach, which is particularly suited to complex domains where disorganized logic needs to be organized, can be described through a catalog of concepts and patterns (for details see [1] and [2] in References):</p><p><strong>Domain and Subdomain </strong>A <em>Domain</em> represents what a business does, including all of the knowledge involving the business and how the business operates. It also represents the problems the business faces that you are trying to solve. A <em>Subdomain</em> is a sub-part of the overall business domain, representing a single and logical domain model which breaks up your whole business domain.</p><p><strong>Bounded Context </strong><em>Bounded Context</em> is a semantic contextual boundary; in it, each component has a specific meaning and performs certain tasks. DDD deals with large models by dividing them into different interconnected bounded contexts. <em>Bounded Context</em> is a central pattern in DDD’s strategic design that deals with large models and teams. The integration between <em>Bounded Contexts </em>is known as <em>Context Mapping. </em>There are different styles of integration (such as RPC, RESTful HTTP, and messaging), and various kinds of <em>Context Mapping</em> relationships (such as Partnership, Customer-Supplier, and Anticorruption Layer).</p><p><strong>Ubiquitous Language </strong><em>Ubiquitous Language</em> is a common language that is developed by the team to create the software model that functions with the <em>Bounded Context</em>. It is used among team members and users. The <em>Ubiquitous Language </em>is rigorous, ideally based on the domain model used in the software.</p><p><strong>Entity, Value Object, Aggregate, Service, Service Object, and Domain Event </strong>An<strong> </strong><em>Entity</em><strong> </strong>is an object with a distinct identity that runs through time and different representations. It is also known as a reference object. A <em>Value Object</em><strong> </strong>is an object that matters only as the combination of its attributes. Two value objects with the same values for all their attributes are considered equal. An<strong> </strong><em>Aggregate</em> is a cluster of entities and value objects that can be treated as a single unit. An aggregate has one aggregate root. References from outside the aggregate should only go to the aggregate root. Aggregates are the basic element of transfer of data storage. Each aggregate forms a transactional consistency boundary. This means that within a single aggregate, all composed parts must be consistent when the controlling transaction is committed to the data storage, according to business rules. A <em>Service</em><strong> </strong>is a standalone operation within the context of your domain. A <em>Service Object </em>collects one or more <em>Services</em> into an object. A<strong> </strong><em>Domain Event</em> is a record of some business-significant occurrence in a bounded context.</p><h3><strong>Banking Industry Architecture Network (BIAN) Distilled</strong></h3><p>BIAN is an association of banks and solution providers with a shared vision of defining a component business blueprint for the banking industry. The figure below shows the BIAN reference model, including the Service Landscape (left side) and Service Domain specification (right side). More details about BIAN can be found in the practitioner’s guide ([3] in References).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nJfzZO22aTDPvaB37Xjs3g.png" /><figcaption>BIAN Reference Model</figcaption></figure><p><strong>Service Domain </strong><em>Service Domain</em> is the foundational building block of the BIAN model to represent the business functional components. A <em>Service Domain</em> is defined to be responsible for implementing one pattern of behavior to instances of one type of business asset (<strong><em>asset type</em></strong>). The asset is what the business possesses, including tangible things such as buildings and computers, or intangible things such as relationships and knowledge. The pattern of behavior, representing the way an asset is controlled or leveraged, is called a <strong><em>functional pattern</em></strong>. A Service Domain has the following architectural properties:</p><ul><li>A discrete, non-overlapping, and elemental business capability, responsible for handling the full lifecycle of its business function. As a result, a Service Domain encapsulates its business information and logic, combining people, processes, and systems. BIAN adopts a completely different view of business activity compared to the more conventional process centric approaches.</li><li>Stable over time, supporting incremental development and adoption. It has a business role or purpose that is defined independently of the way it might be implemented. As business practices and techniques evolve, the inner working of a Service Domain may be enhanced with additional features.