<?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 VerticalServe Blogs on Medium]]></title>
        <description><![CDATA[Stories by VerticalServe Blogs on Medium]]></description>
        <link>https://medium.com/@verticalserve?source=rss-79851bb07a23------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*KZFMJ8RM4bmOgyPzaBoVCA.jpeg</url>
            <title>Stories by VerticalServe Blogs on Medium</title>
            <link>https://medium.com/@verticalserve?source=rss-79851bb07a23------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 05:02:05 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@verticalserve/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[Leaving Tableau Is Won or Lost One Calculation at a Time]]></title>
            <link>https://verticalserve.medium.com/leaving-tableau-is-won-or-lost-one-calculation-at-a-time-d46694ed63a8?source=rss-79851bb07a23------2</link>
            <guid isPermaLink="false">https://medium.com/p/d46694ed63a8</guid>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[llm]]></category>
            <dc:creator><![CDATA[VerticalServe Blogs]]></dc:creator>
            <pubDate>Thu, 15 Oct 2026 00:28:29 GMT</pubDate>
            <atom:updated>2026-10-15T00:28:29.014Z</atom:updated>
            <content:encoded><![CDATA[<h4>What translates, what gets flagged for a person, and the category we refuse to translate on purpose.</h4><p>A workbook’s charts are the visible part and the easy part. Its calculated fields are a decade of business logic written by people who have mostly left, documented nowhere else, and load-bearing in ways nobody can enumerate.</p><p>So the useful question about any migration tool is not “does it convert calculations”. It is: which ones, how confidently, and what does it do about the rest.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*-ad7eShqcnblRXZP.png" /></figure><h3>Three verdicts, not two</h3><p>Every calculation comes out of the importer with one of three statuses. The middle one matters most, and most tools do not have it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*wg168rRnWamPUQsl.png" /></figure><blockquote><em>A tool that reports only converted and failed is hiding the middle category, which is where the wrong numbers come from. Something that converted and might be subtly different is not a success.</em></blockquote><p><em>A worked example</em></p><h4>Six real calculated fields from one workbook</h4><p>Taken from a nine-year-old regional sales workbook. Each is shown as the author wrote it, then as it comes out, then the verdict.</p><p><strong>1. Null-safe margin.</strong> The most common shape in any estate.</p><p>ZN([Revenue]) — ZN([Cost]) → COALESCE(revenue, 0) — COALESCE(cost, 0) ✓ translated</p><p><strong>2. A month bucket.</strong> Also everywhere.</p><p>DATETRUNC(‘month’, [Order Date]) → date_trunc(‘month’, order_date) ✓ translated</p><p><strong>3. Days to ship.</strong> Converts, and asks for a glance.</p><p>DATEDIFF(‘day’, [Order Date], [Ship Date]) → date_diff(‘day’, order_date, ship_date) ⚠ translated — verify</p><p>Flagged because interval functions disagree at boundaries: whether an order placed at 23:50 and shipped at 00:10 is zero days or one. Usually it does not matter. At quarter ends, on an SLA report, it does.</p><p><strong>4. Running total.</strong> A table calculation with a direction-independent meaning.</p><p>WINDOW_SUM(SUM([Sales])) → declared as running_total, emitted as a window pass outside the GROUP BY ✓ translated</p><p><strong>5. Regional average, held constant.</strong> A FIXED level of detail.</p><p>{ FIXED [Region] : AVG([Order Value]) } → a derived table aggregated by region, joined back at that grain — which is what keeps it constant however the view is grouped ✓ translated <strong>6. The one that stops.</strong> IF ISMEMBEROF(‘Regional Managers’) THEN [Region] = USERNAME() ELSE TRUE END → not translated ✗ reported</p><p>This is row-level security written as a formula. It could be translated and deliberately is not — see below.</p><p>Illustrative composite. The figures are internally consistent and every step is one the product supports today.</p><h3>What translates cleanly</h3><p>The deterministic tier is a rewrite from Tableau’s function set to the engine’s. It is unglamorous, testable code rather than a model guessing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*aFfutW7IDPjM8u2I.png" /></figure><h3>Table calculations: translated, not approximated</h3><p>A running total or a percent-of-total is a window over aggregated rows, not an aggregate. These translate into a declared calculation that the compiler emits as a window pass <em>outside</em> the grouping — which is what makes the number right.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*5DqykeoYa_r_JruL.png" /></figure><h4>Why addressing is the dividing line</h4><p>A table calculation in Tableau is computed along a direction through the visible table — across then down, or restarting at each pane. Some calculations have a meaning independent of that direction and some do not. The ones that do translate; the ones that are <em>defined</em> by the addressing are reported, because guessing which direction the author meant is exactly the kind of thing that produces a plausible wrong number.</p><h3>Level-of-detail expressions</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*o3oE6nx7o9_V8bfr.png" /></figure><p>This is the honest boundary, and it is a property of the semantics rather than a limitation of effort. An INCLUDE expression means something different on each sheet it appears on. A model-level definition cannot encode “whatever is on the sheet”.</p><h3>The category we refuse on purpose</h3><p>One group of functions is not translated even though it technically could be, and this is a design decision worth explaining.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*aM5kpehdS6zvdSgb.png" /></figure><h4>Security should not live in a formula</h4><p>USERNAME() inside a calculated field is how a great many Tableau estates implement row-level security, and it has a specific failure mode: the rule is invisible unless you open the workbook, and it only applies to the sheets that happen to use that field. Translating it would faithfully reproduce that weakness. Instead the importer reports it, and you write a row policy on the datasource - which applies to every dashboard, exploration, export and AI answer, including ones built years later. <a href="https://insightaiviz.com/blog-role-steward.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-calculation-by-calculation">How that is set up →</a></p><p>This is the one place the importer deliberately makes you do work, and it is the place where doing the work has the largest payoff.</p><h3>What the report actually gives you</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*yWWbG7j8Ga341GRw.jpg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Pz3bhkhcXajlkm1A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*9rwWynkELNNEFkdp.png" /></figure><h3>Working the list</h3><ol><li><strong>Sort by how often the field is used, not by difficulty.