<?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 Bogdan Yemets on Medium]]></title>
        <description><![CDATA[Stories by Bogdan Yemets on Medium]]></description>
        <link>https://medium.com/@bogdan.yemets?source=rss-8f0cde6ebd27------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*RcpC-TbMW3Nv97Y3tFwLpw.jpeg</url>
            <title>Stories by Bogdan Yemets on Medium</title>
            <link>https://medium.com/@bogdan.yemets?source=rss-8f0cde6ebd27------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 12:10:04 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@bogdan.yemets/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[Common mistakes in location-based app development]]></title>
            <link>https://medium.com/think-clockwise/common-mistakes-in-location-based-app-development-7738d98259ae?source=rss-8f0cde6ebd27------2</link>
            <guid isPermaLink="false">https://medium.com/p/7738d98259ae</guid>
            <category><![CDATA[location-based-services]]></category>
            <dc:creator><![CDATA[Bogdan Yemets]]></dc:creator>
            <pubDate>Thu, 13 Aug 2026 10:05:54 GMT</pubDate>
            <atom:updated>2026-08-13T10:05:54.801Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*JPDOy43QLPSJ0WA_7goKow.jpeg" /></figure><p>Location-based app development has a habit of making early shortcuts expensive. The map software development kit (SDK) demos beautifully with 50 markers on a laptop; the problems surface later — under real datasets, on real devices, at app store review, and on the first invoice past a provider’s free tier. By that point, fixes cost weeks instead of the hours they would have taken at the start of the project.</p><p>Part of the reason is that the constraints hide in the fine print. The Mapbox API, for example, renders a maximum of 100 location markers in real time — a limit you’d normally discover after committing to the provider, unless someone checked first. Every major map provider, mobile OS, and app store carries a handful of rules like this, and any one of them can reshape an architecture.</p><p>Every <a href="https://clockwise.software/digital-product-development/">digital product development company</a> that’s put location features in front of real users ends up with a list like this. Below are the 5 mistakes we run into more than any others in geolocation projects — what each one costs, when it surfaces, and what we do instead, with examples from products we’ve shipped.</p><h4>Choosing a map provider without considering your existing infrastructure</h4><p>The map provider decision usually gets made backwards: compare Google Maps, Mapbox, and Azure Maps on features and per-request pricing, pick a winner, and discover later that the real cost lives in how the provider fits your stack, your data, and your contracts.</p><p>In practice, we look at four things before features:</p><p><em>Existing cloud ecosystem</em></p><p>If your platform already runs on Azure (DevOps, Application Insights, identity), Azure Maps cuts integration and billing overhead. The same logic applies to Google Cloud and the Maps Platform.</p><p><em>Data ownership and export</em></p><p>Some providers restrict caching and storing geocoded results. If your product’s value is the location dataset itself, this clause matters more than price.</p><p><em>Specialized routing needs</em></p><p>Truck routing, vehicle-size restrictions, and hazard data are unevenly supported. Generic driving directions won’t route a double-decker bus.</p><p><em>Cost curve at your scale</em></p><p>Free tiers are generous; the pricing past them is not. Model your costs at 10x your launch traffic, not at launch traffic.</p><p>The route planning solution we built shows how this plays out. The client already had a location tracking platform and an Azure-based toolchain, so Azure Maps was the natural fit, and it also solved their hardest domain problem: routing double-decker buses around low bridges based on vehicle type and size. A provider without vehicle-constrained routing would have meant building that logic from scratch on top of a generic directions API.</p><p>The safer move is to treat the provider as an architectural decision. Write down your constraints first, then shop.</p><h4>Underestimating map performance with large datasets</h4><p>A map with 50 markers works fine everywhere. A map with 5,000 is a different piece of engineering, and the gap between the two is where many location-based apps quietly fall apart.</p><p>Two of our projects hit this wall in different ways.</p><p>On the fleet management platform we developed for a Denver waste collection company, thousands of container markers plus a long container list pushed app load time to about 5 seconds. Drivers won’t wait that long, and dispatchers definitely won’t. We got it down to 1.5 seconds — not with one fix, but with 5 smaller ones stacked:</p><ul><li>Marker clustering on the map</li><li>Optimized component lifecycles and a trimmed DOM structure</li><li>Smaller data request payloads</li><li>Pagination for the container list</li><li>Viewport-aware loading, so the app renders what the user actually sees</li></ul><p>On a location-based chat project, the constraint came from the provider itself: the Mapbox API renders a maximum of 100 location markers in real time. The product needed to show far more users than that, so we built a custom clustering algorithm — users see clusters by area and drill into individual locations by zooming.</p><p>A common risk here is assuming the map SDK will handle scale for you. It won’t, past a point, and each provider’s point is different. This is worth deciding early because clustering, viewport-based loading, and payload design affect your API contracts and your backend, not just the frontend.</p><h4>Not testing on real devices in real conditions</h4><p>Emulators are great for logic. They are useless for the conditions that break location-based apps: battery optimization, GPS drift, spotty connectivity, background process kills, and the specific ways each Android vendor “helps” by shutting down services.</p><p>The fleet management project gave us a clean example. Drivers used smartphones with a then-latest Android version that automatically turned off GPS to save battery. Orders would start fine, then location updates would silently stop mid-route. No emulator reproduces that. The fix was integrating a wake lock so the device couldn’t kill location services during active work — and we only found the problem because real drivers were running real routes on real hardware.</p><p>The same project surfaced a second field-only issue: the system had to support dispatchers and drivers across multiple US time zones, and timestamps kept mismatching. We standardized server time on UTC and let users switch to local display time.