<?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 Business and Technology University  (BTU) on Medium]]></title>
        <description><![CDATA[Stories by Business and Technology University  (BTU) on Medium]]></description>
        <link>https://medium.com/@mariam.kobiashvili?source=rss-45270e692e56------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*4YqVk-s4E7bXb88EblSaSA.png</url>
            <title>Stories by Business and Technology University  (BTU) on Medium</title>
            <link>https://medium.com/@mariam.kobiashvili?source=rss-45270e692e56------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 16:02:57 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@mariam.kobiashvili/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[AI Can Find the Best Answer. Who Decides What “Best” Means?]]></title>
            <link>https://medium.com/@mariam.kobiashvili/ai-can-find-the-best-answer-who-decides-what-best-means-76a738c41256?source=rss-45270e692e56------2</link>
            <guid isPermaLink="false">https://medium.com/p/76a738c41256</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[georgia]]></category>
            <category><![CDATA[ai-governance]]></category>
            <category><![CDATA[ai-ethics]]></category>
            <category><![CDATA[decision-making]]></category>
            <dc:creator><![CDATA[Business and Technology University  (BTU)]]></dc:creator>
            <pubDate>Thu, 19 Nov 2026 10:35:18 GMT</pubDate>
            <atom:updated>2026-11-19T10:35:18.607Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hFsSxsxvHQ9VUN0J6dIphg.png" /></figure><h3>Main</h3><p>BTU’s source comparison finds that choosing an objective, assigning accountability and deciding when a person must intervene are separate governance questions. A small computational illustration makes the distinction concrete: identical service data can support different correct selections when the objective changes.</p><h3>Key Takeaways</h3><ul><li>AI’s ability to recommend a policy does not itself authorize that policy on an organization’s behalf.</li><li>In BTUAI’s six-case constructed example, two objectives select different cases under three time limits but converge when all cases fit. These results are not an AI benchmark or observed business outcomes.</li><li>Georgia’s rules for consequential, solely automated decisions illustrate why governance must be grounded in local law rather than a universal “human approves everything” slogan.</li></ul><h3>Part One · The choice hidden inside the answer</h3><p>Imagine receiving a perfectly accurate analysis that still does not tell you what to do. A customer-service team has limited time. One option closes more requests. Another addresses requests from customers who have already waited longer. Every number is correct. The disagreement remains because the organization has not resolved what it wants to prioritize.</p><p>AI can help expose that disagreement. It can explain alternative policies, test their implications and identify missing information. The important boundary is not between a machine that only calculates and a person who alone can reason. Modern AI can participate in reasoning too. The boundary is between producing a recommendation and possessing the authority to determine which trade-off an organization will accept.</p><p>This is what a careful claim about AI and human judgment should mean. It is not proof that machines will never perform a particular cognitive task. It is a claim about how responsibility is organized when decisions affect others. A better model can make a choice easier to understand without eliminating the institution that must justify it.</p><h3>More information and a better objective are different improvements</h3><p>Some decisions are difficult because a fact is missing. A company may not know which delivery date it promised or whether a document is current. Better retrieval can help. Other decisions remain difficult even after the relevant facts are known. A company may understand the consequences of several options yet still need to decide which commitment takes precedence.</p><p>Treating both problems as information shortages creates a misleading expectation: eventually, enough data will reveal the uniquely correct policy. Often the policy has already been embedded in the definition of success. A dashboard that rewards closed cases encourages one behavior; a dashboard that gives priority to older unresolved cases encourages another. AI can implement either measure accurately.</p><p>Human judgment is not automatically better. People misunderstand evidence and accept convincing recommendations without sufficient scrutiny. A person placed at the end of an automated process may have neither the information nor the authority to change anything. A final approval click therefore should not be confused with meaningful control.</p><p>NIST’s AI Risk Management Framework makes room for different human–AI arrangements, including systems that may not need human oversight. UNESCO’s ethical recommendation, meanwhile, locates ultimate accountability with people and existing legal entities. These positions can coexist: an organization can automate particular actions while remaining responsible for the objectives, limits and consequences of the system.</p><h3>Georgia brings the distinction into focus</h3><p>For Georgian organizations, a generic international demonstration is not a deployment assessment. The system must handle the actual Georgian-language request, the organization’s policy and the applicable law. A customer may refer indirectly to an earlier promise that changes the significance of an otherwise routine message. Correct translation alone does not establish that the case has been understood.</p><p>Georgia’s Personal Data Protection Law provides a concrete local boundary. Article 19 concerns solely automated decisions with legal or other significant effects, specifies exceptions for profiling-based decisions and provides safeguards. It is not a ban on every automated task. The Georgian original also qualifies its human-involvement provision for statutory or delegated-law cases under paragraph 1(c). Applying the provision requires examining the real process and the full legal context, not merely whether someone used AI.</p><h3>Part Two · What BTUAI examined</h3><p>On 6 November 2026, BTUAI compared selected provisions from NIST AI RMF 1.0, UNESCO’s ethical recommendation and Georgia’s law. The same four questions guided the reading: scope, accountability, human involvement, and qualifications or exceptions. The sources differ in legal status: a voluntary framework, an ethical recommendation and national legislation are not interchangeable rules.</p><p>The resulting distinction is useful. Organizations need to identify the objective, the accountable decision-maker and the conditions for intervention separately. A narrow automation policy may answer the last question while leaving the first two unresolved. Conversely, keeping an accountable institution does not require a person to authorize every low-risk technical operation.</p><h3>A calculation that makes the hidden objective visible</h3><p>We then constructed six non-urgent service cases. Their processing times were 10, 15, 20, 30, 40 and 50 minutes; their respective prior waiting times were 1, 2, 3, 12, 15 and 18 days. These are deliberately invented inputs, not measurements from a company. No customer data or AI response benchmark was involved.</p><p>A Python calculation evaluated all 64 possible subsets. One rule maximized the number of completed cases. The other maximized the selected cases’ combined days already spent waiting. Ties used the other objective, then lower processing time. Cases had to be completed in full, with no parallel work or new arrivals.</p><p>With 60 minutes available, the first rule selected A, C and D: three cases, 60 minutes, and 16 days of prior waiting combined. The second selected A and F: two cases, the same 60 minutes, and 19 days of prior waiting. Neither calculation was wrong. The objective changed the answer.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/733/1*lxmmXgnKwELmsS6R-FcEmw.png" /></figure><p><strong>Parentheses show selected case counts. Source: BTU calculation using authored inputs, not observed service outcomes.</strong></p><p>The selections also differed at 90 and 120 minutes. At 165 minutes, both rules selected every case. This convergence matters: the illustration does not imply that objectives always conflict. Nor does the waiting-days total measure harm, fairness or time saved. A real service policy might require deadlines, accessibility needs and other safeguards absent from this intentionally simplified example.</p><p>The calculation is a transparent illustration of objective sensitivity, not evidence that AI cannot propose or choose objectives. Its practical contribution is to make an apparently neutral recommendation easier to interrogate. Asking “Is the arithmetic correct?” is necessary. Asking “What did we ask it to optimize?” may be just as important.</p><h3>A stronger role for AI, a clearer role for institutions</h3><p>The sensible ambition is not to keep people busy approving everything. It is to use AI to make relevant facts, alternatives and consequences more visible, while keeping authority and accountability explicit. A responsible reviewer needs enough information to challenge a result and a real route to change it.</p><p>AI can improve the preparation of a decision. It cannot, by its analytical capability alone, release an organization from explaining why it chose one acceptable objective over another. That obligation is where the discussion of human judgment becomes useful for actual businesses.</p><h3>Questions and Answers</h3><h3>Does accountability require manual approval of every action?</h3><p>No. Controls should reflect the task, risk and applicable law. Responsibility for a system differs from approving each technical operation.</p><h3>Can AI recommend the objective too?</h3><p>Yes. Recommending objectives and explaining their consequences can be valuable. Institutional authorization is a separate step.</p><h3>Did this study show that people outperform AI?</h3><p>No. It combined documentary comparison with a constructed numerical illustration. No human-versus-model performance test occurred.</p><h3>Why report the equal result at 165 minutes?</h3><p>Because it limits the conclusion. When all six cases fit, these two objectives select the same set; disagreement is conditional rather than inevitable.</p><p>Full BTUAI Article: <a href="https://btuai.ge/en/why-ai-cannot-replace-human-judgment/">https://btuai.ge/en/why-ai-cannot-replace-human-judgment/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=76a738c41256" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why the cheaper kilogram can produce the larger parcel bill]]></title>
            <link>https://medium.com/@mariam.kobiashvili/why-the-cheaper-kilogram-can-produce-the-larger-parcel-bill-c1075fcadd84?source=rss-45270e692e56------2</link>
            <guid isPermaLink="false">https://medium.com/p/c1075fcadd84</guid>
            <category><![CDATA[consumer-behavior]]></category>
            <category><![CDATA[business]]></category>
            <category><![CDATA[logistics]]></category>
            <category><![CDATA[georgia]]></category>
            <category><![CDATA[ecommerce]]></category>
            <dc:creator><![CDATA[Business and Technology University  (BTU)]]></dc:creator>
            <pubDate>Thu, 19 Nov 2026 10:26:46 GMT</pubDate>