</strong> A manual calculation on a sheet nobody opens is not work. Cross-reference with your usage data.<a href="https://insightaiviz.com/blog-shelfware.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-calculation-by-calculation">Which is probably most of the estate →</a></li><li><strong>Do the verify tier in one sitting.</strong> Each is a few minutes and they are mostly date boundaries. Doing them in a batch is far faster than one at a time.</li><li><strong>Turn every </strong><strong>USERNAME() into a policy.</strong> This is the highest-value item on the list every time, and it usually reveals that the old rule was applied less consistently than anyone believed.</li><li><strong>Reconcile with an exported crosstab.</strong> Not by reading the SQL. Export from the original, compare cell for cell, and investigate every difference.</li></ol><h3>What we have not built</h3><ul><li><strong>No language-model fallback tier.</strong> We could offer one for the manual set and have chosen not to yet, because a plausible translation of business logic is worse than an honest gap — and the reader cannot tell them apart.</li><li><strong>No automated parity check.</strong> Comparing against an exported crosstab is currently your job. It should be the product’s, and it is the thing we would build next in this area.</li><li><strong>Power BI DAX is not translated at all.</strong> Different language, different model, and the model layer needs the XMLA endpoint.</li></ul><p><a href="https://insightaiviz.com/blog-tableau-migration.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-calculation-by-calculation">What else comes across →</a> · <a href="https://insightaiviz.com/blog-running-both-estates.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-calculation-by-calculation">The quarter you run both →</a></p><h3>About — InsightViz</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*0duaOqRWVqWrtolT.jpg" /></figure><p><strong>InsightViz</strong> is AI-native business intelligence where the dashboard explains itself — press Why on any chart and it works out what drove the change from your own rows. Part of the <a href="https://www.insightlake.com/">InsightLake</a> enterprise AI suite by VerticalServe.</p><p><strong>VerticalServe Inc</strong> — Niche Cloud, Data &amp; AI/ML Premier Consulting Company, Partnered with Google Cloud, Confluent, AWS, Azure…50+ Customers and many success stories..</p><p><strong>Product:</strong> <a href="https://insightaiviz.com/?utm_source=medium&amp;utm_medium=syndication">https://insightaiviz.com</a></p><p><strong>Website:</strong> <a href="http://www.verticalserve.com/">http://www.VerticalServe.com</a></p><p><strong>Contact:</strong> contact@verticalserve.com</p><p><em>Originally published at </em><a href="https://insightaiviz.com/medium/blog-calculation-by-calculation.html?v=2"><em>https://insightaiviz.com</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d46694ed63a8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Most of Your BI Bill Pays for People Who Never Build Anything]]></title>
            <link>https://verticalserve.medium.com/most-of-your-bi-bill-pays-for-people-who-never-build-anything-07ac061e5d5e?source=rss-79851bb07a23------2</link>
            <guid isPermaLink="false">https://medium.com/p/07ac061e5d5e</guid>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[llm]]></category>
            <dc:creator><![CDATA[VerticalServe Blogs]]></dc:creator>
            <pubDate>Wed, 14 Oct 2026 21:27:38 GMT</pubDate>
            <atom:updated>2026-10-14T21:27:38.784Z</atom:updated>
            <content:encoded><![CDATA[<h4>The viewer tax, and why rationing access costs more than the invoice does.</h4><p><em>Most of a BI licence bill is people who never build anything. They open a dashboard, change a filter, read a number and close it. They are priced like authors, and that single fact quietly decides how much of your organisation gets to see its own data.</em></p><p>Pull your BI licence list and sort it by what people actually do. In most estates the shape is the same: a small group who build, a slightly larger group who edit occasionally, and a long tail — often eighty per cent or more — who have never authored anything. They open a dashboard someone else built, change a filter, read a number and close it.</p><p>That long tail is the bill. And because it is priced per head, it is also the thing your organisation quietly optimises against.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*XPFg_qiZw10yRda3.png" /></figure><p><em>A worked example</em></p><h4>A 2,400-person manufacturer, at renewal</h4><p>One BI platform, nine years in. The renewal arrived as a single figure with a 4% uplift and a three-year option. Finance asked for the line items before signing. Rates are illustrative — yours will differ by contract — but the shape does not.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Tv-XR-TcH02qZRHZ.png" /></figure><p>Illustrative composite. The figures are internally consistent and every step is one the product supports today.</p><p>Three things fell out of that table within an hour, none of which the total had shown.</p><ul><li><strong>Viewers are 52% of the bill</strong> and 84% of the seats. They author nothing.</li><li><strong>478 viewer seats had not logged in for a quarter</strong> \u2014 £68,832 a year for people who did not open anything.</li><li><strong>Reserved capacity was sized against a 2021 peak</strong> that had not recurred, and nobody owned the line.</li></ul><p>The negotiated discount on offer was 6%, or about £18,000. The unused viewer seats alone were nearly four times that, and they were invisible until somebody asked for the breakdown.</p><h3>Four costs, only one of which is on the invoice</h3><p>Finance sees the renewal. The other three never appear on a budget line, which is exactly why they persist.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*0u6QFlOImYbeqXBF.png" /></figure><p>The fourth one deserves a moment. Every organisation that rations BI access grows a parallel reporting estate in spreadsheets, and that estate has no row-level security, no audit trail, no refresh schedule and no owner. It is the direct output of a pricing model, and it is usually discovered during an audit rather than during a budget review.</p><h4>The second-order cost is worse than the first</h4><p>An organisation that rations data access gets worse at making decisions, in a way that never appears on a budget line. The invoice is visible. The meeting that went the wrong way because the number arrived as a three-week-old screenshot is not.</p><p><em>A worked example</em></p><h4>What the rationing actually cost, at the same manufacturer</h4><p>With viewer seats capped, access was allocated by the plant managers once a year. Four shift supervisors per plant, across eleven plants, did not have one.</p><p><strong>One plant, one week</strong></p><p><strong>The workaround — A morning email.