</p><p>Our route planning project took this further by design: before general rollout, beta testing was done by actual London bus drivers on live routes, evaluating both functionality and usability. In practice, the depth of field testing depends on your access to end users — but even a week of structured testing with 5 real devices catches a class of bugs no QA lab will.</p><h4>Not running a proof of concept for critical integrations</h4><p>If your product depends on a third-party service behaving in a specific way — a maps API, a government database, an existing internal platform — that assumption deserves a proof of concept before a development budget gets committed to it.</p><p>Here’s what usually happens without one: development runs for 3 months around an integration, then the API turns out to be unable to do the one thing the product needs (rate limits, missing data fields, licensing terms), and the architecture has to bend around the gap.</p><p>On the route planning project, the discovery phase existed specifically to de-risk this. The product had to integrate with Azure Maps, government hazard databases, and the client’s existing tracking platform — 3 separate integration risks. We validated prototypes and a proof of concept covering the critical functionality before full development started. The PoC did double duty, too: it was the artifact that secured labor union approval for the product’s launch, which was a hard gate for the whole project.</p><p>A PoC is a narrow technical answer to a specific question: “Can Azure Maps route a vehicle of this size around this bridge?” If the answer is yes, you’ve spent a couple of weeks buying certainty. If it’s not, you’ve saved a rebuild.</p><h4>Treating location privacy and permissions as a post-launch task</h4><p>This one’s ours to add, because it derails more geolocation products than any map bug: the tracking features get designed first, with permissions penciled in for later. Later arrives as an App Store rejection or a feature that legally can’t ship.</p><p>The background location is the sharpest edge. On a location-based chat project, the core request was continuous real-time access to other users’ locations — including while the app runs in the background. Both Google Play and the App Store treat background location as a restricted capability with explicit review requirements. We implemented background tracking over WebSockets and mapped the approach against both stores’ rules before building, so the feature shipped instead of bouncing off review.</p><p>Privacy controls also turned out to be product features. The users can share real-time location with favorite friends only, all friends, everyone, or nobody — and can mark private places where their presence is hidden. In our experience, users who can control their visibility tend to keep location sharing on; users who can’t tend to switch it off, which quietly removes the “location-based” part of the product.</p><p>The safer move: design the permission flow and privacy model in discovery, alongside the feature list, and read the current store policies before committing to any background behavior.</p><h3>What to settle before development starts</h3><p>The common thread across all 5 mistakes is timing — the answers exist, they just get found too late. Before committing to a build, we recommend having written answers to these:</p><ul><li><strong>Provider fit:</strong> Which map provider matches your cloud ecosystem, data ownership needs, and routing requirements — and what does it cost at 10x launch traffic?</li><li><strong>Scale ceiling:</strong> How many markers, users, or vehicles will the map realistically show, and what’s the rendering strategy once that number passes a few hundred?</li><li><strong>Field conditions:</strong> Which real devices, OS versions, and network conditions will the app face, and when does field testing start?</li><li><strong>Integration proof:</strong> Which third-party dependency is critical enough to break the product, and has a PoC confirmed it actually works?</li><li><strong>Privacy model:</strong> What location data do you collect, who sees it, and does the plan pass current Google Play and App Store background-location rules?</li></ul><p>If any of these gets a shrug instead of an answer, that’s your discovery phase agenda. Two to four weeks of answering them upfront is consistently cheaper than any of the fixes described above — and unlike the fixes, it happens on your schedule instead of production’s.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7738d98259ae" width="1" height="1" alt=""><hr><p><a href="https://medium.com/think-clockwise/common-mistakes-in-location-based-app-development-7738d98259ae">Common mistakes in location-based app development</a> was originally published in <a href="https://medium.com/think-clockwise">Think Clockwise</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The analytics features users never see but can’t live without]]></title>
            <link>https://medium.com/think-clockwise/the-analytics-features-users-never-see-but-cant-live-without-15169356e10a?source=rss-8f0cde6ebd27------2</link>
            <guid isPermaLink="false">https://medium.com/p/15169356e10a</guid>
            <category><![CDATA[analytics]]></category>
            <dc:creator><![CDATA[Bogdan Yemets]]></dc:creator>
            <pubDate>Fri, 31 Jul 2026 11:06:47 GMT</pubDate>
            <atom:updated>2026-07-31T11:06:47.405Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*eYXFVWRbcSrwrxqskEIdYg.jpeg" /></figure><p>Every analytics platform gets judged by its dashboards. Charts, filters, color-coded tiles — the parts people screenshot and send to their boss. Meanwhile, the functionality that makes those dashboards possible stays hidden by design. As an <a href="https://clockwise.software/">outsourcing software development company</a> that has built 10+ data analytics platforms, we can put a number on it: roughly half of the engineering effort goes into features users will never click, open, or even know exist.</p><p>There are 4 of them: data ingestion, data pipelines, data storage, and data export. Remove any one, and your platform turns into a pretty interface with zero utility. Here’s what each layer does, how we usually build it, and why it deserves at least as much attention as the visuals on top.</p><h3>Data ingestion: How information gets in</h3><p>Before your platform analyzes anything, data has to arrive. Ingestion is the front door — the machinery that pulls information from every source you rely on: your CRM, social media APIs, ad platforms, file uploads, third-party services.