            <atom:updated>2026-11-19T10:26:46.451Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0QqtWcGE9Qm3Q222NeIuRQ.png" /></figure><h3>Main summary</h3><p>Public parcel-forwarding prices offer a useful lesson about comparison: the unit price is only meaningful when the unit is defined consistently. A BTU observation of US to Georgia forwarding conditions shows how packaging and billing rules can overturn the apparent advantage of a lower kilogram rate.</p><h3>Main conclusions</h3><ul><li>A forwarding quote needs both a rate and a rule for chargeable weight; a large, light box can make the second more important than the first.</li><li>In twenty predefined parcel conditions, a comparison of two published standard rules changed direction as packaging changed. Six conditions favored Spacecargo’s transport component and fourteen favored Inex’s, without establishing a market-wide winner.</li><li>Transparent comparisons require coherent terms, visible additions and a distinction between transport cost and the complete amount paid for an imported purchase.</li></ul><h3>The missing quantity behind a visible price</h3><p>Imagine comparing two electricity bills without checking whether both prices refer to the same quantity of energy. Parcel forwarding presents a related problem. Prices are often advertised per kilogram, yet the number of chargeable kilograms can differ across providers for the same physical object. The rate is visible; the quantity it multiplies requires more work.</p><p>BTU studied this problem through public conditions for consumer air parcels moving from US warehouses to collection in Tbilisi. Eight forwarders were selected before substantive observation. Their pages were examined on 13 November 2026. The research combined source comparison with predefined mass and dimension conditions. It did not purchase shipments or measure delivery quality.</p><p>This distinction matters because the goal was to understand price architecture. A calculation based on a rule can demonstrate how an offer behaves, but it cannot tell us what typical customers buy or whether packages arrive safely. The relevant economic finding is a conditional relationship, not a consumer prevalence estimate.</p><h3>The box changes the comparison</h3><p>Inex’s published standard US terms specify USD 9.70 per physical kilogram. Spacecargo’s agreement specifies USD 8.85 and uses volumetric weight when it exceeds physical weight. Under the documented dimensional formula, the three external measurements in centimetres are multiplied and divided by 6, 000. A box can therefore carry more billing kilograms than kilograms on a scale.</p><p>BTU fixed four physical masses and five box sizes, creating twenty comparable conditions. Each was evaluated under both rules. The comparison holds the route, customer type and collection scope constant and excludes promotions, optional courier service, customs charges and the purchase price. It compares a transport component rather than pretending to calculate every buyer’s eventual invoice.</p><h3>Same route and standard transport component in USD</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/692/1*puJJ3FqfsLL33ZExLIkhYQ.png" /></figure><p>Sources: Inex and Spacecargo public conditions; BTU calculations, 13 November 2026. Excludes promotions, currency conversion and additional services. The range reflects rounding sensitivity, not a provider quotation.</p><p>The two 2 kg examples make the mechanism visible without changing the item’s mass. A compact 30×20×20 cm box has 2 kg volumetric weight. Spacecargo’s component is USD 17.70 against Inex’s USD 19.40. Increase the box to 40×30×30 cm and volumetric weight becomes 6 kg. The same rules now yield USD 53.10 and USD 19.40 respectively.</p><p>Across the deliberately selected conditions, Spacecargo is lower in six cases and Inex in fourteen. Those counts are useful as a stress test of the rules. They are not shares of real parcels or evidence of how many buyers would save. A different mix of boxes would change the count.</p><p>The underlying boundary is more portable than the count. Before rounding, the lower rate loses its advantage when billable mass is roughly 9.6% above actual mass. Spacecargo’s published rate is about 8.8% lower than Inex’s, using the latter as the denominator. A small headline discount can thus be outweighed by the definition of the billing quantity.</p><h3>Precision means acknowledging the rule that is missing</h3><p>One box in the set produces 13⅓ dimensional kilograms. Spacecargo’s document specifies 100 g precision for the relevant weight range but does not specify the direction of rounding in that provision. The study therefore uses adjacent 13.3–13.4 kg bounds. For the 5 kg example, this gives USD 117.71–118.59, rather than inventing a definitive quote. Both endpoints remain above the compared USD 48.50 physical-weight component.</p><p>Other providers add different conditions. Onex’s published oversized-parcel rule requires both a dimension sum above 150 cm and volumetric weight at least twice actual weight. Its description also refers to a discount. Without the confirmed discount, the researchers classified the condition but did not manufacture a payable amount.</p><p>The distinction between a general agreement and a category-specific exception also required care. Camex and Amanati had public descriptions whose scope or priority needed clarification before a single numerical rule could be assigned. They were retained in the source assessment and excluded from the unambiguous price calculation. This is a limitation of the available comparison, not a finding about actual billing or legal compliance.</p><h3>More information is not necessarily a clearer offer</h3><p>The observed issue is not simply that shoppers need to read more. They need pieces of information to fit together. A price page, a service agreement and a news announcement can each be understandable separately while leaving uncertainty about which rule applies today to a particular parcel.</p><p>A further layer is unavoidable additions. USA2GEORGIA’s terms, for example, include an operational handling component linked to declared value. Currency conversion adds another distinction between a dollar calculation and the GEL settlement amount. A comparison that ignores these layers may be arithmetically correct but answer the wrong economic question.</p><p>Packaging itself deserves attention. If every external dimension rises by 5%, volume grows by approximately 15.8%. Where volume determines a continuously calculated transport component, the effect follows that multiplication rather than the change in any single side. This does not justify stripping necessary protection from fragile goods. The relevant choice includes damage risk, repacking conditions and fees, not just empty space.</p><p>BTU’s analytical use of Georgian consumer research materials puts this observation in the context of price clarity and delivery predictability. A useful business response would be a quote showing physical mass, dimensional mass, the applied exception and additions as separate lines. Any parameter awaiting warehouse measurement should be visible. Whether this improves perceived trust is a testable question, rather than an outcome demonstrated by this study.</p><h3>The practical conclusion</h3><p>A parcel, with defined external dimensions and service conditions, is a better comparison unit than a bare kilogram rate. That principle does not identify one universally superior provider. It identifies the information needed to make a defensible decision. The most helpful price is not necessarily the shortest price label, but the one a buyer can connect to the eventual account.</p><h3>Questions and answers</h3><h3>Is this a ranking of forwarders</h3><p>No. The calculations isolate documented standard transport components. Service performance and complete landed costs require additional evidence.</p><h3>Can a promotion change the result</h3><p>Yes. It changes the rate and therefore the comparison boundary. Weight rules and additional charges still need to be checked.</p><h3>What does the rounding interval mean</h3><p>It preserves uncertainty in a source provision instead of assigning a rounding direction without evidence. The illustrated price ordering survives either endpoint.</p><h3>What should follow this pilot</h3><p>A consent-based comparison of official quotes and final invoices would test billing predictability directly. Public rules establish the framework; transaction evidence would test its application.</p><p>Full BTUAI Artcile: <a href="https://btuai.ge/en/parcel-prices-and-the-size-of-the-box/">https://btuai.ge/en/parcel-prices-and-the-size-of-the-box/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c1075fcadd84" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[An AI Agent Should Make Your Judgment More Useful]]></title>
            <link>https://medium.com/@mariam.kobiashvili/an-ai-agent-should-make-your-judgment-more-useful-42e6fa4f7d54?source=rss-45270e692e56------2</link>
            <guid isPermaLink="false">https://medium.com/p/42e6fa4f7d54</guid>
            <category><![CDATA[human-ai-collaboration]]></category>
            <category><![CDATA[decision-making]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[business-productivity]]></category>
            <dc:creator><![CDATA[Business and Technology University  (BTU)]]></dc:creator>
            <pubDate>Wed, 18 Nov 2026 12:41:15 GMT</pubDate>
            <atom:updated>2026-11-18T12:41:15.917Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mDgYfWoiuBVY73hDPxOBeg.png" /></figure><h3>Main</h3><p>An agent can prepare a better basis for a decision without becoming the decision-maker. A small Georgian-language text trial shows what that preparation can contain, while stopping short of claiming that anyone became more productive.</p><h3>Key Takeaways</h3><ul><li>Human augmentation requires a better human outcome, not merely a well-written AI response.</li><li>BTUAI’s six authored Georgian cases met 23 criteria unambiguously; the author-scorer also credited one conditional answer. No human performance was measured.</li><li>Useful decision support preserves uncertainty, exposes unsupported assumptions and leaves consequential choices with the responsible person.</li></ul><h3>Part one The work before the decision</h3><p>The most useful AI agent may be the one that stops you from acting on a plausible mistake. A preliminary note says the warehouse can supply an order. Another says a new process will save ten hours. A third assumes that the newest contract file is the approved one. Each sounds actionable. Each may hide a question that should be answered first.</p><p>That is where the argument for human augmentation becomes concrete. An agent can retrieve information, compare records and check calculations before a person chooses what to do. The resulting division of work is valuable only if the person can understand, inspect and challenge the output. Otherwise, a decision has merely acquired a more fluent intermediary.</p><p>An agent is more than a text generator when it can use tools, choose steps and respond to results. Yet a satisfactory paragraph tests only part of that larger system. Technical guidance from Anthropic distinguishes predefined workflows from dynamically directed agents and recommends adding complexity only where it helps. The label alone tells an employer little about usefulness.</p><p>A knowledgeable employee brings priorities, context and responsibility to a task. The system can bring organized information. These contributions are complementary when the employee sees why a conclusion follows and what remains unknown. They are less complementary when a confident recommendation hides its assumptions. Agreement is not always assistance.