</li><li>Canonical, or the ‘same for everyone’, supporting high levels of interoperability. It defines the generic functional building blocks that make up any bank. Its role/purpose can be consistently interpreted from one deployment to another. This means a bank or solution provider can switch out the underlying system for a Service Domain and replace it with an alternative solution.</li></ul><p>BIAN members have spent several years defining a comprehensive collection of the Service Domains that cover most banking activities.</p><p><strong>Service Landscape </strong>The <em>Service Landscape</em> organizes all the <em>Service Domains</em> into groups based on large <strong><em>business areas</em></strong> and more narrowly defined <strong><em>business domains</em></strong> within these areas. The layout is simply intended to help with the identification and selection of Service Domains.</p><p>In recent years, BIAN has extended the Service Domain specifications to include Control Records, Behavior Qualifiers, Information Profile and Business Object Model, and Service Operations.</p><p><strong>Control Record </strong>The <em>Control Record</em> is defined as the combination of the asset type and the generic artifact that characterizes the functional pattern of the Service Domain. The <em>Control Record</em> is comprised of all the business information governed by the Service Domain as it completes a full cycle of its work. For example, for <em>Current Account</em> Service Domain, the asset type is <em>Current Account</em>, and the generic artifact for <em>Fulfill</em> functional pattern is <em>Arrangement</em>; therefore, the Control Record is <em>Current Account Arrangement</em>.</p><p><strong>Behavior Qualifier </strong>A further level of breakdown for the Control Record is often required to define a sufficiently narrow business context. The <em>Behavior Qualifier</em> is introduced to characterize the finer grained activity that is a part of the overall work done by a functional pattern. The <em>Behavior Qualifier</em> retains the core behavioral characteristics of its associated functional pattern. For example, for <em>Current Account</em> Service Domain, Behavior Qualifiers such as <em>Direct Debit</em>, <em>Deposits</em>, <em>Withdrawals</em>, and <em>Payments</em> are used to describe the specific features of the Service Domain characterized by the <em>Fulfill</em> functional pattern. Additional levels of breakdown can be applied consistently as needed.</p><p><strong>Information Profile and Business Object Model </strong>In addition to the Control Record instances (and the contained Behavior Qualifier instances), <em>Information Profile </em>contains information used in the control and management of the Service Domain as a service center and the analysis and reporting of usage and performance of the Control Record instances the Service Domain manages. Using the modeling approach for <em>Information Profile</em>, the recent BIAN <em>Business Object Model (BOM)</em> provides a more precise conceptual basis to describe the business semantics and data of BIAN Service Domains. The BOM contains models for a growing number of Service Domains, Control Records, and mappings to ISO20022 business components and elements.</p><p><strong>Service Operation and Service Group </strong>A Service Domain offers a collection of service operations and usually consumes the service operations of other Service Domains as needed to complete its work. BIAN defines the <em>Service Operation</em> to indicate a service dependency between Service Domains. BIAN defines a general set of <strong><em>action terms</em></strong> that characterize the purpose of an offered service. These action terms can be grouped into categories. A Service Operation specification is defined using the applicable action term and optionally a behavior qualifier, with payload specified as an organized list of semantic attributes (as BOMs) covering the key business information provided and returned for the service exchange. The <em>Service Operations</em> are grouped into <em>Service Groups</em> based on the categories of the action terms.</p><h3><strong>Mapping Between BIAN and DDD</strong></h3><p>The BIAN approach to specifying Service Domains using the asset-leverage model (one asset type + one functional pattern) provides a unique domain decomposition model for the banking business domain. The BIAN reference model also provides a ubiquitous language for both business and technical practitioners in describing and modeling business capabilities. BIAN concepts and models, such as Control Record, Behavior Qualifier, and BOM, can be used to construct DDD domain models such as Entity and Aggregate. The figure below shows the mapping between BIAN and DDD.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x9j5EiY5o2Pl5zowLLhMbw.png" /><figcaption>Mapping Between BIAN and DDD</figcaption></figure><p><strong>BIAN deals with the DDD Domain and Subdomains for overall banking business </strong>As a conceptual business component framework for banking, BIAN decomposes the overall banking business Domain into a set of Subdomains called BIAN Service Domains. The BIAN Service Landscape organizes all the BIAN Service Domains into larger Business Areas and smaller Business Domains, which are Subdomains themselves. The BIAN Service Landscape and Service Domains provide a generic reference framework from which a component design blueprint for a particular bank can be derived.