</p><p>In our projects, this layer usually runs on webhooks, API calls, or file servers for collecting the data, with Lambda functions paired with SQS, Kafka, or AWS Batch handling the queues.</p><p>The queues are the underrated part. Sources push data at unpredictable rates, third-party APIs throttle requests, and services go down at 3 a.m. A queue absorbs the spikes, retries the failures, and keeps a flaky external API a minor annoyance instead of a hole in your monthly numbers. Users open the platform in the morning, see complete charts, and assume the data simply appears. That assumption is the queue doing its job.</p><p>One habit we keep from years of these projects: when a platform depends on pulling data from an outside source, we study the API documentation and build a small proof of concept first. Rate limits and corner cases love to hide until week six of development — a PoC drags them into the open while changing course is still cheap.</p><h3>Data pipelines: Where raw data becomes usable</h3><p>Ingested data arrives messy — mixed formats, duplicates, gaps, mismatched field names. Pipelines transform, filter, group, and save that incoming stream so it lands ready for analysis. We build them with message brokers like Apache Kafka and Amazon Kinesis, orchestrated by tools like Apache Airflow and AWS Step Functions.</p><p>This is the layer that quietly eats entire jobs. One of our clients, a telecom broker, used to spend days every month manually matching data across sources. We designed a system where pipelines handle the heavy lifting, and the same process now takes minutes.</p><p>Another client, a top Australian marketing agency, ran cross-channel analysis through complex custom Excel formulas, updating every metric by hand each month. We translated those formulas into algorithms inside their platform, and it now pulls in fresh data monthly, analyzes it, and visualizes the results on its own. Users just see updated numbers. The pipeline is the reason those numbers are correct.</p><h3>Data storage: The choice that ages fastest</h3><p>All that processed data needs a home. For modest volumes, a general-purpose relational database like PostgreSQL or MySQL handles the job well. Past a certain scale, though, analytical queries start straining a row-based database, and a solution built for business intelligence workloads — we often go with Cassandra or Amazon Redshift — becomes the smarter pick.</p><p>This decision deserves extra care for one reason: storage is the most expensive layer to change later. Swapping a chart library is a sprint. Migrating a production database with years of accumulated data is a project of its own, with downtime risks attached. That’s why we lock in the storage choice during discovery, based on current data volume, expected growth, and whether existing data needs migrating into the new platform.</p><p>Databases collect zero praise from users. Complaints show up the moment a dashboard takes eleven seconds to load — and by then, the fix costs ten times what the right early decision would have.</p><h3>Data export: Getting insights back out</h3><p>The best analysis earns its keep only when results reach the right people and tools. Export is the return trip: pushing processed data back into your CRM, sending insights to tools like Notion, or generating files your team can share.</p><p>Technically, it mirrors ingestion — APIs, webhooks, and file servers, wired in the opposite direction. For a telecom provider, we configured exports so users could download dashboard data in CSV and XLS formats, ready to view, edit, and pass along in other apps. The download button is the visible bit. The system assembling those files on demand, straight from live analytical data, is the part everyone forgets to budget for.</p><p><strong>The invisible layer, layer by layer</strong></p><p><strong>Data ingestion</strong> is where everything starts. It pulls data in from all your sources — no matter where they live. Typical tools: webhooks, API calls, file servers, Lambda + SQS, Kafka, AWS Batch.</p><p><strong>Data pipelines</strong> take that raw incoming data and do the real work on it — transforming, filtering, grouping, and saving it into a usable shape. Typical tools: Apache Kafka, Amazon Kinesis, Apache Airflow, AWS Step Functions.</p><p><strong>Data storage</strong> keeps everything sitting ready for fast analytical queries, so nothing crawls when you actually need answers. Typical tools: PostgreSQL, MySQL, Cassandra, Amazon Redshift.</p><p><strong>Data export</strong> is the last mile — it delivers results out to people and other systems. Typical tools: APIs, webhooks, file servers.</p><p>That’s the flow, start to finish: data comes in, gets shaped, gets stored, goes out. Want me to drop this into a doc, or tweak the phrasing anywhere?</p><h3>Why founders under-budget this layer</h3><p>Analytics demos show dashboards, so estimates get anchored to dashboards. Then development starts, and the invisible layer claims a serious share of the budget — a full data analytics platform typically lands between $100,000 and $200,000 over 6+ months, and the plumbing accounts for a healthy portion of that.</p><p>The safer move is deciding this layer early. During a discovery stage (from 3 weeks and about $12,000), we map your data sources, volumes, formats, and destinations, then design the architecture around them. A platform planned this way scales on schedule instead of surprising you at 10x the data volume. Skipping this step tends to produce the opposite: a beautiful interface in month two, and an ingestion rewrite in month eight.</p><h3>The quiet layer wins</h3><p>Dashboards collect the applause; these four layers do the work. Nearly every analytics platform we’ve built started the same way — a founder describing their data situation on a call, and us sketching which combination of ingestion, pipelines, storage, and export would turn that tangle into clean Monday-morning insights. The visible features come later. By then, they’re the easy part.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=15169356e10a" width="1" height="1" alt=""><hr><p><a href="https://medium.com/think-clockwise/the-analytics-features-users-never-see-but-cant-live-without-15169356e10a">The analytics features users never see but can’t live without</a> was originally published in <a href="https://medium.com/think-clockwise">Think Clockwise</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Making logistics data actionable for ops team: What worked for our clients]]></title>
            <link>https://medium.com/@bogdan.yemets/making-logistics-data-actionable-for-ops-team-what-worked-for-our-clients-13a2d8059f29?source=rss-8f0cde6ebd27------2</link>