</p><h3>The local setting changes the test</h3><p>Georgia, the country in the South Caucasus, offers a practical setting for this question. Geostat reports that 94.9% of its enterprises had internet access in 2025, while 15.3% used a website. Neither figure measures the consistency of internal records or readiness for AI. Connectivity and usable organizational knowledge are different requirements.</p><p>A Georgian-language handoff must preserve a deadline, quantity, exception or condition as accurately as it preserves the general meaning. An employee should not need to reconstruct the original material merely to decide whether the summary is trustworthy. International demonstrations do not establish that local working documents meet this test.</p><p>Research involving people provides a different kind of evidence. Brynjolfsson, Li and Raymond’s 2025 study covered 5,172 support workers and found a 15% average increase in issues resolved per hour with AI assistance, with unequal effects across experience levels. It was one firm, not a forecast for Georgian organizations or every autonomous agent.</p><p>The organizational question is therefore specific: which obstacle is being removed? Finding a document, reconciling quantities and identifying an unapproved term are different problems. Review effort also matters. If every output must be rebuilt from scratch, apparent automation has created another queue rather than usable capacity.</p><h3>Part two Six preliminary notes put to the test</h3><p>BTUAI ran a bounded Georgian-language text trial on 6 November 2026. Six cases were written for the study, each containing records and a flawed preliminary conclusion. They concerned training-centre payments, a hotel quotation, stock availability, customer ratings, a design studio’s time plan and contract versions. No real employees or businesses supplied these cases.</p><p>The responding system was asked to assess the note, explain the checkable basis, identify an unresolved question and propose a next step. Four substantive criteria per case were fixed before execution: evidence-consistent interpretation, numerical or record reconciliation, missing information and retained human choice. The responder did not receive the scoring rubric.</p><p>Two working sessions on the same AI platform produced and assessed the outputs. The scorer also wrote the cases. This was not blinded expert evaluation or independent human review. There was one run, no replacement of unsuccessful responses and no live external action.</p><p>The author-scorer judged 23 criteria unambiguously met and credited one conditional answer: 78 unique payers, provided the stated duplicates represented 12 excess records within the total. Six cases passed only with that scoring judgment included. That is a result about the selected outputs on a small authored set. There was no human comparison group, no second model and no measurement of what anyone subsequently decided. The score should not be turned into a productivity claim.</p><h3>The value of refusing false precision</h3><p>The inventory case recorded 64 units, with 19 reserved and seven damaged. The missing fact was overlap: were any damaged units already among the reserved stock? The available quantity could be as low as 38 or as high as 45. A new order required 40. The response therefore did not guarantee supply. It identified the exact information needed to decide.</p><p>This is a useful form of support because it makes uncertainty actionable. A person can ask the warehouse about overlap rather than accept an invented exact quantity. The system has not chosen the business priority or made a customer promise. It has improved the preparation available for that choice.</p><p>The time case exposed a different error. A presentation normally required six hours. The proposed AI workflow required one hour for drafting, two for review and one for correction, plus five hours of initial setup. Two presentations therefore required 13 hours in the first week rather than 12. The proposed ten-hour saving became a one-hour increase. Without repeating setup, the planned weekly difference would be four hours. These are authored assumptions, not observed work times.</p><p>The other cases required similar restraint. Fifty voluntary ratings supported a weighted average of 4.4, not a claim about all customers. A discounted room charge of GEL 3,240 did not include an unknown breakfast total. A newer unsigned contract did not establish approval. The pattern was a clearer basis for human judgment, not evidence that judgment improved.</p><h3>The claim that still needs people</h3><p>High author-assigned scores also reveal the limits of this test. Six short cases cannot establish reliability across unfamiliar documents, longer workflows or live tools. A correct handoff can still be misunderstood, ignored or accepted too readily. Human augmentation must be evaluated at that final step.</p><p>A stronger study would compare people performing comparable work with and without assistance, measuring quality and total time, including review and correction. It should also test whether people notice incorrect advice. Until then, the honest conclusion is narrower: this text system prepared useful decision-support material in six authored cases. An agent strengthens human capability only when that preparation becomes better human work.</p><h3>Questions and answers</h3><h3>Can an agent disagree with an employee</h3><p>Yes. When a working note conflicts with the records, useful support exposes the difference and makes it checkable.</p><h3>What does the assessment establish</h3><p>Twenty-three criteria were unambiguously met, and the scorer also credited one conditional answer. This is neither a general reliability rate nor a percentage improvement in human performance.</p><h3>Was time saved measured</h3><p>No. The time example is scenario arithmetic. The trial did not measure people’s completion time.</p><h3>Who makes the final decision</h3><p>In this pilot’s design, the responsible person does. Real organizations must define decision and action authority in advance.</p><h3>Research basis</h3><p>Research question: Can a text-based AI system check a flawed working note and prepare a clearer basis for a human decision?</p><p>Method and data: On 6 November 2026, BTUAI ran six purpose-written Georgian scenarios once, covering payments, a hotel quotation, inventory, ratings, time planning and contract versions. These are synthetic cases, not real business records. Four criteria were set in advance for each case: evidence-consistent interpretation, numerical or record reconciliation, identification of missing information and retention of human choice. All six responses were retained unchanged, with no rerun or replacement.</p><p>Assessment: A separate session on the same AI platform generated the responses without receiving the rubric. The case-author AI session scored them; this was neither independent human review nor blinded expert assessment. Of 24 criteria, 23 were unambiguously met and one was conditionally credited: 78 unique payers is correct if the 90 records include 12 excess duplicate records. Five cases therefore passed unambiguously; the sixth passed with that conditional credit.</p><p>Main result and units: The responses exposed unsupported assumptions and questions needing clarification. Scenario arithmetic gives 38–45 available units; a weighted average of 4.4 points across 50 ratings; and GEL 3,240 for discounted hotel rooms, excluding an undetermined breakfast total. Planned time for two presentations rises from 12 to 13 hours in the first week; without repeating setup, the later difference is four hours. These are scenario calculations, not measured savings.</p><p>Limitations: Six simple cases and one run do not establish general reliability. There were no human participants, control group, comparison model or live-system actions. Human productivity, time, learning, decision quality and financial outcomes were not measured. The exact tested model version was not retained. The workplace study and Geostat indicators provide context, not validation of this trial.</p><p>Full BTUAI Article: <a href="https://btuai.ge/en/how-an-ai-agent-strengthens-human-capability/">https://btuai.ge/en/how-an-ai-agent-strengthens-human-capability/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=42e6fa4f7d54" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The real price of a larger pack]]></title>
            <link>https://medium.com/@mariam.kobiashvili/the-real-price-of-a-larger-pack-86ef692b5d0d?source=rss-45270e692e56------2</link>
            <guid isPermaLink="false">https://medium.com/p/86ef692b5d0d</guid>
            <category><![CDATA[pricing]]></category>
            <category><![CDATA[household-finances]]></category>
            <category><![CDATA[consumer-behavior]]></category>
            <category><![CDATA[retail]]></category>
            <category><![CDATA[georgia]]></category>
            <dc:creator><![CDATA[Business and Technology University  (BTU)]]></dc:creator>
            <pubDate>Wed, 18 Nov 2026 10:50:40 GMT</pubDate>
            <atom:updated>2026-11-18T10:50:40.969Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*naCQdDKuOuG0h7j0vuoCzg.png" /></figure><p>A Georgian retail comparison shows why a bargain depends on both the offer and the quantity a household can use</p><h3>Main summary</h3><p>BTU’s calculations connect displayed grocery prices with the amount of product a household expects to use. Three matched comparisons in Georgia show that a larger format can offer a substantial unit-price advantage or a very small one, making the specific offer more informative than package size alone.</p><h3>Main conclusions</h3><ul><li>In the matsoni offers examined on 11 November 2026, the 400-gram format’s unit-price advantage over the 110-gram format was 23.2% at Agrohub and 54.4% at Goodwill Galleria Tbilisi on Wolt.</li><li>The same Goodwill channel showed only a 0.8% per-egg advantage for a 15-pack of Koda category-0 eggs over a 10-pack. An identical quantity of 30 eggs cost just GEL 0.15 less in larger packs.</li><li>BTU’s contribution is to connect these actual offer comparisons with conditional consumption thresholds. Prices are observed listings; thresholds are calculations, while household use remains the condition to establish.</li></ul><h3>An economy that can be half a tetri</h3><p>Consider two packs of the same brand and stated category of eggs. One holds 10 and costs GEL 6.45. The other holds 15 and costs GEL 9.60. The larger pack is cheaper per egg, but only by half a tetri: 64 tetri instead of 64.5. A tetri is one hundredth of a Georgian lari.</p><p>These were the prices displayed on Goodwill Galleria Tbilisi’s public Wolt page in the comparison read on 11 November 2026. Standardizing the purchase to 30 eggs gives GEL 19.35 in smaller packs and GEL 19.20 in larger ones. The saving exists, but its size changes how compelling the offer looks.</p><p>That is the starting point of BTU’s analysis: a package can be larger without offering a substantial economy. The useful question is how much an appropriate quantity costs, followed by how much of it will actually be used.</p><h3>The promotion changes the arithmetic</h3><p>Matsoni, a fermented milk product familiar in Georgia, provides a contrasting case. The comparison examined Santino 3.2% matsoni in 110-gram and 400-gram formats within each of two retail channels. At Agrohub’s own website, the prices were GEL 1.20 and GEL 3.35. At Goodwill Galleria Tbilisi on Wolt, they were GEL 1.20 and GEL 1.99.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/729/1*lLl7v88AVMME-HYYRnV0FA.png" /></figure><p>Sources: primary Agrohub pages and Goodwill Galleria Tbilisi’s Wolt catalogue, read 11 November 2026. Pack prices are retailer figures; unit prices are BTU calculations. Delivery and service charges are outside this product-price comparison. Agrohub’s small matsoni and both Goodwill matsoni sizes displayed discounts.