</strong> A supervisor with a seat exported the overnight yield dashboard to PDF at 06:40 and emailed it to the four who did not. By the 14:00 shift handover it was eight hours old.</p><p><strong>The consequence — A decision on stale numbers.</strong> A line running out of tolerance was visible on the dashboard at 11:20 and in the supervisors’ inboxes the following morning. One shift of scrap.</p><p><strong>The tell — Nobody logged it as a BI problem.</strong> It was logged as a process issue. The connection between a £12 seat and a shift of scrap was never made, because those two numbers live in different systems and different budgets.</p><p>Illustrative composite. The figures are internally consistent and every step is one the product supports today.</p><h3>Why it is priced this way</h3><p>Not malice — architecture. When the vendor hosts your data and runs the engine, every viewer is a real cost to them: storage, compute, a session held open. Per-seat pricing is a rational way to recover that, and it worked when BI meant a few hundred analysts inside a reporting function.</p><p>It stops working the moment the honest answer to <em>who should see this?</em> is <em>everyone who could act on it</em> — which, for an operational dashboard, it usually is. The warehouse manager who could fix the delivery problem is not in the reporting function and does not have a seat.</p><blockquote><em>The pricing model is not charging you for value. It is charging you for the vendor’s hosting costs, in a unit — the human being — that has nothing to do with them.</em></blockquote><h3>What changes when the data is already yours</h3><p>InsightViz keeps a governed copy of your data in your own cloud storage, in an open format, and the query engine is part of the product. There is no vendor hosting your data and no per-session cost for anyone to recover, so there is no reason to charge for someone who only reads.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*0RI_nM7uIX9cMfm9.png" /></figure><p>The storage is the boring part and the important part. It is Parquet on your own object store — S3, Azure Blob or Google Cloud Storage, in your account, under your keys. Any tool that reads Parquet can read it, including the next BI tool you buy. That is the difference between a copy of your data and a hostage.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*b0r8C1bv5EmPvLpa.jpg" /></figure><h3>What the people who only read actually get</h3><p>Removing the seat charge only matters if the tool is usable by someone who was never trained on it. Four things carry most of that weight:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*c-N8e4fGjlwHokd4.png" /></figure><p>Note what the last row replaces. Most of a BI ticket queue by <em>count</em> is small questions — break this down by segment, compare it with last quarter — and those are the ones that make analysts feel like a service desk. They stop being tickets.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*faVyIOFftq7zQC7Q.jpg" /></figure><h3>The honest limits</h3><p>Three, because a pricing argument that admits nothing is a pricing argument nobody believes.</p><ul><li><strong>Someone still pays for the compute.</strong> It moves from a licence line to your cloud bill. For most estates it is dramatically cheaper — a dashboard reads aggregated Parquet, not the warehouse — but it is not free, and it is a line you now own.</li><li><strong>You need somewhere to run it.</strong> An AWS, Azure or GCP account, and someone who can deploy into it. If you have no cloud footprint, the hosted option exists, but then you are back to paying someone for hosting — just not per reader.</li><li><strong>Governance becomes your job.</strong> When everyone can see the data, who sees which rows matters more, not less. The rules are set once on the datasource rather than per workbook, which is the better design, but the first set you write will be wrong and will need revisiting.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Z4ZhQhBm4NJ8Vtfo.jpg" /></figure><h3>The question to take to your renewal</h3><p>Not “can we negotiate the rate?” — you can, by a few per cent, once a year. The question is:</p><blockquote><em>How many people in this organisation would be better at their job if they could see this, and what are we paying to stop them?</em></blockquote><p>If the answer to the second half is most of the invoice, that is worth an afternoon. <a href="https://insightaiviz.com/why-switch.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-viewer-tax">The full business case →</a></p><h3>About — InsightViz</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*3XkX02VJanhRTD1H.jpg" /></figure><p><strong>InsightViz</strong> is AI-native business intelligence where the dashboard explains itself — press Why on any chart and it works out what drove the change from your own rows. Part of the <a href="https://www.insightlake.com/">InsightLake</a> enterprise AI suite by VerticalServe.</p><p><strong>VerticalServe Inc</strong> — Niche Cloud, Data &amp; AI/ML Premier Consulting Company, Partnered with Google Cloud, Confluent, AWS, Azure…50+ Customers and many success stories..</p><p><strong>Product:</strong> <a href="https://insightaiviz.com/?utm_source=medium&amp;utm_medium=syndication">https://insightaiviz.com</a></p><p><strong>Website:</strong> <a href="http://www.verticalserve.com/">http://www.VerticalServe.com</a></p><p><strong>Contact:</strong> contact@verticalserve.com</p><p><em>Originally published at </em><a href="https://insightaiviz.com/medium/blog-viewer-tax.html?v=1"><em>https://insightaiviz.com</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=07ac061e5d5e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Should You Build Your Own AI BI Layer? Three Times the Answer Is Yes]]></title>
            <link>https://verticalserve.medium.com/should-you-build-your-own-ai-bi-layer-three-times-the-answer-is-yes-66339665bf38?source=rss-79851bb07a23------2</link>
            <guid isPermaLink="false">https://medium.com/p/66339665bf38</guid>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[data-analysis]]></category>
            <dc:creator><![CDATA[VerticalServe Blogs]]></dc:creator>
            <pubDate>Wed, 14 Oct 2026 19:06:25 GMT</pubDate>
            <atom:updated>2026-10-14T19:06:25.204Z</atom:updated>