            <guid isPermaLink="false">https://medium.com/p/13a2d8059f29</guid>
            <category><![CDATA[logistics]]></category>
            <dc:creator><![CDATA[Bogdan Yemets]]></dc:creator>
            <pubDate>Thu, 25 Jun 2026 10:49:47 GMT</pubDate>
            <atom:updated>2026-06-25T10:49:47.139Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/747/1*NaQQUjgrA4TsMo9nCISKlw.png" /></figure><p>Here is a pattern we run into on most logistics projects: the data is already there. GPS feeds, ERP records, warehouse logs, driver check-ins, fuel reports. Nobody is starved for numbers. What they are missing is a way for a dispatcher to glance at one screen and know what to do in the next ten minutes.</p><p>A dashboard doesn’t fix that on its own. We have seen plenty of dashboards that look great in a demo and sit unused a month later. The ones that stick share a few habits — we’ll share them in this blog post.</p><p>A bit of context on where this comes from. As a logistics development company, we have shipped 10+ logistics apps over the last 10 years, including operator dashboards for the top bus provider in London, one of Ukraine’s largest construction materials manufacturers, and a US waste hauler. Across that work, the dashboards we built have cut delivery costs by roughly 30% and route-planning time by about 40%. Here is what that looked like on two of those projects, where ops teams went from chasing information to acting on it.</p><h4>UDK: one screen instead of 5 phone calls</h4><p>UDK is one of the largest construction materials manufacturers in Ukraine. Their delivery operation ran across accounting, the warehouse, and external carriers, and none of those systems talked to each other. To confirm the status of a single order, someone made phone calls. A lot of them. Reconciling what really happened meant matching spreadsheets by hand.</p><p>We built a centralized ERP platform called WebOffice. It pulled the scattered data from accounting, the warehouse, and delivery partners into one interface. Order placement, carrier auctions, vehicle loading schedules, delivery tracking: all of it lived in the same place and updated in real time.</p><p>The dashboard was only the visible part. We had to make a dispatcher trust it enough to skip the phone call. It only happens when you reconcile and validate the data before anyone looks at it.</p><p>The first-year numbers: we cut 5+ phone calls per order and saved $100,000 on delivery costs. The phone calls matter more than they sound. Every one of them was a person stopping their work to ask another person a question the system should have answered.</p><p><strong>What worked for UDK:</strong></p><ul><li>One platform pulling accounting, warehouse, and carrier data into a single interface</li><li>Data reconciled and validated before it hit the screen, so dispatchers trusted it</li><li>A live view of the whole flow: order placement, carrier auctions, loading schedules, delivery tracking</li></ul><h3>A Denver waste hauler: getting drivers off the office floor</h3><p>The second project we want to share was for a waste transportation company based in Denver. Their dispatch ran on paper and conversation. Drivers came into the office to pick up work orders, came back for new ones, and called in when something changed. Tracking the work, and proving it later for accounting, was a daily headache. The reporting side, in particular, had become a mess that no one could untangle at month-end.</p><p>We built a fleet management platform with two halves: a dispatcher console on the web and a mobile app synced to it for drivers. Dispatchers assign work orders digitally and watch their status change in real time. Drivers get their next job on the phone, so the trips back to the office stopped.</p><p>Reporting is where the data turned useful instead of decorative. The platform generates reports on driver performance, completed orders, and routes used, automatically. The dispatcher sees the whole fleet at a glance. Accounting gets clean records without chasing anyone. Same operation, far less friction.</p><p><strong>What worked for the Denver team:</strong></p><ul><li>A web dispatcher console paired with a driver mobile app, kept in sync</li><li>Digital work orders with live status, so drivers stopped driving back for their next job</li><li>Automatic reports on driver performance, completed orders, and routes used</li><li>Clean records flowing straight to accounting</li></ul><h4>What made the data usable</h4><p>Two projects, two industries, one pattern. The data was already there in both cases. The job was turning it into something an ops team could act on during a shift. A few things made the difference.</p><h4>Reconcile the data before you show it</h4><p>A dashboard that displays wrong numbers is worse than no dashboard, because people stop trusting it and quietly go back to phone calls. On both projects we spent real effort normalizing and validating inputs from systems that updated at different speeds and sometimes disagreed. Boring work. It is also the reason the screens earned trust.</p><h4>Build around events</h4><p>A dispatcher does not need a weekly average. They need to know that this pickup got missed and this truck went off route, right now, while there is still time to fix it. We designed both consoles to surface the thing that changed, so the exception finds the operator instead of the operator hunting for it.</p><h4>Match the tool to the person using it</h4><p>The UDK dispatcher and the Denver driver work on different platforms built for different teams and workflows. Rather than creating role-based views within a single system, each project was designed around the specific needs of its users. The dispatcher platform focuses on coordination and oversight, while the driver platform provides the tools and functionality drivers need in the field. This approach ensures that each team has exactly the capabilities they need, without unnecessary features or complexity.</p><h4>Where this leaves you</h4><p>If your team has the data but still runs on phone calls and spreadsheets, the fix is rarely more reporting. It’s automation and one trustworthy view that shows what changed and what to do about it. Both clients got there the same way: they cleaned up the inputs first, then shaped the output around the people who use it every day.</p><p>Most of those decisions happen before the tech part, when you settle which inputs to trust, which events to surface, and who gets to see what. Get that part wrong and you end up with another dashboard that demos well and goes unused a month later. We have made those calls enough times to know where they tend to go sideways, and that kind of judgment is hard to pull from a template.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=13a2d8059f29" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Platform redesign that delivered 8x faster loading and much smoother user experience]]></title>