</p><p><strong>BIAN Service Domains map to DDD Bounded Contexts </strong>Each BIAN Service Domain, as a functional building block, defines a Bounded Context, a contextual boundary in which an asset is leveraged in a specific way. The BIAN asset-leverage model provides a unique and consistent way of dividing the large and complex banking Domain into a set of discrete, non-overlapping, interconnected Bounded Contexts.</p><p><strong>BIAN provides a Ubiquitous Language </strong>BIAN also provides a common language to describe business capabilities, what a bank does, what it needs to know, and what services are offered and required. It is developed by the member banks and has been increasingly adopted by the banks and software providers. The common language is rigorous and has been incorporated into the business architecture and point solutions by banks and solution providers. The common language increases the interoperability in one project or between projects within a bank, between banks and solution providers.</p><p><strong>BIAN Control Records and Behavior Qualifiers map to DDD Aggregates </strong>BIAN Control Record consists of the business information for completing a full cycle of its work. BIAN Behavior Qualifier is a breakdown of the Control Record to define narrower business contexts and more specific purposes for the service operation. All composed parts of a Control Record or Behavior Qualifier must be consistent when the controlling transaction is committed. BIAN Control Records and Behavior Qualifiers map to DDD Aggregates. BIAN BOMs, which are used to model Control Record, Behavior Qualifier and payload of Service Operation, map to DDD Domain Objects (Aggregate, Entity and Value Object).</p><p><strong>BIAN Service Operations and Service Groups map to DDD Services and Service Objects </strong>A BIAN Service Domain offers a collection of Service Operations that are consumed by other Service Domains to complete their work. BIAN Service Operations map to DDD Services while the Service Groups maps to the DDD Service Objects.</p><p><strong>BIAN Events map to DDD Domain Events </strong>BIAN has focused on defining Service Operations, which are used in conventional orchestration that coordinates service calls between Service Domains. To support event-based exchange, BIAN is currently extending the Service Domain specifications to include the Events produced and consumed by the Service Domains. These Events map to DDD Domain Events.</p><h3><strong>Summary</strong></h3><p>In this article, I demonstrated that BIAN can be used as a framework for the Domain-Driven Design approach by establishing the relationship between BIAN concepts and models and the corresponding ones in the Domain-Driven Design. BIAN can be used at a strategic design level by providing a unique and consistent way of decomposing the banking business domain into a set of Bounded Contexts. It can also be used at a tactic level where the information models are used as the starting point for designing and constructing domain objects such as Aggregates and Entities.</p><h3><strong>References</strong></h3><p>[1] Eric Evans,<a href="https://www.amazon.com/Domain-Driven-Design-Tackling-Complexity-Software/dp/0321125215"> Domain-Driven Design: Tackling Complexity in the Heart of Software</a>, 2003</p><p>[2] Vernon Vaughn,<a href="https://www.amazon.com/Domain-Driven-Design-Distilled-Vaughn-Vernon-ebook/dp/B01JJSGE5S"> Domain-Driven Design Distilled</a>, 2016</p><p>[3] Guy Rackham,<a href="https://bian.org/wp-content/uploads/2020/07/BIAN-Semantic-API-Pactitioner-Guide-V8.1-FINAL.pdf"> BIAN Semantic API Practitioner’s Guide V8.1</a>, 2020</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5aa6a55185c7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Global Open Banking Landscape]]></title>
            <link>https://biaohao.medium.com/global-open-banking-landscape-cc6ae03991bc?source=rss-cf1dc593d6bd------2</link>
            <guid isPermaLink="false">https://medium.com/p/cc6ae03991bc</guid>
            <category><![CDATA[open-banking]]></category>
            <category><![CDATA[regulation]]></category>
            <category><![CDATA[banking]]></category>
            <category><![CDATA[fintech]]></category>
            <dc:creator><![CDATA[Biao Hao]]></dc:creator>
            <pubDate>Tue, 23 Jun 2020 03:25:25 GMT</pubDate>