</p><p>The two matsoni comparisons yield different unit-price advantages, even though pack sizes do not change. The result belongs to each current offer. It does not isolate a causal effect of package size from promotion, and it should not be turned into a general ranking of the retailers.</p><p>For Georgian households, the question matters beyond occasional bargain hunting. GeoStat’s 2025 figures put food, beverages and tobacco at 38.7% of monetary consumption expenditure. That annual statistic, published 25 May 2026, establishes the relevance of recurring grocery decisions; the November price observations supply the new comparison.</p><h3>The quantity that reaches the table</h3><p>For a perishable product, purchased weight and useful quantity may differ. To make this relationship explicit, BTU divides the price of a large pack by the small format’s price per gram. The answer is the amount that must be used from the larger pack to match the cost-per-used-gram benchmark of a fully consumed small format.</p><p>At Agrohub’s prices, this break-even point is approximately 307.1 grams out of 400. At Goodwill’s prices, it is approximately 182.4 grams. Using more than the relevant threshold makes the larger pack cheaper per consumed gram, under the stated assumptions. These are price-based conditions, not estimates of Georgian food waste or instructions about shelf life.</p><p>Actual purchases still come in whole packs. If the requirement is 110 grams, the small matsoni costs less in cash. If the requirement is 220 grams, two small packs at Goodwill cost GEL 2.40, whereas one large pack costs GEL 1.99. The larger format then requires less money to obtain what is needed, even with some contents remaining. A useful comparison therefore considers both the standardized unit and the practical shopping quantity.</p><p>The eggs show the other edge of the problem. If one egg from the 15-pack remains unused, its cost per used egg becomes about 68.6 tetri, above the fully used 10-pack’s 64.5-tetri benchmark. A tiny unit-price advantage gives little room for unused product. This arithmetic explains a condition; it does not claim that a customer actually wastes that egg.</p><h3>Why the smallest pack can be sensible</h3><p>BTU’s everyday-consumption analytical model treats packaging as a choice involving household need, budget and convenience. It helps organize the questions behind the calculation rather than supplying measured packaging preferences.</p><p>Who will use the product? Will it serve several meals or one occasion? Can the household store it safely under its labelled conditions? A family’s size alone cannot settle those questions, because not every member necessarily consumes the same product.</p><p>Available cash also matters. In the Agrohub comparison, the larger matsoni requires GEL 2.15 more upfront. Someone balancing several essential purchases may choose the smaller format while fully understanding its higher unit price. A lower bill today and a lower price per kilogram are different advantages.</p><p>For businesses, this suggests a constructive role for a range of pack sizes. A smaller format can serve a specific occasion; a larger one can provide meaningful economy. The relevant commercial question is which need each format addresses, rather than whether every customer can be moved to the largest size.</p><h3>Make the comparison usable</h3><p>Georgia’s consumer-rights law provides for clear selling and unit prices, with statutory exceptions. Common units help customers compare, but the products must also match in relevant characteristics. Fat content, ingredients and egg categories can change what is being bought.</p><p>The findings support a simple discipline: examine the actual offer, compare an appropriate unit and consider usable quantity. Retail sales alone reveal purchases, not how much of a product a family consumed. Understanding that final step would require direct information about use.</p><p>Larger packs can create real value. The value is strongest when the price advantage is meaningful and the quantity fits household life. The practical improvement is to make those conditions visible, rather than asking the word “large” to stand in for an economic argument.</p><h3>Questions and answers</h3><h3>Is the cheapest kilogram always the best purchase?</h3><p>The best fit also depends on the amount needed, whole packs available, cash required and safe use. A lower unit price answers only one part of the choice.</p><h3>Do the results show a trend in Georgian prices?</h3><p>They describe three matched offer comparisons read on one date. A trend requires comparable observations over time.</p><h3>What is original in BTU’s analysis?</h3><p>The normalization of actual offers and the connection between their unit-price advantage and conditional consumption thresholds, supported by a household-decision framework.</p><p>Full BTUAI Article: <a href="https://btuai.ge/en/when-a-larger-pack-saves-money/">https://btuai.ge/en/when-a-larger-pack-saves-money/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=86ef692b5d0d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Marketing to the Customer Is No Longer Enough]]></title>
            <link>https://medium.com/@mariam.kobiashvili/marketing-to-the-customer-is-no-longer-enough-a197cf58bbef?source=rss-45270e692e56------2</link>
            <guid isPermaLink="false">https://medium.com/p/a197cf58bbef</guid>
            <category><![CDATA[marketing]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[consumer-behavior]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[ecommerce]]></category>
            <dc:creator><![CDATA[Business and Technology University  (BTU)]]></dc:creator>
            <pubDate>Tue, 17 Nov 2026 10:31:51 GMT</pubDate>
            <atom:updated>2026-11-17T10:31:51.965Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vwctgy9lj3d6SOwAo8HSnA.png" /></figure><p><em>As AI shopping agents move from recommendations to transactions, brands need to become persuasive to humans and legible to machines.</em></p><h3>The Global Question</h3><p>For decades, digital commerce optimized a journey that still assumed one thing: a person was clicking the links. Agentic commerce changes that assumption.</p><p>A shopper can increasingly tell an AI system what outcome they want and delegate part of the work: research, comparison, filtering and eventually checkout.</p><h3>A Second Buyer Enters the Market</h3><p>A recent BCG article coauthored with leaders from Yale’s Center for Customer Insights frames the change clearly. Brands will increasingly compete not only for customer attention but also for the trust of the AI systems shaping customer decisions.</p><p>The human buyer still responds to identity, story and experience. The machine buyer is better at comparing observable facts at scale.</p><h3>Why Operations Become Marketing</h3><p>An agent can compare reviews, prices, fulfillment, availability and service evidence across multiple sources. That makes the gap between brand promise and brand performance easier to expose.</p><p>The future of brand building is therefore not less emotional. It is more accountable.</p><h3>The Infrastructure Is Arriving</h3><p>Google’s Universal Commerce Protocol is designed to let platforms, agents and merchants share a common language for discovery and checkout. Merchant structured-data systems already support explicit price, availability, shipping and return information.</p><p>This is a sign that agentic commerce is moving from concept to infrastructure.</p><h3>Do Not Declare SEO Dead</h3><p>The rise of GEO-Generative Engine Optimization-has created a new consulting vocabulary. Google’s own guidance is more conservative: good optimization for generative AI search remains grounded in good search practices.</p><p>The real shift is not a new acronym. It is the need for clear, structured and current information.</p><h3>The Numbers Need Careful Reading</h3><p>McKinsey estimates that agents could mediate USD 3–5 trillion in global consumer commerce by 2030. BCG’s survey of nearly 300 CMOs finds 96% see end-to-end AI transformation, but only 31% have made agentic execution real.</p><p>Forrester’s official B2B prediction is also more measured than some viral summaries: it forecasts that 20% of B2B sellers will face AI-agent quote negotiations in 2026, not that 90% of all B2B purchases will become agent-to-agent.</p><h3>Why Georgia Is an Interesting Small-Market Case</h3><p>Georgia is already a meaningful online-shopping market. In 2025, 37.4% of internet users aged 15+ had bought or ordered goods or services online; among 15–29-year-olds the figure was 61.1%.</p><p>For a small economy, the coming shift could be especially important for exporters. A Georgian wine producer, hotel, designer or B2B supplier may be evaluated by an AI agent before a human ever sees the company’s homepage.</p><h3>What Businesses Should Notice</h3><p>The practical task is not to market only to machines. It is to build a dual interface: storytelling for people and evidence for machines.</p><p>That means consistent product identity, structured product facts, independent trust signals, transparent policies and current inventory and pricing.</p><h3>What Georgia Can Teach Other Small Economies</h3><p>Smaller firms cannot always win the attention economy through advertising scale. But agent-mediated commerce can increase the value of something smaller companies can improve relatively quickly: information quality.</p><p>A company with clean data and reliable operations can become easier for an agent to recommend than a louder but less transparent competitor.</p><h3>Conclusion</h3><p>The next front door to commerce may not be a search result or a social feed. It may be an agent acting for a customer.</p><p>Brands will still need to be memorable. They will also need to be verifiable.</p><p>Full BTUAI Article: <a href="https://btuai.ge/en/when-the-buyer-is-an-ai-agent-how-georgian-business-needs-to-change/">https://btuai.ge/en/when-the-buyer-is-an-ai-agent-how-georgian-business-needs-to-change/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a197cf58bbef" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Georgia needs more than another stock market listing]]></title>
            <link>https://medium.com/@mariam.kobiashvili/georgia-needs-more-than-another-stock-market-listing-58475c3dcb24?source=rss-45270e692e56------2</link>
            <guid isPermaLink="false">https://medium.com/p/58475c3dcb24</guid>
            <category><![CDATA[georgia]]></category>
            <category><![CDATA[capital-markets]]></category>
            <category><![CDATA[business]]></category>
            <category><![CDATA[stock-market]]></category>
            <category><![CDATA[investment]]></category>
            <dc:creator><![CDATA[Business and Technology University  (BTU)]]></dc:creator>
            <pubDate>Tue, 17 Nov 2026 10:29:25 GMT</pubDate>