            <link>https://medium.com/think-clockwise/platform-redesign-that-delivered-8x-faster-loading-and-much-smoother-user-experience-db4f9d58f7d5?source=rss-8f0cde6ebd27------2</link>
            <guid isPermaLink="false">https://medium.com/p/db4f9d58f7d5</guid>
            <category><![CDATA[platform-modernization]]></category>
            <dc:creator><![CDATA[Bogdan Yemets]]></dc:creator>
            <pubDate>Thu, 21 May 2026 15:33:16 GMT</pubDate>
            <atom:updated>2026-05-21T15:33:16.121Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*5STjD1W6gqRK67TGIscJaw.jpeg" /></figure><p>Maintaining a legacy system eventually becomes a struggle against your own data. Most product owners know the frustration of watching a system stall because the code is too old to handle more users or new features.</p><p>But how do you remove years of technical debt without breaking the core product?</p><p>Me and my teammates found an answer to that question while working on a lead parsing platform that had hit its performance limit. And our work resulted in an 8x speed increase and an interface that finally stays out of the user’s way.</p><h3>The background: Where we started</h3><p>Our client had a platform that handled lead data across dozens of websites. As the database grew, the platform became so slow that it couldn’t handle incoming traffic or user requests. Adding new features made the system noticeably heavier, which turned simple tasks into a struggle. So, he started looking for a <a href="https://clockwise.software/">software development outsourcing company</a> that could handle this problem, and found us.</p><p>Our lead developer, Alex, saw the problem and suggested a different path. Instead of patching an outdated backend and a legacy frontend, he proposed to rebuild one core page using a modern tech stack on the frontend to prove how much faster the platform could run. The plan was simple: if the client liked the results, we would gradually migrate the rest of the system to avoid any downtime. A client agreed, so we started working.</p><h3>The push to 8x: How we rebuilt for speed</h3><p>The first test page Alex built proved we could do better. Once the client saw the performance jump, we committed to a full migration. We wanted to remove every friction point that had built up over the years. This meant rethinking how the platform handled data at every level.</p><h3>Cutting the weight</h3><p>The biggest drag on the system was the legacy tech. We started by removing AngularJS entirely. By moving to Vue.js, we cut the frontend package size by 3x. Naturally, this change led to fewer downloads and less parsing. Also, we moved the heavy lifting to a high-speed data server, which cleared congestion on the old backend.</p><h3>Maximizing the workspace</h3><p>Beyond improving the platform’s technical foundation, we also revisited the overall lead analysis flow. The existing workspace used screen space inefficiently, and users often had to dig through a cramped interface to find the data they needed. To solve this, we reimagined the leads analysis workflow with a new column-sorting system and seamless subdomain switching, while also rebuilding the core leads module to handle near real-time lead inflow across dozens of websites without the previous lag. As a result, we reclaimed 50% more effective space on the leads table, allowing users to view twice as much data at once.</p><h3>From tables to charts</h3><p>The legacy system relied on simple tables, making it hard to spot patterns. We replaced those with a dynamic leads chart that tracks confirmed and missed leads over time. It allowed the client to see trends at a glance.</p><p>By focusing on a mix of bug fixing, refactoring, and adding these new features, we turned a sluggish tool into an optimized frontend experience. If we sum it up, here is what actually changed:</p><ul><li><strong>Removed AngularJS</strong> to eliminate legacy overhead</li><li><strong>Reduced frontend package size by 3x</strong> for faster loading</li><li><strong>Deployed a new data server</strong> for near-instant responses to lead inflows</li><li><strong>Optimized the leads table</strong> to show 50% more information at once</li><li><strong>Built instant filtering and aggregation</strong> to handle complex data segmentation without delays</li></ul><h3>How we organized our development process</h3><p>Because the platform was already live, changes had to be rolled out in small, carefully managed increments without disrupting the client’s daily operations or drawing attention from users. To coordinate that kind of gradual transformation, we introduced an adaptive Kanban process that kept the project moving steadily while allowing the team to respond quickly to shifting priorities.</p><h3>A low-risk approach</h3><p>We followed a logical path from the simplest modules to the most complex ones. This allowed us to build momentum and refine our strategy as we went. Our cycle was straightforward:</p><ol><li>Build the module using the new tech stack.</li><li>Verify performance and stability within our team.</li><li>Release to users and listen to their experience.</li><li>Fix issues and polish features based on real-world use.</li><li>Move to the next module.</li></ol><p>This method ensured that we weren’t just refactoring old code, but simultaneously delivering new, high-value features.</p><h3>Who was involved?</h3><p>To make that possible, we brought together a small cross-functional team:</p><ul><li><strong>Project manager</strong>, who kept communication clear and priorities aligned</li><li><strong>Front-end engineer</strong>, who led the Vue.js migration and UI optimization</li><li><strong>Two backend engineers</strong>, who improved data logic, server performance, and the new parsing system</li><li><strong>QA engineer</strong>, who made sure every release stayed stable and bug-free</li></ul><h3>Conclusion</h3><p>By systematically removing technical debt and optimizing the core modules, we turned a stalling lead-parsing platform into a high-performance tool that handles real-time data with ease. Now, the system provides a scalable foundation for any new features the client needs to build.</p><p>So, the moral of the story is simple:If your current software is starting to hold your business back, it might be time to rethink the engine under the hood. And yes, it can be done without downtime.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=db4f9d58f7d5" width="1" height="1" alt=""><hr><p><a href="https://medium.com/think-clockwise/platform-redesign-that-delivered-8x-faster-loading-and-much-smoother-user-experience-db4f9d58f7d5">Platform redesign that delivered 8x faster loading and much smoother user experience</a> was originally published in <a href="https://medium.com/think-clockwise">Think Clockwise</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How we designed ERP architecture for an enterprise with 1000+ employees]]></title>