            <atom:updated>2020-07-16T21:57:06.415Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Biao Hao, Executive Architect, IBM Global Solution Center in Dallas | Bharat Bhushan, CTO, Banking &amp; Financial Markets, IBM Europe</em></p><p>Open Banking is transforming the world of banking, from the way consumers will use banking services, to how it will reorganize and flatten the banking industry, and lower the barriers to entry. In 2015, European Parliament adopted Revised Payments Services Directive (or PSD2 in short), with the intention to increase competition and promote innovation in the payments industry. Since then, PSD2, as a regulation driven approach, has been adopted by regulators in many countries around the world. In countries without regulations, Open Banking is driven mostly by FinTechs and banks, with the desire to provide their customers new or better products and services. Even in countries with regulations, regulators and financial institutions are taking different spins and perspectives on this topic. For example, the UK regulator has defined a very prescriptive standard for the top 9 banks to adopt, while the remaining 5,000 banks in Europe have chosen to follow whatever standards they like, as long as they meet the spirit of the regulation.</p><p>These varied approaches in the market often create confusion among regulators and financial institutions. As a global organization that is present in over 90% of financial institutions, IBM can create a unique perspective on the topic and demonstrate what financial institutions can do to comply or compete in this area.</p><p>Global Open Banking Landscape, an initiative sponsored by IBM’s Academy of Technology and Industry Academy, has set out to map the current developments and state of adoption of Open Banking around the world. With contributions from more than 80 IBM volunteers across 20 countries and 5 continents, the initiative produced a report (refer to the slide deck below) covering regulatory and industry drivers, regulations and technical standards, use cases, adoption levels by financial institutions and consumers, and advice on how to turn the challenge of Open Banking into an opportunity.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.slideshare.net%2Fslideshow%2Fembed_code%2Fkey%2FkfT1YXgsZLUaN0&amp;display_name=SlideShare&amp;url=https%3A%2F%2Fwww.slideshare.net%2FBiaoHao%2Fglobal-open-banking-landscape&amp;image=https%3A%2F%2Fcdn.slidesharecdn.com%2Fss_thumbnails%2Faot-globalopenbankinglandscapev1-200621043820-thumbnail-4.jpg%3Fcb%3D1593021587&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=slideshare" width="600" height="500" frameborder="0" scrolling="no"><a href="https://medium.com/media/70924343540f6a1686b26f905d43b61b/href">https://medium.com/media/70924343540f6a1686b26f905d43b61b/href</a></iframe><p>Below are excerpts from the report.</p><p><strong>Regulation vs. Market Driven</strong></p><p>As illustrated in the diagram below (chart 5 of the slide deck), regions with Open Banking regulations include Australia, Brazil, European Economic Area, India, Japan, and United Kingdom, while market driven countries include Canada, China, New Zealand, Singapore, Sri Lanka, Switzerland, Turkey, United Arab Emirates, and the United States.</p><figure><img alt="Maps of Regions" src="https://cdn-images-1.medium.com/max/1024/1*BA_9gPlUgtWuByEZhOMyEQ.png" /></figure><p><strong>Opportunities and Prioritization Matrix</strong></p><p>Open Banking use cases, represented as opportunities, are ranked based on complexity, business impact, and customer value as illustrated in the diagram below (chart 6 of the slide deck).</p><figure><img alt="Opportunities and Prioritization Matrix" src="https://cdn-images-1.medium.com/max/1024/1*qbONylcSusvqjJKd1q4aYQ.png" /></figure><p>Short summary of select countries: China, India, Singapore, Australia, the United States, and Brazil.</p><p><strong><em>China</em></strong></p><p>While there are no regulations, Open Banking is widely adopted by FinTechs and large banks to grow their customer base and meet the growing needs of their customers.</p><ul><li>Large banks build their own ecosystems with broad open banking capabilities including consumer and business accounts / lending, payments and settlement, investment and wealth management, foreign currency exchange, rewards, bank information, and more.</li><li>“Super apps” from large FinTechs have a huge customer base (Alipay has 400+M users, and WeChat has 600+M users). Smaller banks leverage these “super apps” to deliver their financial services capabilities.</li></ul><p><strong><em>India</em></strong></p><p>The banking industry in India has two flavors of Open Banking, Account Aggregators, initiated by Reserve Bank of India, and voluntary adoption by the banking industry and FinTechs.</p><ul><li>In 2016, the Indian banking regulator, Reserve Bank of India (RBI), adopted concepts from Open Banking / PSD2 via a directive for the registration and operations of Account Aggregators (AAs). The AAs, similar to TPPs, facilitate the exchange of customer financial data among banks, FinTechs, and other parties.</li><li>From 2009 to 2010, the Government of India had already started building unique identification (UID — Aadhaar) for India and had made the API available. Subsequently, the Government of India built various APIs, known as the India Stack, for nationwide use. This included Unified Payments Interface (UPI) that unleashed changes in Indian payments, banking and FinTech industries.