            <content:encoded><![CDATA[<h4>And the question that actually decides it, which is neither cost nor capability.</h4><p>Any vendor writing this post has an answer before they start, so here is ours up front: sometimes you should build it, and we would rather say which cases than pretend there are none. A pitch that claims buying is always right is a pitch nobody senior believes.</p><p>The build case for BI has also got genuinely stronger. DuckDB, Parquet, Arrow and a competent model make a semantic layer and a dashboard renderer a plausible quarter’s work for a good team. That is new, and it deserves a straight answer rather than a reflex.</p><h3>Three cases where building is right</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*8VrIUGX2MHxLtPl-.png" /></figure><p>Notice what is not on that list: cost. “We can build it cheaper” is almost always an estimate of version one against a mature product, which is not the comparison you will live with.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*882xe2s-dFiTJAnb.png" /></figure><h3>The question that actually decides it</h3><blockquote><em>Not “can we build this?” — you probably can. It is “who owns it in year two, when the person who built it has changed team?”</em></blockquote><p>This is the failure mode nobody prices in, and it is specific rather than vague. An internal BI layer built by one strong engineer recreates the key-person dependency it was meant to remove — one level down, and now in code nobody else has read.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*O63tGK_b-6mx3JbG.png" /></figure><p>Ask it concretely before you commit: name the two people who will own this in eighteen months. If you cannot, or if the honest answer is “whoever is free”, that is the decision. Not a preference — the decision.</p><p><em>A worked example</em></p><h4>The estimate, and what it became</h4><p>A platform team at a fintech scoped their own BI layer. Two strong engineers, a quarter. The estimate was not unreasonable — version one shipped roughly on time. This is the estimate beside what the first eighteen months actually cost.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*qDZ19s41YYcs4isX.png" /></figure><p>Illustrative composite. The figures are internally consistent and every step is one the product supports today.</p><h4>The incident in row three</h4><p>Nine months in, a panel’s cached result was keyed on the query text alone. Two users with different row entitlements shared a cache entry, and one of them saw the other’s region for about forty minutes before someone noticed. Nothing was malicious and nothing left the building — but it was reportable, and the four weeks of rework was the cheapest part of it.</p><p>Version one was on time. Version three — with security that holds everywhere, scheduling that recovers, provenance a compliance team accepts, and a way off the old tool — was four times the estimate. Version three is what you are comparing against a product.</p><h3>What people underestimate in a BI build</h3><p>The rendering is the fun part and it is not the work. Four things consume a build, and three of them are invisible from the outside.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*9TSEDSZAskLU7nUP.png" /></figure><h4>The one that catches everyone</h4><p>Caching and security are the same problem, and most builds discover this after go-live. If a panel’s result is cached on the query alone, two people with different row access share a cache entry and one of them sees the other’s numbers. The key has to include the policy — and the cleanest way to guarantee that is for the policy to already be inside the SQL the key is computed from.</p><p>That is a design decision you make once, early, and it is very unpleasant to retrofit.</p><h3>The honest comparison</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*pvQwAlb2sZYii2ZJ.png" /></figure><h3>The middle path most estates should take</h3><p>The interesting option is rarely build-everything or buy-everything. It is: buy the governed layer, build the two or three things that are genuinely yours.</p><ol><li><strong>The definitions are yours either way.</strong> A model is a readable, versioned document. That is the asset, and no product should hold it hostage.</li><li><strong>The data is yours either way.</strong> Parquet in your own bucket. If you removed the product, every file stays readable.</li><li><strong>Build the embedded surface if analytics is your product.</strong> Query through the API, render it yourself, keep your own look and URL structure.</li><li><strong>Do not build the compiler or the policy engine.</strong> These are where the subtle, expensive, security-relevant bugs live, and they are entirely undifferentiated. Nobody has ever won a customer because of their ownGROUP BY generator.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*2W3bujmlIE5HlDuw.jpg" /></figure><h3>If you have already built it</h3><p>Most organisations reading this have something — a Streamlit app, a set of scheduled notebooks, a dbt layer with a thin viewer on top. That is not wasted, and it is usually the most valuable input available.</p><p>The dbt models are your definitions, already argued over. The scheduled queries tell you the real refresh cadences. The notebook nobody can retire tells you which questions actually matter. Bring all three to the evaluation rather than starting from a blank requirements document.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*mGoCRB5StYelWkZe.png" /></figure><h3>The test to run before you decide</h3><ol><li><strong>Write down the two owners for year two.</strong> If you cannot name them, buy.</li><li><strong>Ask your security lead to describe how row security survives caching.</strong> If the build team has not thought about it, that is the size of the gap.</li><li><strong>Count the dashboards you would have to migrate.</strong> Build gives you no importer. Every one of them is rebuilt by hand, by someone, on a deadline.</li><li><strong>Price version three, not version one.</strong> Version one is a quarter. Version three — with scheduling, audit, export, provenance and migration — is the product you are actually comparing against.</li></ol><p><a href="https://insightaiviz.com/blog-first-ninety-days.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-build-vs-buy">If you buy: the first ninety days →</a> · <a href="https://insightaiviz.com/blog-measuring-it.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-build-vs-buy">How to tell whether it worked →</a></p><h3>About — InsightViz</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*s79QTWOFL0lEst10.jpg" /></figure><p><strong>InsightViz</strong> is AI-native business intelligence where the dashboard explains itself — press Why on any chart and it works out what drove the change from your own rows. Part of the <a href="https://www.insightlake.com/">InsightLake</a> enterprise AI suite by VerticalServe.</p><p><strong>VerticalServe Inc</strong> — Niche Cloud, Data &amp; AI/ML Premier Consulting Company, Partnered with Google Cloud, Confluent, AWS, Azure…50+ Customers and many success stories..</p><p><strong>Product:</strong> <a href="https://insightaiviz.com/?utm_source=medium&amp;utm_medium=syndication">https://insightaiviz.com</a></p><p><strong>Website:</strong> <a href="http://www.verticalserve.com/">http://www.VerticalServe.com</a></p><p><strong>Contact:</strong> contact@verticalserve.com</p><p><em>Originally published at </em><a href="https://insightaiviz.com/medium/blog-build-vs-buy.html?v=2"><em>https://insightaiviz.com</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=66339665bf38" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Screenshot a Dashboard. Get It Rebuilt on Your Own Data.]]></title>