            <atom:updated>2026-11-17T11:30:13.489Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*X-bL7OH-72JKUZv1ydouGw.png" /></figure><h3>Main summary</h3><p>For Georgian businesses, the opportunity in prospective international listings is access to investors willing to share the risks of growth. Building a liquid domestic equity market is a separate challenge, even when a company succeeds abroad.</p><h3>Main conclusions</h3><p>· International offerings could expand Georgian firms’ access to equity when new shares bring money into the businesses. Sales by existing shareholders have a different immediate destination for the proceeds.</p><p>· National Bank data show that bonds supplied 11.4% of Georgian non-financial companies’ debt financing at end-2025. This describes the mix of debt, not the share of equity in total financing.</p><p>· Georgia’s domestic equity trading-activity ratio was 0.7% in 2025 under the National Bank’s definition. A listing overseas does not automatically produce more trading in shares available in Georgia.</p><p>An announcement about a prospective stock market listing is often read as an announcement about investment. For a growing business, the two can be connected. They are not interchangeable.</p><p>On 11 November, a Galt &amp; Taggart representative told Georgian business outlet BMG that some Georgian companies were interested in issuing shares internationally. No prospective issuers were named. The suggested nine-to-twelve-month horizon was an expectation, not a timetable for confirmed transactions.</p><p>The discussion nevertheless raises an important question for a small economy: how can success in global capital markets become a wider improvement in business finance?</p><p>BTU’s analytical assessment is that two objectives need to be distinguished. Companies need capital that fits the risks and duration of their growth plans. Domestic investors need a functioning market in which shares can actually be bought and sold. One international offering may help with the first without resolving the second.</p><h3>The kind of money matters</h3><p>A manufacturer developing an export product and a young technology business can face the same financial tension. Investment must happen before the full revenue opportunity becomes clear. Money is needed today, while the timing of future cash flows remains uncertain.</p><p>Debt imposes repayment obligations. Ordinary equity gives investors an ownership claim and exposure to the business’s results. Neither is universally superior. They allocate risk differently, and a company’s financing needs may change as it develops.</p><p>Georgia’s National Bank offers useful structural context in its Capital Market Overview, published on 15 September and principally covering end-Q2 2026. Its figures are not new November data. They describe the setting into which prospective offerings would arrive.</p><p>At end-2025, all bonds accounted for 11.4% of Georgian non-financial companies’ debt financing, including public and private offerings in local and foreign markets. The complementary share, calculated by BTU, is 88.6% for credit excluding bonds. These percentages exclude shareholders’ equity from their denominator. They cannot tell us that the same share of all corporate activity is financed by loans.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/624/1*ZffKpBrp3kXvyU2VyIJ27A.png" /></figure><p>Source: NBG Capital Market Overview Q2 2026; 88.6% is BTU’s complementary calculation. Indicators have different denominators and must not be added.</p><p>The significance is not that every borrower should become a listed company. An ordinary equity investor can participate in uncertain growth without requiring a loan-like fixed repayment schedule. Existing owners, however, give up part of the ownership and accept responsibilities towards new shareholders. Equity has a cost even when it has no scheduled principal repayment.</p><p>The policy opportunity is a wider range of appropriate choices. A company with stable cash flows may be well served by borrowing. Another business may need a private investor. Public equity can become relevant when scale, governance and financing requirements justify its obligations.</p><h3>Follow the proceeds</h3><p>For an offering to finance business expansion directly, it matters who receives the money. Newly issued shares can bring proceeds into the company. A sale of shares already owned by a founder transfers money to that founder. An offering can contain both.</p><p>The London Stock Exchange’s current admissions guidance explicitly includes admission without new capital raising. This is useful when evaluating announcements, but it should not become an excuse to dismiss transactions that enable an ownership change. Those transactions may improve governance or diversify ownership. Their full value simply should not be counted as new money for factories or jobs.</p><p>The same discipline applies to share prices. A higher valuation changes the price at which existing ownership claims trade. It does not deposit cash into the company’s account. A stronger valuation may help a future capital raising, but the financing arrives through that later transaction.</p><p>For Georgian businesses, a useful assessment therefore starts with the issuer’s proceeds and their purpose. Paying down debt may strengthen a company’s resilience. Funding expansion may open another opportunity. Each objective needs its own evidence of execution rather than a universal promise of higher employment or exports.</p><h3>Two markets can develop at different speeds</h3><p>At end-2025, shares of two companies admitted to the Georgian Stock Exchange had a combined market capitalisation of GEL 631 million, according to the National Bank. The bank reported a domestic equity trading-activity ratio of 0.7% for 2025: annual transactions divided by average market capitalisation for 2024 and 2025.</p><p>These figures concern the domestic market, not all Georgian-linked companies traded abroad. Capitalisation is a valuation measure. The trading ratio concerns activity. Neither is a measure of newly raised funds.</p><p>Thin trading can make an otherwise attractive investment difficult to sell at an acceptable price. More listed names alone do not create buyers. An international offering can give a Georgian company access to a larger investor pool, while the domestic market remains shallow.</p><p>Local investment funds are not automatically an equity-demand solution either. At end-Q2 2026, corporate-debt and fixed-income funds dominated the GEL 546 million managed by Georgia’s 24 funds. The composition shows why growth in managed assets and growth in demand for company shares need separate analysis.</p><h3>The business must be ready</h3><p>Investor access depends on the company’s ability to explain its revenues, obligations, significant related-party dealings and development risks. That is an organisational capability, not merely a convincing presentation.</p><p>Current FCA procedures require relevant investor documents to meet disclosure and approval requirements. The exact requirements depend on the market and offering. Regulatory approval does not guarantee an investor’s return.</p><p>For a growing Georgian firm, preparation can mean making financial information less dependent on the founder, clarifying responsibilities and improving ongoing disclosure. These changes require resources. They help explain why a public listing is not the right immediate goal for every small business.</p><p>BTU’s conclusion is that international listings should be judged by the financing choices they open and the capital actually received by companies. Domestic equity-market development should be judged separately by investable businesses, informed investors and sustained trading. Success abroad can support that wider process, but it cannot substitute for it.</p><h3>Questions and answers</h3><p>Does a listing always raise new capital?</p><p>No. New share issuance can finance the company; sales by existing holders deliver proceeds to those holders. Some admissions involve no capital raising.</p><p>Is Georgia’s 0.7% trading ratio an investment return?</p><p>No. It compares annual transactions with the National Bank’s specified average market-capitalisation denominator for 2024 and 2025.</p><p>Are new Georgian international listings confirmed?</p><p>The interview discussed interest and prospective timing. It did not identify completed transactions or named new issuers.</p><p>What should readers watch next?</p><p>Identifiable transaction documents should show the issuer, new shares, proceeds received by the company and intended use. Subsequent reporting can establish what was delivered.</p><p>Full BTUAI Article: <a href="https://btuai.ge/en/georgian-businesses-need-an-alternative-to-debt/">https://btuai.ge/en/georgian-businesses-need-an-alternative-to-debt/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=58475c3dcb24" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Georgia’s electricity problem is not just rising imports]]></title>
            <link>https://medium.com/@mariam.kobiashvili/georgias-electricity-problem-is-not-just-rising-imports-8a75c059727f?source=rss-45270e692e56------2</link>
            <guid isPermaLink="false">https://medium.com/p/8a75c059727f</guid>
            <category><![CDATA[electricity-distribution]]></category>
            <category><![CDATA[electricity]]></category>
            <category><![CDATA[balance]]></category>
            <category><![CDATA[georgia]]></category>
            <dc:creator><![CDATA[Business and Technology University  (BTU)]]></dc:creator>
            <pubDate>Mon, 16 Nov 2026 11:25:07 GMT</pubDate>
            <atom:updated>2026-11-16T11:25:07.003Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*t8TdH9FsgSixMKeJ462_Jg.png" /></figure><h3>Main summary</h3><p>Georgia’s electricity imports rose moderately in the first nine months of 2026, but its net import requirement rose much faster. BTU’s analysis shows that the central change was a loss of export volumes, which becomes clearer when trade, domestic demand and the generation mix are examined together.</p><h3>Main conclusions</h3><p>· BTU calculates that Georgia’s net electricity imports increased by 314.6 million kWh between January–September 2025 and the corresponding 2026 period. Lower exports account for about 83% of that change.</p><p>· National consumption increased by 240.3 million kWh and domestic production fell by 74.3 million kWh. These independently aggregated changes reconcile with the deterioration in the external balance.</p><p>· The second quarter accounts for the largest deterioration. Its shrinking surplus matters for evaluating new investment, because nameplate capacity does not tell us when useful energy will reach the system.</p><p>Energy investment announcements usually arrive with a prominent capacity figure. A new plant will provide a certain number of megawatts; an associated substation will connect it to the grid. These are important details. But they do not, by themselves, reveal which economic problem the project will solve.</p><p>A 13 November BM.ge interview about an Imereti wind project provides a timely example. The investor’s representative described plans for the plant and its substation, with full commercial operation planned for early 2028. That remains a company timetable. To understand the system into which new production would enter, BTU examined Georgia’s actual electricity balance rather than treating future capacity as energy already available. [3]</p><h3>The difference between buying more and selling less</h3><p>ESCO, Georgia’s Electricity System Commercial Operator, publishes monthly physical balances. Comparing January–September in 2025 and 2026 produces two apparently different headlines. Gross electricity imports increased by about 5%. Net imports increased by about 51%. [1–2]</p><p>The explanation is straightforward. Net imports subtract exports from imports. Georgia bought 53.8 million kWh more electricity in the 2026 period, but exported 260.8 million kWh less. Together, those movements increased its net import requirement by 314.6 million kWh.</p><p>BTU’s decomposition attributes approximately 83% of the net deterioration to reduced export volumes. This describes the accounting contribution of each trade flow, rather than identifying the motives of traders or the ultimate causes of changing production. The distinction matters: a system can experience a substantially weaker external balance even when imports rise only modestly.