            <link>https://medium.com/think-clockwise/how-we-designed-erp-architecture-for-an-enterprise-with-1000-employees-57b1229fe250?source=rss-8f0cde6ebd27------2</link>
            <guid isPermaLink="false">https://medium.com/p/57b1229fe250</guid>
            <category><![CDATA[erp-software]]></category>
            <category><![CDATA[erp-solutions]]></category>
            <category><![CDATA[erp-development]]></category>
            <category><![CDATA[erp-system]]></category>
            <dc:creator><![CDATA[Bogdan Yemets]]></dc:creator>
            <pubDate>Thu, 19 Mar 2026 16:20:00 GMT</pubDate>
            <atom:updated>2026-03-19T16:20:00.051Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WNS6bh1o3l985KXR-vnDuw.jpeg" /></figure><p>Producing <strong>10% of a country’s aerated concrete</strong> with<strong> 1,000+ employees</strong> is a massive operation, but you can’t run it on phone calls and spreadsheets. When your production capacity outpaces your administrative speed, those manual processes become a hard ceiling on your revenue.</p><p>UDK, Ukraine’s top producer in the industry, hit this wall. Their production was huge, but their order processing was outdated. Every sale triggered a chain of manual check-ins just to confirm if a truck could leave the yard.</p><p>As an <a href="https://clockwise.software/erp-software-development/">ERP software development company</a>, we knew how to replace those manual hurdles with a high-load digital core.</p><p>Here’s how we handled the discovery, connected their existing legacy software to a modular system, and automated the logistics to keep the scale moving.</p><h3>Discovery phase: How we uncovered the invisible bottlenecks</h3><p>We started by looking at why a company <strong>pumping out 440,000 cubic meters of concrete</strong> was getting stuck in its own internal workflows. In an enterprise this size, the official manual rarely matches reality. The real process lives in how information flows through the organization’s day-to-day work.</p><h3>Auditing the human middleware</h3><p>We shadowed the sales and logistics teams to track how an order travels. We found that instead of a system handling the flow, employees were spending their entire day manually connecting disconnected apps.</p><ul><li><strong>The 5-call cycle:</strong> Processing one wholesale order averaged 5 phone calls between the client, sales, the warehouse, and logistics. This created constant lag and room for error.</li><li><strong>Manual data syncing:</strong> Managers were checking credit limits in one accounting tool, stock levels in another, and then typing that into a third spreadsheet to book a truck.</li><li><strong>The status gap:</strong> Since clients could not track their own orders, <strong>sales spent 40% of their workday</strong> just answering questions about truck locations.</li></ul><h3>Defining the strategy: Choosing where to build</h3><p>The goal was to find the fastest way to get a return on the investment. For example, we decided not to touch the core accounting software because it already worked and suited the company. Instead, we focused on building a web-based layer that handled the entire workflow around it.</p><p>This approach allowed us to:</p><ul><li><strong>Keep the core intact:</strong> We left the existing financial database alone to avoid unnecessary risk and downtime.</li><li><strong>Surface the data:</strong> We built a modern interface that pulled real-time info from those legacy systems and put it directly in front of clients and managers.</li><li><strong>Automate the heavy lifting:</strong> We prioritized the order management and logistics modules. These were the specific areas where manual errors and slow data entry were actually costing the company money.</li></ul><p>By the end of this phase, we had a clear software design. This central system connected production data directly to customer and employee screens and automated the manual steps in between.</p><h3>Architectural strategy: Modernization without disruption</h3><p>Updating an enterprise of this size is risky if you try to replace everything at once. We avoided that high-cost trap by building an interface layer that sits on top of the existing setup. This gave users a modern way to work while the core systems stayed exactly as they were.</p><h3>Building in stages</h3><p>We broke the system into independent parts and prioritized the most critical ones first. This allowed us to fix order management and logistics months before we even looked at other features. By rolling out the software in stages, the business started seeing a return on the investment almost immediately instead of waiting years for a massive launch.</p><h3>The technical priorities</h3><p>The foundation of the system was built on 3 specific technical priorities to ensure long-term success:</p><ul><li><strong>Data unification: </strong>We defined exactly how all data entities, such as customer records and inventory levels, interact across different systems to ensure information is always consistent.</li><li><strong>Scalability and performance:</strong> The architecture was designed specifically for high load, capable of processing the massive amounts of data required to coordinate 90+ heavy trucks and their deliveries every day.</li><li><strong>Stability:</strong> For a 24/7 manufacturing facility like UDK, downtime is not an option. The system was engineered to remain stable and accessible around the clock.</li></ul><h3>Building the core modules</h3><p>The strength of the ERP lies in its specialized modules, each designed to automate a specific vertical of the business. By breaking the system into these focused components, we addressed the unique requirements of every department while maintaining a unified data flow.</p><h3>The client cabinet</h3><p>To eliminate the need for constant phone communication, we built a self-service portal for wholesale partners. This module allows clients to browse the product catalog, view their own documentation, and track order statuses in real-time. By moving these tasks to a digital interface, the sales team was freed from basic information requests.</p><h3>Logistics and distribution engine</h3><p>This is the most complex component of the architecture, focusing on a custom algorithm to optimize shipping.