            <link>https://verticalserve.medium.com/screenshot-a-dashboard-get-it-rebuilt-on-your-own-data-52ef5ffeea0a?source=rss-79851bb07a23------2</link>
            <guid isPermaLink="false">https://medium.com/p/52ef5ffeea0a</guid>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[business-intelligence]]></category>
            <dc:creator><![CDATA[VerticalServe Blogs]]></dc:creator>
            <pubDate>Tue, 13 Oct 2026 12:28:58 GMT</pubDate>
            <atom:updated>2026-10-13T12:28:58.286Z</atom:updated>
            <content:encoded><![CDATA[<h4>How AI reads a picture of a Power BI or Tableau page, and what it refuses to pretend.</h4><p>Every migration has a long tail of pages nobody can export. A Power BI report from a team that left. A slide in the board pack that someone rebuilds by hand each quarter. A photo of a whiteboard from the planning offsite. There is no file to import, but there is a picture, and the picture says exactly which panels the business decided it needed.</p><p>The InsightViz builder has four ways in: describe a dashboard in words, design it panel by panel, import a Tableau workbook — and drop a picture of one. This post is about the last, because it is the one people assume is a trick, and the reason it is not is the same reason the rest of the product can be trusted.</p><h3>What happens to the picture</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*4R5GDMLfzfEYL3iK.png" /></figure><p>The model that reads the image is asked to do one thing: say what each panel <em>is</em> — its title, its chart type, what it appears to measure and by what, roughly where it sits. It is given your model’s catalogue of dimensions and metrics and told to express every panel in those terms and no others. Then the plan goes through exactly the checks a dashboard composed from a sentence goes through: every field it names is looked up, every panel is compiled, and a panel that names nothing your data has is left out and listed.</p><p>The numbers in the picture are never used. A KPI tile that says <em>$998.4K</em> becomes a panel over your Total Sales metric; whatever your data says is what it shows.</p><h3>A worked example</h3><p>Here is a retail dashboard as a flat image — four headline numbers, sales by region, a monthly line, profit by category and a returns table — dropped onto the builder over the sample Superstore model that ships with the product:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*0VmjKcguxiVpaFzE.jpg" /></figure><p>The sample model has sales, profit, quantity, region, category, segment and dates. It has no returns. The picture asks for returns twice. This is what came back, word for word from the builder:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*M1tLUpG9-SyifKbd.png" /></figure><p>Two of those substitutions are ones you would not keep. That is fine. The point is that you can see them before you press save, against the panel they apply to, in plain words — so the decision to delete the panel or to add a returns table to the model is yours, made in thirty seconds, rather than discovered in a meeting when someone asks why the return rate is twelve percent.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*oNKVX9vfsc7TyS5J.png" /></figure><h3>Where it fits in a migration</h3><p>A Tableau workbook should still be imported as a file: the file carries the calculations, the filters and the click-to-filter behaviour, and the import reports on each one. A picture carries none of that. What a picture is good for is everything that is not a Tableau file:</p><ul><li><strong>Power BI pages.</strong> Importing Power BI files is on the roadmap. A screenshot of the page rebuilds its panels over your model today.</li><li><strong>Slides and spreadsheets.</strong> The board-pack chart someone rebuilds by hand each quarter becomes a dashboard panel that refreshes on its own.</li><li><strong>Sketches.</strong> A whiteboard photo from a planning session is a perfectly good brief. The layout is approximate, and the panels are real.</li></ul><h4>What a picture cannot tell you</h4><p>A screenshot shows what a panel looked like, not how it was computed. Filters applied off-screen, a calculation behind a tile, a cross-filter click or a drill path — none of it is in the pixels, so none of it is carried. Colours and exact sizes are approximate. The reading is a model’s reading of an image, so check each panel against the original before you trust it; the builder says that at the top of every list. And the numbers will differ from the picture whenever your data differs from the data the picture was drawn from, which is the point.</p><h3>Then it is a dashboard like any other</h3><p>Nothing is saved until you press save. Once it is, the dashboard is a document like every other: edit any panel by hand, publish it with a note, restore an earlier version, press <strong>Why?</strong> on any chart, keep the explanation as an insight, and have the page mailed to the team on Monday. The picture was only ever a faster way of saying what you wanted.</p><p>Have a page you cannot export? <a href="https://insightaiviz.com/contact.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-rebuild-from-a-picture">Send us a screenshot</a> and we will rebuild it on the call, over a model of your own data.</p><h3>About — InsightViz</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*SKgS3tfh5fYhJ1re.jpg" /></figure><p><strong>InsightViz</strong> is AI-native business intelligence where the dashboard explains itself — press Why on any chart and it works out what drove the change from your own rows. Part of the <a href="https://www.insightlake.com/">InsightLake</a> enterprise AI suite by VerticalServe.</p><p><strong>VerticalServe Inc</strong> — Niche Cloud, Data &amp; AI/ML Premier Consulting Company, Partnered with Google Cloud, Confluent, AWS, Azure…50+ Customers and many success stories..</p><p><strong>Product:</strong> <a href="https://insightaiviz.com/?utm_source=medium&amp;utm_medium=syndication">https://insightaiviz.com</a></p><p><strong>Website:</strong> <a href="http://www.verticalserve.com/">http://www.VerticalServe.com</a></p><p><strong>Contact:</strong> contact@verticalserve.com</p><p><em>Originally published at </em><a href="https://insightaiviz.com/medium/blog-rebuild-from-a-picture.html?v=2"><em>https://insightaiviz.com</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=52ef5ffeea0a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Your Monday Dashboard Should Explain Itself Before the Meeting]]></title>
            <link>https://verticalserve.medium.com/your-monday-dashboard-should-explain-itself-before-the-meeting-0c87182d2568?source=rss-79851bb07a23------2</link>
            <guid isPermaLink="false">https://medium.com/p/0c87182d2568</guid>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[llm]]></category>
            <dc:creator><![CDATA[VerticalServe Blogs]]></dc:creator>