</li></ul><p><strong><em>Singapore</em></strong></p><p>Currently, there is no specific data sharing or Open Banking legislation in Singapore. However, the government provides a guideline for Open Banking APIs, which are defined via a publicly available infrastructure.</p><ul><li>In November 2016, the Monetary Authority of Singapore (MAS), in collaboration with the Association of Banks in Singapore (ABS), published a comprehensive roadmap — <em>Finance-as-a-Service: API Playbook</em>. The playbook set out a comprehensive framework that introduced governance, implementation, and use cases and design principles for APIs.</li><li>API Exchange (APIX) is an open-architecture API marketplace and sandbox platform for collaboration between FinTechs and financial institutions in which participants can integrate and test solutions with each other via a cloud-based architecture.</li></ul><p><strong><em>Australia</em></strong></p><p>In 2018, the Australian government established the Consumer Data Right (CDR), the legislation to give Australians greater control over their data. Subsequently, in 2019, the Consumer Data Standards (CDS) was defined as the technical standards. Australia has several unique Open Banking features including:</p><ul><li>Customer Experience Guidelines — The regulator has undergone consumer testing and experience guidelines research to provide recommendations to the market surrounding what the Open Banking customer flows and consent experience should look like to ensure that the customer experience across banks is consistent.</li><li>Not Just Banking — The CDR legislation is intended to be applied sector by sector across the whole economy, starting with the banking, energy and telecommunications sectors.</li><li>Exposing Payees and Direct Debits — The intent is to allow for the automation of transferring payees and direct debits from competing banks, effectively alleviating the pain and anxiety surrounding switching banks.</li></ul><p><strong><em>United States</em></strong></p><p>Open Banking in the United States is defined by growing consumer demand without any legislative mandate. Successful apps, such as Mint and YNAB, which both offer a single view of a user’s accounts across multiple financial services providers, have prompted banks to provide access to customer account data using APIs, which provide better performance and security than screen scraping. OFX is a legacy standard that was adopted by many Personal Financial Management (PFM) solutions. Financial Data Exchange (FDX) is a new technical standard gaining traction in the U.S.</p><ul><li>The Financial Data Exchange (FDX) is a non-profit organization dedicated to promoting and enhancing a common interoperable standard and operating framework for sharing consumer financial data.</li><li>FDX is open to all financial institutions and FinTechs, with 94 members as of February 2020.</li><li>FDX counts over 8 million consumers at the beginning of 2020, and is expected to have 12 million consumers by April 2020.</li></ul><p><strong><em>Brazil</em></strong></p><p>The financial system is dominated by five large traditional banks. The majority of the population is still unbanked, with very limited access to financial products such as loans and personal credit. Those that do have access to credit face the highest interest rates in the world. Banco Central do Brasil (BCB or Bacen) is the financial system regulator. Its main objective is to foster competition among the large banks, and also allow new entrants such as Neo Banks and FinTechs to create opportunities for lower interest rates and more diverse financial products. Bacen first issued a public consultation for feedback in November 2019. In May 2020, Bacen published the regulation with an implementation calendar (detailed below). In the initial stage, only banks classified as tier S1 and S2 (Regulation #4453/2017) have to comply.</p><ul><li>Phase 1 (November 2020): Share information about available products and services and provide access to data about channels, products and services related to checking and savings accounts, and payments and credit.</li><li>Phase 2 (May 2021): Share customer personal information and transactional data (accounts and credit).</li><li>Phase 3 (August 2021): Initiate payments transactions and credit service proposals.</li><li>Phase 4 (October 2021): Provide information about currency exchange, investments, insurance and retirement funds.</li></ul><p><strong>Acknowledgements</strong></p><p>We would like to thank Chae An and Dr. Thomas Hager for executive sponsorship, Elisabeth Stahl and Connor Loessl for support, and Rahul Kinger for designing and compiling the report. Special thanks to all the participants for their contributions, especially the team leaders, Alex Cuthbert, Alberto Miyazaki, Alden Chan, James Sanders, Chuan Jun Liao, Jurgen Lang, Tim Allcock, Pal Krogdahl, Rhujuta Kene, Deepak Raghaven, Mikio Kamada, Rika Sugiyama, Guy Perera, James Stephens, Kamran Sheikh, and Sevilay Kurt, for leading the teams that compiled contents for 18 countries.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cc6ae03991bc" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>