</p><ul><li>Algorithm focus: The system automatically calculates delivery schedules by considering different vehicle loading types and the most efficient route distributions.</li><li>Real-time sync: By connecting the warehouse directly to the delivery fleet, the engine ensures that truck dispatching perfectly matches the physical reality of the loading bays.</li></ul><h3>Finance and automated documentation</h3><p>The ERP communicates directly with the existing accounting setup to handle automated invoicing as soon as an order is fulfilled. By shifting to automatic updates, we removed the need for double-entry and cut the risk of human error. When a manager updates a price or a stock level, that change reflects across the entire organization immediately. The system also manages complex credit limits and tiered discounts, automatically keeping every sale within approved financial boundaries.</p><h3>The admin panel</h3><p>To give leadership full visibility, we designed a centralized dashboard for management. This panel provides a high-level view of system statistics, production metrics, and user management, allowing for data-driven decisions based on live information from the factory floor and the field.</p><h3>Integration and data integrity</h3><p>To ensure the system stays manageable long-term, we maintained a detailed software design description. This technical roadmap allows any team to maintain or expand the platform, which removes the risk of being locked into a single vendor and keeps the architecture transparent for the company.</p><h3>Results</h3><p>The shift from manual steps to a structured digital system gave UDK a clear return on the investment. By focusing on how the work actually scales, the company hit several major operational milestones.</p><ul><li><strong>Ending the call cycle:</strong> We cleared the administrative noise by giving clients and managers direct access to real-time data. The 5-call requirement per order was completely removed.</li><li><strong>Doubling capacity:</strong> The automated dispatching logic allowed the company to jump <strong>from 45 to 92 trucks per day</strong>. They doubled their output without needing to add more logistics staff to handle the load.</li><li><strong>Direct savings:</strong> Beyond the time recovered, the system cut down delivery inefficiencies enough to <strong>save the company $100,000 in logistics costs</strong> during the first year alone.</li></ul><h3>Lessons learned: How to ship an enterprise ERP</h3><p>Building an ERP for an operation like UDK highlights a few fundamental rules for any enterprise-level project. They might sound simple, but they are usually what decides if the system works as it should or just becomes expensive shelfware:</p><ul><li><strong>Start with the pain point:</strong> Do not attempt to build a massive three-year roadmap that tries to solve everything at once. Focus on the single biggest bottleneck first to gain momentum and see an immediate return.</li><li><strong>Architecture is evolution:</strong> An ERP should not be a rigid, finished product. View it as a living system that can evolve and scale as the business grows and market conditions change.</li><li><strong>User adoption is key:</strong> When designing for people who have spent years using spreadsheets and phone calls, the new interface must be more intuitive and helpful than the manual alternative. If the system is easy to use, the transition happens naturally.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=57b1229fe250" width="1" height="1" alt=""><hr><p><a href="https://medium.com/think-clockwise/how-we-designed-erp-architecture-for-an-enterprise-with-1000-employees-57b1229fe250">How we designed ERP architecture for an enterprise with 1000+ employees</a> was originally published in <a href="https://medium.com/think-clockwise">Think Clockwise</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How we customized maps to safely navigate London’s top-1 bus operator]]></title>
            <link>https://medium.com/think-clockwise/how-we-customized-maps-to-safely-navigate-londons-top-1-bus-operator-970d771eb6b7?source=rss-8f0cde6ebd27------2</link>
            <guid isPermaLink="false">https://medium.com/p/970d771eb6b7</guid>
            <category><![CDATA[location-intelligence]]></category>
            <category><![CDATA[logistics-companies]]></category>
            <category><![CDATA[logistics]]></category>
            <dc:creator><![CDATA[Bogdan Yemets]]></dc:creator>
            <pubDate>Wed, 18 Mar 2026 12:01:44 GMT</pubDate>
            <atom:updated>2026-03-18T12:11:32.447Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cvOoP9OHVKoc7ajfauA5VA.jpeg" /></figure><p>Location tracking and route planning are the nervous system of urban logistics. While basic tracking provides real-time visibility, professional-grade route planning is what calculates the safest path, factoring in complex variables like vehicle dimensions, legal weight limits, and hyper-local urban restrictions that standard consumer apps simply ignore.</p><p>But how do you move heavy goods through a city as ancient and congested as London without hitting a literal or legal wall?</p><p>In this article, we’ll break down how we, as a <a href="https://clockwise.software/logistics/">logistics software development company</a>, engineered a navigation ecosystem specifically tailored to London’s architectural constraints. We’ll detail the custom logic used to solve the low-bridge problem via Azure Maps, how we used NFC-enabled logins to solve driver-side friction, and the architectural steps taken to turn this solution into a scalable white-label platform.</p><h3>Project background: The high stakes of London’s streets</h3><p>The UK experiences nearly<strong> 2,000 bridge strikes every year</strong>. And these are not just minor fender benders. When a double-decker bus or a heavy goods vehicle hits a low-clearance bridge, the fallout is immediate: millions in structural damage, gridlocked commuter routes, and genuine risks to public safety.</p><p>Standard navigation apps don’t solve this because they prioritize the fastest arrival time. They don’t know a vehicle’s height or weight limits.</p><p>Foresolutions brought us in to build something better for London bus operators. The goal was to replace generic GPS logic with a system that treats a vehicle’s physical dimensions as the primary routing constraint. In this environment, the only viable path is the one that actually fits the bus.</p><p><strong>Why standard GPS fails professional drivers</strong></p><p>Standard GPS apps work for a weekend drive, but not for a bus operator. These apps are fundamentally built for passenger cars, which means they miss the data layer needed to move a six-ton vehicle through London.