            <pubDate>Tue, 13 Oct 2026 00:38:14 GMT</pubDate>
            <atom:updated>2026-10-13T00:38:14.467Z</atom:updated>
            <content:encoded><![CDATA[<h4>AI-written weekly stories where every figure is checked against the data — or the story doesn’t get sent.</h4><p>Monday, 8:40. The leadership meeting is at nine. The dashboard was refreshed overnight and is exactly as clear as it was last week: a bar chart, a trend line, a ratio by category. Somewhere in it is the one thing that moved. Finding it, working out why, and deciding whether anyone needs to act is the twenty minutes nobody has before nine — so it happens in the meeting, out loud, from memory.</p><p>Most BI tools answer this with a scheduled PDF. The PDF arrives on time and says nothing the dashboard did not; it is the same chart, in an attachment. This post is about the three things InsightViz sends instead, and the rule that decides what is allowed into the email.</p><h3>From one answer to a story</h3><p>Press <strong>Why?</strong> on a chart and InsightViz works out what drove the change — the two periods, the breakdown by every business dimension, the member that carried it — from the rows, under the reader’s own access rules. That answers one question about one chart, once. At the foot of the explanation is a button that keeps the question: <strong>Make this an insight</strong>.</p><p>An insight is a short plan of parts. Each part is a governed query through the same compiler as every dashboard panel; a run turns the plan into a story — a headline, the figures that matter as tiles, each part’s chart and sentences, a synthesis, and what to do with an owner against each action. This is the one made from the sample dashboard’s sales panel:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*wQ3lt5Bn18o8rlNm.jpg" /></figure><p>There are four ways to get one, and they end in the same place:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*0Yiz6Ix1R_ogZuoR.png" /></figure><h3>Every figure is checked, and the check is on the page</h3><p>A story is sentences, and sentences are where a language model is most tempted to round, combine or invent. So after every run, every number in every sentence is matched back to the rows the parts returned: quoted directly from a cell, re-derived by one step of arithmetic between two cells, or found somewhere in the table other than where the sentence puts it. The mark beside each claim says which. Click one and the sheet behind it opens:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*tcv54ENVF0H4MPj7.jpg" /></figure><p>Two things follow from that check, and they are the rest of this post. A story with figures that did not trace cannot be published without someone saying so on the record. And only a published story ever leaves the building.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*tVFu-XUHsm8kcQyX.png" /></figure><h3>On the page the team already opens</h3><p>A published insight can sit on the dashboard grid like any other panel. It re-runs at the page’s filters: choose East and the story is about East, with East’s figures traced against East’s rows. A slice that was already worked out is served from cache and says when it was computed, because an old number that looks current is worse than a blank.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*gjwjKWNZRXsxsr1c.jpg" /></figure><h3>Then make it arrive</h3><p><strong>Watch</strong> on the dashboard holds the two things that turn a page into something that comes to you. An <strong>alert</strong> checks one panel’s figure on a schedule and mails its recipients when it crosses a line — and every <strong>Why?</strong> proposes one, with a threshold taken from the series’ own variance rather than a round number someone guessed. A <strong>subscription</strong> mails the dashboard as a PDF or an image on a schedule, with the filters it was made with.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ne83aiuP2PlYzIeq.jpg" /></figure><p>What makes the subscription different is the body of the email. Above the attachment it carries <strong>What changed</strong> — the panels that moved most since the previous period, with the driver under each — and, when the dashboard’s weekly insight is published, its headline, its summary and what to do. The person reading on a phone at 8:40 has the two lines that matter and the reason before they open anything.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*DOzO7RiQK0N6APUk.png" /></figure><h4>Why a draft never goes out</h4><p>Publishing is where a story with untraced figures is refused. If the email carried the latest run regardless, the check would be a page in the console that a scheduled job walks straight past. An email is a copy of the data that leaves the building and is forwarded without its context; it gets the stricter rule, not the looser one. A dashboard with a weekly insight still in draft sends its PDF and What changed, and no story.</p><p><em>A worked example</em></p><h4>Monday, 8:40</h4><p>The regional director from <a href="https://insightaiviz.com/blog-why-did-sales-drop.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-before-the-meeting">the first post in this series</a> set this up after the bad August. What she reads on her phone the following Monday:</p><p><strong>Mail — “Executive Overview — this week”.</strong> What changed first: Sales by Region down, the High band carrying most of it; Profit Ratio up, Office Supplies the largest change. Two lines, the driver under each.</p><p><strong>Story — The weekly insight’s headline.</strong> Underneath, the published story’s headline and summary, and the two actions with their owners — the East Technology pipeline, the Corporate order volume.</p><p><strong>Tap — Open the dashboard.</strong> The insight panel on the page is the same story at her filters. She taps the East figure; the claim sheet shows the row it came from.</p><p><strong>Alert — Friday, the week after.</strong> Sales by Region falls below the threshold the explanation proposed. The alert arrives before the next Monday, not during it.</p><p>Illustrative composite. The figures are internally consistent and every step is one the product supports today.</p><p>None of that needed an analyst on Monday morning, and none of it said anything the rows do not support. The analyst’s afternoon went into the question the story raised and could not answer — why Corporate ordered less — which is the afternoon worth having.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*VpckO2o-9PulAV6n.png" /></figure><h3>What it does not do</h3><ul><li><strong>It does not know the world.</strong> A story explains what the rows show. Why a customer ordered less is a question for a person; the story says that, and names what would answer it.</li><li><strong>It is only as good as its traced share.</strong> The share of figures that traced is printed under every story. A story at sixty percent is a draft worth reading and not one worth sending, which is the point of the publish check.