</p><p>The figures describe energy, not money. The commercial value of purchases and sales depends on their timing and prices. A decline in exported kilowatt-hours cannot be converted into a reliable revenue loss using an assumed average price.</p><h3>A second route to the same result</h3><p>The finding can be tested without starting from trade. National consumption increased by 240.3 million kWh, while domestic generation fell by 74.3 million kWh. Adding the extra requirement to the lost production gives the same 314.6 million kWh change in the balance, apart from negligible source rounding.</p><p>This second route is useful because it connects the trade result to what happened inside the electricity system. Demand grew by approximately 2.2%, whereas generation declined by approximately 0.7%. A fairly small change in a large domestic total can translate into a substantial change in the smaller net trade position.</p><p>The generation total also hides a significant rearrangement. Hydropower production fell by 257.0 million kWh. Thermal generation rose by 145.1 million kWh, while the calculated residual for other generation increased by 37.6 million kWh. Much of the hydro decline was therefore offset before it reached the total production figure.</p><p>That is an important economic distinction. Similar aggregate output can come from a different production portfolio. The operating requirements of that portfolio may change even when its total appears stable. The balance establishes the output movements; determining their cost or separating weather, maintenance and dispatch effects requires further evidence.</p><h3>Where the surplus disappeared</h3><p>Dividing the comparison into quarters reveals that the deterioration was uneven. First-quarter net imports increased by 103.8 million kWh. The third-quarter position improved by 184.2 million kWh, moving from net imports in 2025 to net exports in 2026.</p><p>The second quarter tells the strongest part of the story. Its net export surplus fell from 420.7 million kWh to 25.7 million kWh. This 395.1 million kWh reduction outweighed the third-quarter improvement and explains much of the nine-month result.</p><p>Inside that quarter, hydro generation fell by 559.3 million kWh. Thermal output increased by 193.2 million kWh and other generation provided an additional partial offset. Total production nevertheless declined by 354.0 million kWh, while consumption increased by 41.0 million kWh. Together, these changes reconcile with the smaller surplus.</p><p>This is more informative than saying Georgia simply became a larger electricity buyer. It identifies a period in which a previously substantial export cushion nearly disappeared. The comparison is between matched quarters, not a forecast that every subsequent year will follow the same pattern.</p><h3>What investors should take from this</h3><p>The practical question for new generation is how its production fits the system’s requirements. Annual energy output matters, but so does its distribution across months and hours. The proposed Imereti development also illustrates why the plant and its network connection need to be considered together. Generating power and delivering it into the system are linked but separate capabilities. [3]</p><p>For international readers, the units are worth distinguishing. Megawatts measure capacity. Kilowatt-hours measure energy produced or consumed over time. An announced capacity cannot be directly compared with the energy gap in these tables without an operating profile.</p><p>BTU’s assessment is that investment appraisal should start with the observed system requirement, then examine the project-specific contribution. That includes expected generation, connection readiness and interaction with existing sources. A forecast for one plant cannot be borrowed mechanically from the performance of another.</p><p>The relevance extends to businesses outside energy. Manufacturers, retailers, hotels and digital-service providers need electricity when they operate, rather than an abstract share of annual national output. Expansion in these sectors must fit with generation and network development. The national balance offers a framework for understanding that connection, not a prediction of any particular company’s future bill.</p><p>The principal lesson is therefore not that Georgia’s imports suddenly surged. It is that additional consumption and lower domestic generation were reflected mainly in reduced exports. Recognizing that mechanism gives new investment a more demanding test: whether additional capacity becomes useful, deliverable energy when the system needs it.</p><h3>Questions and answers</h3><p><strong>Does a higher net import requirement establish blackouts?</strong> No. It measures the gap between national consumption and domestic generation over the period. Cross-border flows can supply that gap without unmet demand.</p><p><strong>Do reduced exports establish a financial loss?</strong> They establish a lower physical sales volume. A monetary assessment also needs actual transaction prices and contractual conditions.</p><p><strong>Why not estimate household savings from the proposed wind project?</strong> Household effects require project output, purchase prices, network costs and tariff arrangements. The announcement does not establish those inputs.</p><p><strong>What should be updated next?</strong> The comparison should incorporate subsequent actual monthly balances, maintaining matched periods and the same definitions. The fourth quarter is not forecast here.</p><p>Sources: [1] ESCO, <a href="https://esco.ge/energy-balance/by-year-1/balnsi-2026;">https://esco.ge/energy-balance/by-year-1/balnsi-2026;</a> [2] ESCO, <a href="https://esco.ge/energy-balance/by-year-1/balance-2025;">https://esco.ge/energy-balance/by-year-1/balance-2025;</a> [3] BM.ge, <a href="https://bm.ge/news/chinuri-sichuan-energy-development-i-imeretis-qaris-eleqtrosadguris-proeqtshi-140-milionamde-investitsias-chadebs.">https://bm.ge/news/chinuri-sichuan-energy-development-i-imeretis-qaris-eleqtrosadguris-proeqtshi-140-milionamde-investitsias-chadebs.</a></p><p>Original article: <a href="https://btuai.ge/en/the-hidden-change-in-georgias-electricity-balance/">https://btuai.ge/en/the-hidden-change-in-georgias-electricity-balance/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8a75c059727f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[An AI agent should stop making you read the same work twice]]></title>
            <link>https://medium.com/@mariam.kobiashvili/an-ai-agent-should-stop-making-you-read-the-same-work-twice-06656a8cb0fa?source=rss-45270e692e56------2</link>
            <guid isPermaLink="false">https://medium.com/p/06656a8cb0fa</guid>
            <category><![CDATA[business]]></category>
            <category><![CDATA[task-division]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[work-routine]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Business and Technology University  (BTU)]]></dc:creator>
            <pubDate>Mon, 16 Nov 2026 11:22:03 GMT</pubDate>
            <atom:updated>2026-11-16T11:22:03.408Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x0ZzIawdowXfCJR8zd6CMw.png" /></figure><h3>Main</h3><p>The useful promise of an AI agent is to preserve the state of work and show what has actually changed. A small Georgian-language demonstration made that mechanism visible, but a shorter queue is not evidence of saved human hours.</p><h3>Key Takeaways</h3><p>· A useful agent distinguishes a newly received message from genuinely new information and retains the underlying case history.</p><p>· In BTUAI’s single synthetic run on 6 November 2026, twelve incoming records produced five new or changed cases and seven suppressed repeats, with fourteen distinct cases retained overall.</p><p>· The demonstration measured information processing, not employee productivity; human review costs and real service outcomes remain unmeasured.</p><h3>Part one The work between decisions</h3><p>Imagine starting the day with twelve messages about orders already in progress. Most tell you something you knew yesterday. Some contain a genuine amendment. One asks for an action that has not happened. Another contradicts an existing record. Before doing anything useful, you must reconstruct the state of the work.</p><p>That reconstruction is a familiar kind of routine. It does not disappear when an AI system turns twelve messages into twelve polished summaries. The writing has changed, but the reader must still identify what matters. A more useful assistant keeps the previous state, compares new evidence against it and preserves the difference between an agreed change and an unresolved question.</p><p>An agent, in this discussion, is a system that can select steps, use permitted tools and continue toward a goal based on their results. That does not mean every operation should depend on a language model. Exact duplication checks and arithmetic often belong in ordinary software. Interpretation matters when the same fact is expressed differently or when “please cancel” must be distinguished from “cancelled.” Anthropic’s agent-design guidance similarly favors the simplest adequate system rather than complexity for its own sake.</p><p>The organizational question is therefore specific: can preparation be delegated without transferring decisions the system has no authority or evidence to make? An unresolved delivery conflict is not necessarily failed work. Correctly exposing it may be the useful result. The person receives a narrow question, connected to its sources, rather than having to discover the contradiction in a long conversation.</p><h3>Why Georgia offers a useful perspective</h3><p>Georgia’s official enterprise ICT survey reported internet access at 94.9% of enterprises in 2025, while 15.3% used a website. The release appeared in May 2026. Neither measure establishes workflow integration. Their relevance is to distinguish digital access from the ability to maintain a coherent, usable work record.</p><p>For businesses handling Georgian-language messages, the proposed benefit is concrete: connect a request to its case, keep an agreed amendment and surface an unresolved condition. Trade, tourism, education and professional services offer possible settings, not measured sectoral results. A small team may value continuity across several responsibilities, but additional correction work could erase any benefit.</p><p>Real workplace evidence is a different category. Brynjolfsson, Li and Raymond’s 2025 study of 5,172 customer-support workers at one firm found an average 15% increase in issues resolved per hour with AI assistance. Effects differed across workers. That result concerns a human-assistance system and cannot establish either autonomous-agent performance or Georgian productivity gains. It illustrates the need to measure completed work, not merely the quantity of text generated.</p><h3>Part two A bounded Georgian language demonstration</h3><h3>Separating changes from repeated intake</h3><p>On 6 November 2026, BTUAI used a deliberately small synthetic stream to examine one part of routine reduction. Twelve initial records described twelve order cases. A second batch contained twelve incoming messages: six unchanged updates, three changed cases, two new cases and one retransmission. All records were author-created. No company messages or human participants were involved.</p><p>An AI worker actually processed the Georgian texts, preserving source identifiers and preparing a structured final state, a change queue and an exception queue. Python supported record processing and verification. The exercise ran once; a separate AI reviewer checked its outcome against the input. This was not human expert review. It was not a controlled comparison between models, ordinary software and people.