</p><ul><li><strong>Dimensional awareness.</strong> Consumer apps don’t know a vehicle’s height, weight, or width. This is how you end up with a double-decker wedged under a low bridge or stuck in a narrow tunnel.</li><li><strong>Granular hazard reporting.</strong> Professional drivers need lane-specific data and real-time updates on public transport risks to hit strict arrival windows.</li><li><strong>System integration.</strong> A fleet solution needs to connect to central dispatch and respect labor union interface requirements, which you won’t find in a consumer-grade app.</li></ul><p>These gaps shaped the need for a new solution, one that would account for all those specifics. So, we got to planning it.</p><h3>Project discovery: Securing approvals and validating the roadmap</h3><p>In the high-stakes world of London transport, a new tool must be both technically sound and socially viable. To build an app that meets both points, we started with a discovery phase, focusing on 4 key objectives:</p><ol><li><strong>Getting the green light from labor unions.</strong> We presented medium-fidelity prototypes and a clear proof of concept to union representatives. It was about proving the app would make a driver’s life easier rather than just adding another layer of digital noise to their day.</li><li><strong>Testing the tech stack.</strong> We had to be sure that pulling data from Azure Maps and government hazard databases would be fast and accurate.</li><li><strong>Mapping the architecture.</strong> We designed the blueprint for how the driver-facing app and the operator web platform would talk to each other.</li><li><strong>Spotting bottlenecks early</strong>. Searching for things that could trip the team up later, like data lag or a clunky onboarding process, helped us solve those headaches before they became expensive development problems.</li></ol><p>With the design docs, feature requirements, and architecture locked in, we were ready to move into development.</p><h3>Project development: Solving the technical puzzle</h3><p>We focused on 3 pillars to turn the app into something drivers can rely on:</p><ul><li><strong>Custom routing via Azure Maps.</strong> To address the low-bridge problem, we dove into the Azure Maps API. By plugging in the exact height, weight, and width of a double-decker, we made the system automatically filter the London road network. It only shows paths that are physically possible and legal. This means a driver is never routed into a hazard that a normal GPS would not even notice.</li><li><strong>Frictionless authentication.</strong> Manual logins are a headache at the start of a shift. We used NFC tech to fix that. A driver just taps a programmed card against the device to get instant access to route and shift data. It keeps the focus on the road and cuts out unnecessary screen time.</li><li><strong>Real-time sync and hazard alerts.</strong> We built a solid data pipeline between the mobile app and the operator dashboard to keep everyone on the same page. If a new risk pops up, like a sudden road closure or an accident, administrators can pin it or pull it from live traffic sources. Those updates trigger instant notifications for any driver heading toward those coordinates.</li></ul><h3>Scalability: Building for a global white-label future</h3><p>Beyond solving the immediate needs of London’s transit system, we designed the software with a long-term commercial vision. The architecture was built from the ground up to be a white-label asset, allowing Foresolutions to offer the same level of precision to other transport authorities worldwide.</p><p>To make the platform adaptable, we focused on 2 main principles:</p><ul><li><strong>Multitenancy by design.</strong> We used a multi-tenant structure so a single software instance can serve different organizations while keeping their data completely isolated. It allows for fast onboarding of new fleets without deploying new infrastructure every time a new client comes on board.</li><li><strong>Modular UI and role-based access.</strong> We built the interface with modules that are easy to rebrand or adjust for specific regional needs. Also, we set up role-based access to ensure that whether a user is a driver or a fleet manager, they only see the tools and data relevant to their job.</li></ul><p>By combining deep map customization with a scalable backend, we delivered a platform that makes urban transport safer and provides a solid foundation for growth later.</p><h3>Project outcome: Safer streets and a roadmap for the future</h3><p>By moving from theoretical planning to field execution, the solution is already providing the data necessary to refine its impact on urban safety.</p><p>The primary outcomes of this deployment include:</p><ul><li><strong>Fewer accidents and safer streets.</strong> The client expects a major drop in injuries and structural damage. By making it impossible for a double-decker to even get routed toward a low bridge, we addressed the root cause of the most expensive delays in the UK transport network.</li><li><strong>Continuous feedback loops.</strong> Real-time insights from drivers now fill a solid product backlog. This means every update to the web platform or the mobile app is based not on theories, but on what is happening on the road.</li><li><strong>Ready for global expansion.</strong> Since we built this on multi-tenancy principles, the platform is already primed for white-labeling. Foresolutions can now offer this high-precision tool to other companies worldwide without rewriting the core engine.</li></ul><h3>Conclusion</h3><p>Building a navigation solution for London’s bus network showed us that high-stakes logistics need something beyond a generic map. We focused on vehicle-specific constraints and driver safety, not speed alone. That shift turned a basic tracking tool into something drivers can rely on in critical situations.</p><p>The project also reflects a broader move toward tools that account for real city conditions, instead of ideal routes on a map.</p><p>We’ll measure success in simple terms. Fewer accidents and more reliable public transport. With a scalable, white-label architecture already in place, the platform can be adapted for other cities facing similar challenges.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=970d771eb6b7" width="1" height="1" alt=""><hr><p><a href="https://medium.com/think-clockwise/how-we-customized-maps-to-safely-navigate-londons-top-1-bus-operator-970d771eb6b7">How we customized maps to safely navigate London’s top-1 bus operator</a> was originally published in <a href="https://medium.com/think-clockwise">Think Clockwise</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>