</li><li><strong>It compares periods the way the data is grouped.</strong> A brief that asks about “early versus late 2025” is answered from monthly rows unless the plan groups them. The planner is told to group them, but a story can still lean on single months — read the question and the grain printed on each part.</li><li><strong>Schedules run in UTC.</strong> Pick the hour accordingly; the panel says which time zone it means.</li></ul><p>If your Monday starts with someone screenshotting a chart into a chat and asking “anyone know why?”, that is the meeting this was built to get ahead of. <a href="https://insightaiviz.com/contact.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-before-the-meeting">Bring one of your dashboards</a> and we will set the weekly story up on it.</p><h3>About — InsightViz</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*NhvGVgP_ODzdVYXt.jpg" /></figure><p><strong>InsightViz</strong> is AI-native business intelligence where the dashboard explains itself — press Why on any chart and it works out what drove the change from your own rows. Part of the <a href="https://www.insightlake.com/">InsightLake</a> enterprise AI suite by VerticalServe.</p><p><strong>VerticalServe Inc</strong> — Niche Cloud, Data &amp; AI/ML Premier Consulting Company, Partnered with Google Cloud, Confluent, AWS, Azure…50+ Customers and many success stories..</p><p><strong>Product:</strong> <a href="https://insightaiviz.com/?utm_source=medium&amp;utm_medium=syndication">https://insightaiviz.com</a></p><p><strong>Website:</strong> <a href="http://www.verticalserve.com/">http://www.VerticalServe.com</a></p><p><strong>Contact:</strong> contact@verticalserve.com</p><p><em>Originally published at </em><a href="https://insightaiviz.com/medium/blog-before-the-meeting.html?v=2"><em>https://insightaiviz.com</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0c87182d2568" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Every BI Tool Has an AI Chat Box Now. Most Shouldn’t Be Trusted.]]></title>
            <link>https://verticalserve.medium.com/every-bi-tool-has-an-ai-chat-box-now-most-shouldnt-be-trusted-51a6d986d1a1?source=rss-79851bb07a23------2</link>
            <guid isPermaLink="false">https://medium.com/p/51a6d986d1a1</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[data-analytics]]></category>
            <dc:creator><![CDATA[VerticalServe Blogs]]></dc:creator>
            <pubDate>Sun, 11 Oct 2026 21:56:11 GMT</pubDate>
            <atom:updated>2026-10-11T21:56:11.995Z</atom:updated>
            <content:encoded><![CDATA[<h4>What separates an assistant you can put in front of a CFO from a demo that works on easy questions.</h4><p><em>Every BI vendor now has a chat box. Most of them are a demo: impressive on clean data and easy questions, unreliable the moment the question gets specific. The difference is not the model — it is what the model is allowed to touch, and what it does when it does not know.</em></p><p>A manager in a Tuesday review asks: <em>which region has the weakest profit ratio, and how has it trended this year?</em> The dashboard on screen does not break it down that way. In most organisations that question becomes a message to an analyst and an answer on Thursday, by which point the meeting has moved on and the decision has been made without it.</p><p>Here is that exact question, asked and answered in the room:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*D7qG7vQ2GepbfsM9.jpg" /></figure><p>The region was the easy part. The valuable part is the shape of the year — that the weakness is two bad months rather than a slow decline, which points at discounting or a handful of large low-margin orders rather than at a structural problem. Getting that on Thursday is worth much less than getting it while the room is still discussing it.</p><p><em>A worked example</em></p><h4>The rest of that meeting</h4><p>The first answer named the West and the two months behind it. What happened over the next six minutes is the part a single-question demo never shows — because the value is in the follow-ups, and follow-ups are where a chat box on raw data comes apart.</p><p><strong>Six minutes, five questions</strong></p><p><strong>Q2 — “Was February a volume problem or a margin problem?”.</strong> Units sold held; average selling price fell 11%. So margin, not volume — which rules out the demand explanation somebody had already offered.</p><p><strong>Q3 — “Which category drove it?”.</strong> Furniture, down 340 basis points against a category average of 40. The other two categories barely moved.</p><p><strong>Q4 — “Is that seasonal? Show me February the year before.”.</strong> The prior February shows no such fall. It is not seasonal, which was the last comfortable explanation available.</p><p><strong>Q5 — “Break furniture down by customer segment”.</strong> Almost all of it sits in one segment — corporate — which points at contract pricing rather than at anything the stores did.</p><p><strong>The outcome — A specific question for a specific person.</strong> The meeting ends with “who approved the corporate furniture discounts in February?” rather than “can someone look into the West?”. One of those gets answered.</p><p>Illustrative composite. The figures are internally consistent and every step is one the product supports today.</p><p>Each of those five would have been a separate ticket. More importantly, questions three, four and five only exist <em>because</em> the previous answer arrived while the room was still thinking. A three-day turnaround does not produce that chain at all — it produces one question, answered after the meeting, and no chain.</p><h3>Why most BI chat boxes cannot be trusted</h3><p>The common design hands the model a database connection and asks it to write SQL. It fails in three predictable ways, and every one of them is quiet.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*hwyQ7xJCfPNiBz0o.png" /></figure><p>None of these announce themselves. They do not error. They arrive as a well-formatted number in a meeting, and they are discovered weeks later, if at all.</p><h4>Why prompting does not fix this</h4><p>The usual response is a longer system prompt — “always use the revenue_net column, never join orders to line_items without…”. That is a rule written in prose and enforced by hope. It works most of the time, which is the worst possible failure rate for a number somebody is going to quote to a board.</p><h3>The fix is a smaller surface, not a bigger model</h3><p>The assistant in InsightViz is never given raw data and never given a database connection. It is given your business definitions — the metrics and dimensions your teams agreed on, with plain-English names, descriptions and the other words people use for them — and its only job is to choose from that list. InsightViz turns the choice into the query.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*IG2-3DYNoM3W5onh.