</p><p><strong>What the shorter list still had to contain</strong></p><p>In one case, the original delivery date for seventy units was 16 November. The next message said that the supplier now named 18 November while the procurement record still said 16 November. No rule made either source authoritative. Retaining both dates and flagging the conflict was more useful than producing a falsely certain update.</p><p>Another message requested cancellation of an order. The result kept that as a request, not as an accomplished cancellation. A separate agreed amendment changed a quantity from eighty to eighty-five. Together, the cases required the system to distinguish a confirmed correction, conflicting evidence and an action requiring authorization.</p><p>Just as importantly, silence did not erase work. Cases with no subsequent message remained in the final state. The original twelve cases plus two new ones produced fourteen retained cases. This is why the complete record and the short review queue must coexist. A short briefing should remove repeated reading, not outstanding obligations.</p><h3>A result with a deliberately narrow meaning</h3><p>The run identified the changes embedded in this small set. It did not establish general accuracy. The records were clean, identifiers were stable and the important facts were available. Real workflows can involve scans, missing attachments, ambiguous names, contradictory policies and failed tools. No ordinary-software comparator was tested, so superiority over a simpler solution remains unestablished.</p><p>Human review was not timed. No employee’s working day, wage cost or service outcome was observed. A shorter queue may help, but its hardest entries may still consume substantial attention. Verifying the complete record also takes effort. A real evaluation would need to count those costs rather than treating them as free.</p><h3>Conclusion</h3><p>The credible promise is a change in where attention goes. Let the system reconstruct state, keep evidence and suppress genuine repetition; let the person see the unresolved choices. The demonstration showed that arrangement in one bounded text task. Whether it produces more time for customers, learning or better decisions remains a question for actual work, not an extrapolation from a smaller list.</p><h3>Questions and answers</h3><h3>Where is a sensible starting point</h3><p>Compare current messages with an existing case record, where sources and changes are verifiable.</p><h3>Were seven cases removed</h3><p>No. Seven repeated records were excluded from the change queue; all distinct cases were retained.</p><h3>Why keep human decisions in the process</h3><p>Conflicting dates and a cancellation request cannot be converted into final commitments without sufficient evidence and authority.</p><h3>Did the test prove saved time</h3><p>No. It demonstrated processing of a synthetic stream, not a measured productivity gain.</p><h3>Research basis</h3><p>Research question: Can an AI worker separate new and changed cases from a recurring Georgian-language stream while preserving case history and unresolved issues?</p><p>Method: On 6 November 2026, a single small synthetic demonstration was executed for BTUAI. An AI worker compared constructed texts and produced the final case state, change queue and exception queue. Python supported processing and count checks; a separate AI reviewer compared output against the source records. This was not human expert review, a real-company field study or a blinded test.</p><p>Data and period: The author-created stream dated 6 November 2026 contained 24 records: 12 baseline and 12 subsequent records. The second intake comprised six unchanged updates, three changed cases, two new cases and one retransmission. Context comes from Geostat’s 2025 enterprise ICT indicators, released on 29 May 2026, the 2025 QJE workplace study and Anthropic’s 2024 design guidance. These contextual sources do not independently validate the pilot result.</p><p>Main result and units: BTU’s calculation places five cases in the change queue: three changed and two new. Seven repeated intake records were excluded from that queue, while 14 distinct cases remained in the final state (12 + 2). The two exceptions, conflicting delivery dates and a cancellation request, are a subset of the five queued cases, not additional cases. Queue-entry reduction is (12 − 5)/12 × 100 = 58.3%. The denominator is second-intake records, not distinct cases or working hours.</p><p>Limitations: One small constructed stream used stable identifiers and clean text. Human time, review effort, costs and service outcomes were not measured; there were no repeated runs or ordinary-software comparator. The 58.3% result is not time, cost or staffing savings and does not estimate general accuracy or Georgian business productivity. Geostat’s indicators do not measure AI adoption.</p><h3>Primary sources</h3><p>BTUAI author-created Georgian synthetic records, execution output and separate AI review, 6 November 2026: no public URL.</p><p>Geostat 2025: <a href="https://www.geostat.ge/media/79412/Use-of-Information-Communication-Technologies-in-Enterprises---2025.pdf">https://www.geostat.ge/media/79412/Use-of-Information-Communication-Technologies-in-Enterprises---2025.pdf</a></p><p>QJE 2025: <a href="https://academic.oup.com/qje/article/140/2/889/7990658">https://academic.oup.com/qje/article/140/2/889/7990658</a></p><p>Anthropic 2024: <a href="https://www.anthropic.com/research/building-effective-agents">https://www.anthropic.com/research/building-effective-agents</a></p><p>Full Report is available here: <a href="https://btuai.ge/en/how-an-ai-agent-frees-people-from-routine/">https://btuai.ge/en/how-an-ai-agent-frees-people-from-routine/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=06656a8cb0fa" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Handoff Is Where Human–AI Teams Succeed]]></title>
            <link>https://medium.com/@mariam.kobiashvili/the-handoff-is-where-human-ai-teams-succeed-6f896e2d8304?source=rss-45270e692e56------2</link>
            <guid isPermaLink="false">https://medium.com/p/6f896e2d8304</guid>
            <category><![CDATA[business]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[future-of-work]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Business and Technology University  (BTU)]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 11:52:31 GMT</pubDate>
            <atom:updated>2026-11-13T11:52:31.940Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0ZtCOzJhgDRQGvmBK0coOQ.png" /></figure><h3>Main</h3><p>BTU researchers assess that the useful boundary between people and AI depends on whether the next person can verify and act on the work received. Georgia offers a practical setting for this question: multilingual service tasks need a reliable connection between prepared information, local conditions and a decision that actually resolves the case.</p><h3>Key Takeaways</h3><ul><li>BTU’s analytical contribution distinguishes preparation, a focused checkpoint and exception resolution, rather than treating collaboration as a single transfer from AI to a person.</li><li>The same customer case can move between AI preparation and human judgment as facts or conditions change; a fixed division cannot cover every turn.</li><li>An organization gains from faster generation only when the receiving employee can finish the work without an expanding review queue.</li></ul><p>A team introduces AI, and the first visible change is encouraging. Drafts arrive sooner. Documents are compared more quickly. A colleague who once began every response from scratch now receives a plausible starting point. Yet the customer may still wait just as long.</p><p>The missing step is often the handoff. Someone has to determine what the draft establishes, what it assumes and what remains unresolved. When that person cannot see the original evidence, the shortcut ends with another round of research. The organization has accelerated preparation without necessarily accelerating completion.</p><p>This is the focus of BTU’s analysis of how people and AI should share work. The original contribution is a distinction between three moments: preparing evidence, checking a consequential issue and resolving an exception. Those moments need different information. Asking for a comprehensive report at each one can make collaboration slower and harder to understand.</p><p>Qualitative comparison: different handoff functions. Source: BTU analytical synthesis; not measured outcomes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/806/1*j-yTkBIE8Xn1bfemrE8xCA.png" /></figure><h3>Give the next person an answer they can use</h3><p>A good handoff is shaped by the receiving person’s next action. Someone comparing two documents needs to see what changed and where the relevant original passage sits. Someone handling a customer request needs to distinguish the customer’s preference from the organization’s existing promise. A polished summary that merges them can create confusion precisely because it reads so smoothly.</p><p>The test is not whether the output looks complete. It is whether the recipient can establish what is reliable and make the next decision. That may require less text, but better evidence: a source, a changed condition or a clearly stated unresolved question.</p><p>This is not a recommendation to make every employee approve everything. Review that focuses only on wording can miss the actual issue. In a quotation, the consequential elements may be price and conditions. In educational material, the essential check may concern whether the explanation is correct. Human attention should go to the elements that determine the outcome.</p><h3>A case can change hands more than once</h3><p>Consider a hotel in Georgia receiving a request to move a reservation. This is an illustration, not a reported deployment. An agent might assemble the booking conditions, compare dates and prepare a bilingual response. If the request fits the policy, the employee’s check may be straightforward.</p><p>Then the guest mentions an earlier promise that differs from the standard conditions. The task has changed. Repeating the general policy will not settle the matter. A person must reconcile the evidence and choose an appropriate response. After that decision, AI may again help prepare the message.</p><p>The useful boundary therefore moves within the same case. It is not permanently attached to a job title or a whole conversation. Nor does sending work back mean the collaboration failed. A limited correction can return to the agent while verified material remains intact.</p><p>The reason for that return matters. A copied date requires a different repair from a changed contractual premise. If the agent does not know what changed, it may repeatedly polish a conclusion that is no longer valid. The employee needs to identify the affected basis, not restart the entire assignment.</p><h3>More output can become a queue</h3><p>An employee’s attention is finite. When an agent prepares several cases at once, the receiving person may become the constraint. Completed drafts wait, customers wait and information can become outdated before anyone acts on it.</p><p>BTU’s assessment is that preparation should be coordinated with the capacity to finish cases. That does not supply a universal number of simultaneous assignments. It identifies a management question: is another generated answer useful now, or would resolving the current exception produce more value?</p><p>International evidence reinforces the need to consider the person using the tool. A 2025 Quarterly Journal of Economics study of 5,172 support workers found a 15% average increase in issues resolved per hour with AI assistance, with differing effects by experience. Employees could edit or ignore suggestions. The finding is workplace-specific, not a Georgian productivity forecast.</p><h3>Local conditions belong inside the handoff</h3><p>Georgia makes the language dimension concrete. An effective English demonstration does not establish accuracy in Georgian. Even a fluent translation can miss the meaning of a local service condition or the wording used in the source document. The person reviewing a Georgian answer needs the relevant Georgian evidence, not simply reassurance that the model performs well elsewhere.</p><p>BTU’s professional-agent production standard also distinguishes professional mandates, necessary knowledge and circumstances requiring human involvement. Its assessment applies to the configuration actually examined, rather than automatically covering later tools or permissions. Applied to collaboration, that principle means the division of work should be reconsidered when the process changes.</p><p>Technical guidance offers a compatible lesson. Anthropic’s 2024 discussion recommends simple arrangements and feedback before adding complexity. NIST’s voluntary AI risk framework connects governance, context, measurement and management. Neither supplies a measured Georgian return or a binding local rule; both help frame disciplined evaluation.</p><h3>Measure what the customer receives</h3><p>The meaningful result is a completed case, not a collection of drafts. Faster preparation can coexist with extra corrections, repeated customer contacts or more referrals to experienced staff. Those effects belong in the assessment alongside speed.</p><p>The strongest human–AI teams will not necessarily delegate the largest share of visible activity. They will make the next step easier for whoever receives the work. Preparation will bring evidence. Checkpoints will answer specific questions. Human judgment will resolve genuine exceptions, then return the case to a useful path.</p><p>That is a more demanding standard than counting generated responses. It is also closer to what customers and employees need: less reconstruction, fewer unresolved cases and work that actually reaches an end.</p><h3>Questions and Answers</h3><p><strong>What is the hidden cost of a handoff?</strong></p><p>Finding sources and rebuilding context can erase the time saved during preparation.</p><p><strong>Should every employee receive the same AI output?</strong></p><p>No. Experience and available time affect what evidence and workload are useful.</p><p><strong>Can the 15% research result be used as a Georgian target?</strong></p><p>Not as a predicted effect. A local process needs its own comparison of completed outcomes.</p><p>Full BTUAI.ge article: <a href="https://btuai.ge/en/how-people-and-ai-should-share-work/">https://btuai.ge/en/how-people-and-ai-should-share-work/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6f896e2d8304" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[When Global Shipping Changes, Georgia’s Test Is Reliability]]></title>
            <link>https://medium.com/@mariam.kobiashvili/when-global-shipping-changes-georgias-test-is-reliability-cc570f61356d?source=rss-45270e692e56------2</link>
            <guid isPermaLink="false">https://medium.com/p/cc570f61356d</guid>
            <category><![CDATA[logistics]]></category>
            <category><![CDATA[global-economy]]></category>
            <category><![CDATA[international-trade]]></category>
            <category><![CDATA[georgia]]></category>
            <category><![CDATA[middle-corridor]]></category>
            <dc:creator><![CDATA[Business and Technology University  (BTU)]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 11:43:40 GMT</pubDate>
            <atom:updated>2026-11-13T11:43:40.041Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AtHuuaE0A9sxL9wvIM9yAA.png" /></figure><h3>The difference between moving more cargo and earning more from it</h3><p>Consider two businesses that appear to be enjoying the same shipping boom. One earns more because the number of containers carried is rising. The other earns more because ships are delayed and available space has become scarce. Both can report stronger revenues, but they tell very different stories about the future. The first may reflect lasting demand; the second can disappear when the bottleneck clears.</p><p>That distinction is at the centre of the global container market in late 2026. On 5 November, shipping group Maersk reported that worldwide container demand had increased 2–4% year on year during the third quarter. The company also described growing industrial demand associated with energy investment, electrification and artificial intelligence infrastructure. Its assumption for the whole of 2026 is roughly 4% market growth. That is a forward-looking assumption, not an observed annual outcome.</p><p>Capacity constraints tell the other half of the story. Severe weather and congestion in Asian ports left vessels waiting longer and absorbed usable shipping capacity. The Shanghai Containerised Freight Index measure of average spot rates rose 46% from the preceding quarter, according to Maersk. It would be wrong to interpret this as a uniform increase in transport costs for every business. It is also wrong to treat high prices as straightforward evidence of an equally dramatic rise in cargo volumes.</p><p>For countries positioned between larger trading regions, this combination creates both an opening and a trap. Disrupted routes can make alternatives more attractive. But building an economic strategy around disrupted routes is risky if those disruptions are temporary.</p><h3>Georgia’s opportunity is more tangible than a map suggests</h3><p>Georgia, on the eastern shore of the Black Sea, is part of the network often called the Middle Corridor: a chain of rail, sea and land connections linking markets across the South Caucasus and the Caspian region. International discussions sometimes make it sound like a single new route between China and Europe. In reality, it is an operational system requiring coordination across ports, trains, shipping services and borders. It is also used for regional trade that does not span the whole Eurasian continent.</p><p>The country has measurable transport activity behind its ambitions. Georgia’s Ministry of Economy and Sustainable Development reported that seaports and terminals handled 9.7 million tonnes of cargo during the first six months of 2026, 22% more than in the corresponding period of 2025. Container handling reached 409,397 TEU, a 9% rise. TEU refers to a standard twenty-foot container equivalent rather than a count of individual physical boxes. Over the same six months, Georgian Railway moved 6.9 million tonnes of freight, up 10%.</p><p>Another indicator comes from the operator of Poti port. APM Terminals Poti handled 525,211 TEU in the first nine months of 2026, a 9% year-on-year increase. This figure has a different geographic scope and period from the national first-half figure, so the two cannot be subtracted to calculate cargo moved by other terminals or during a particular quarter.</p><p>These increases are significant, but they should not be mislabelled. Total seaport cargo is not identical to international transit, and rail tonnes are not directly comparable with container TEU. Nor can figures from the first nine months of the year demonstrate that a November market announcement caused growth in Georgia.</p><p>The distinction is consequential for Georgia’s economy. A container handled at a port may be bringing goods into Georgia, carrying Georgian exports, or passing through on an international transit route. Higher total container throughput therefore does not establish an equivalent increase in transit traffic, domestic logistics income, or the volume of long-distance China–Europe shipments.</p><p>Table 1. Selected indicators of Georgia’s transport sector:</p><p>Time is the second constraint. A container can leave a vessel quickly yet spend much longer waiting for the onward rail or road leg. For an international shipper, the competitive product is an end-to-end route across several operators and borders, not a terminal’s reported capacity in isolation.</p><p>This makes coordination a concrete investment question: can vessel arrivals, terminal handling, rail schedules and border procedures function as a predictable service? The ability to reduce uncertainty across those steps is a more durable source of value for Georgia than temporary increases in global ocean rates.</p><p>Independent analysis adds a further caution. PMCG’s September 2026 review put Georgia’s total transit flows in 2025 at 13.7 million tonnes and Middle Corridor-related flows at 4.6 million tonnes. Its assessment emphasised the importance of regional commerce alongside long-distance China–Europe shipments. For Georgia, the corridor’s present-day economic relevance is therefore broader and more practical than a single geopolitical slogan.</p><h3>What would make the advantage last?</h3><p>The business test of any transport route is the journey, not the map. A shipper wants to know how much it costs to move a defined cargo from a defined origin to a destination, how long it will take, and how reliably that promise will be met. Cheap service at one terminal can be offset by unpredictable waiting elsewhere. A faster railway segment may make little difference if containers lose time at a border or in a port.</p><p>BTU’s analytical assessment is that Georgia’s durable opportunity lies in converting geographic access into predictable service. That requires physical capacity but also information sharing between operators, transparent handling processes, better connections between sea and rail, and clear performance indicators. A customer choosing a route for several years needs a reliable system, not a fortunate week in the freight market.</p><p>Georgia’s importers and exporters have another stake in the outcome. More expensive or less predictable shipping can tie up working capital in inventories and raise landed costs. How much is passed to consumers depends on contracts, competitive conditions and business margins. Meanwhile, logistics providers may find higher-value services in warehousing, customs support, multimodal coordination and shipment tracking. More containers alone do not prove that these activities have become more profitable.</p><p>This is also a test for investment decisions. The higher freight rates now helping carriers might decline if congestion subsides or shipping capacity increases. Port and rail projects involve long horizons and should be evaluated against sustained cargo demand, operating costs and end-to-end route performance rather than the latest global price spike.</p><p>Georgia does not need to become the cheapest or fastest route for every shipment to be relevant. It needs to be a credible choice for specific trade flows whose customers value its overall combination of time, price and reliability. The evidence suggests an active transport system with room to develop. It does not yet establish how much of global growth Georgia will capture.</p><p>The broader lesson extends well beyond this small Black Sea economy. When global logistics is disrupted, the winners over time may not be those that briefly receive the most redirected traffic. They may be those that use the moment to build connections customers continue to trust when the world becomes less disrupted.</p><p>Full BTUAI.ge article: <a href="https://btuai.ge/en/global-shipping-enters-a-new-phase/">https://btuai.ge/en/global-shipping-enters-a-new-phase/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cc570f61356d" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>