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*6MYnkXIkAHuKT4dZ.png" /></figure><p>Each failure mode above is closed by construction rather than by instruction:</p><ul><li>The joins are declared once in the model, so the assistant never picks one. There is no prompt to get wrong.</li><li><em>Net revenue</em> is defined once and used by every dashboard <em>and</em> every answer, so the two cannot disagree. If the definition is wrong, it is wrong everywhere at once — which is a fixable problem rather than a reconciliation problem.</li><li>A field that is not in the catalogue cannot be selected, so an invented one is refused and named. The answer becomes “this model has no customer tier” instead of a number.</li></ul><h4>The access rules are the same rules</h4><p>Row and column access is applied when the query is built, not filtered out of the results afterwards. A regional manager asking “which region is weakest” gets an answer about their own region, because the restriction is inside the query. A column someone may not see is not hidden from the chart — it is absent from the vocabulary, so it cannot be charted, exported, or asked about at all.</p><h3>Every answer carries its working</h3><p>An assistant that is right most of the time is only useful if a reader can check it without asking the person who built it. Every answer, and every panel on every dashboard, will show the question that was asked, the query that ran, and which access rules applied.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*2COPviOXk_pQe7p5.jpg" /></figure><p>This is the feature that decides whether an organisation actually adopts an AI answer or quietly keeps asking the analyst to double-check it. When someone challenges a figure in a meeting, the answer is on screen rather than in somebody’s head.</p><h3>The question you no longer have to type</h3><p>Since this post was first written the assistant has moved onto the chart. The Tuesday question above — <em>which region has the weakest profit ratio, and why</em> — no longer needs to be typed at all. Every chart and headline number carries a <strong>Why?</strong> button; press it and the change is worked out from the rows before a model is asked anything: the two periods, the contribution of every region, category and segment, the member that moved the other way, and where the period sits in the trend.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*PRrIiuU2N9bh6fmU.jpg" /></figure><p>The conversation is still there — <em>Ask about this</em> sits under the explanation — but its evidence is now the panel: the claims, the trend, the breakdowns, and at most one further governed query when those cannot answer. That is the same rule as the rest of this post, applied one level down: the query may only use the approved definitions, and now the <em>sentence</em> may only use the rows. <a href="https://insightaiviz.com/blog-why-did-sales-drop.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-ask-the-dashboard">The full worked example is here →</a></p><h3>What actually changes in the organisation</h3><p>Three things, in roughly this order, over the first quarter.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*pzX4HTTTG1de-Ktr.png" /></figure><p>The third is the one that pays for the project. Metric drift is invisible while every team has its own workbook; it becomes obvious and fixable the moment there is a shared vocabulary that everything else draws on.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ZtPNwAfqnKdzYORG.jpg" /></figure><h3>What it does not do</h3><p>It does not replace your analysts, and anyone selling that is selling a demo. It answers questions your model can already express. A genuinely new question — one that needs a table you have not connected, a join nobody declared, a metric nobody defined — still needs a person, and that is the work analysts should be doing rather than re-cutting the same report six ways.</p><p>It is also not a replacement for a dashboard. A number someone looks at every morning belongs on a page, not behind a question that has to be re-asked and re-read each time. The assistant is for the long tail, which happens to be most of the questions.</p><p>And it will not guess. When a question needs a field the model does not have, it says so. In a tool people are going to quote in meetings, refusing is the feature that matters most.</p><blockquote><em>An assistant that is confidently wrong once will not be trusted again, whatever it does for the rest of the year.</em></blockquote><p><a href="https://insightaiviz.com/why-switch.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-ask-the-dashboard">See the full business case →</a> · <a href="https://insightaiviz.com/blog-semantic-layer-ai.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-ask-the-dashboard">How the semantic layer bounds the assistant →</a></p><h3>About — InsightViz</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*NYIhtlCfxQPLqtIr.jpg" /></figure><p><strong>InsightViz</strong> is AI-native business intelligence where the dashboard explains itself — press Why on any chart and it works out what drove the change from your own rows. Part of the <a href="https://www.insightlake.com/">InsightLake</a> enterprise AI suite by VerticalServe.</p><p><strong>VerticalServe Inc</strong> — Niche Cloud, Data &amp; AI/ML Premier Consulting Company, Partnered with Google Cloud, Confluent, AWS, Azure…50+ Customers and many success stories..</p><p><strong>Product:</strong> <a href="https://insightaiviz.com/?utm_source=medium&amp;utm_medium=syndication">https://insightaiviz.com</a></p><p><strong>Website:</strong> <a href="http://www.verticalserve.com/">http://www.VerticalServe.com</a></p><p><strong>Contact:</strong> contact@verticalserve.com</p><p><em>Originally published at </em><a href="https://insightaiviz.com/medium/blog-ask-the-dashboard.html?v=1"><em>https://insightaiviz.com</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=51a6d986d1a1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Half Your Dashboards Haven’t Been Opened in 90 Days]]></title>
            <link>https://verticalserve.medium.com/half-your-dashboards-havent-been-opened-in-90-days-237dd7439479?source=rss-79851bb07a23------2</link>
            <guid isPermaLink="false">https://medium.com/p/237dd7439479</guid>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[data-analytics]]></category>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[VerticalServe Blogs]]></dc:creator>
            <pubDate>Sun, 11 Oct 2026 21:54:38 GMT</pubDate>
            <atom:updated>2026-10-11T21:54:38.462Z</atom:updated>
            <content:encoded><![CDATA[<h4>Count the dashboards nobody reads before you migrate anything. It changes the scope, the cost and the timeline.</h4><p><em>Between half and four fifths of a typical BI estate was not opened by anyone in the last ninety days. Nobody has looked, because nothing in the tooling makes it easy and nobody is rewarded for deleting things. It is the cheapest audit available and it changes the plan.</em></p><p>Ask a BI lead how many dashboards they have and you get a number. Ask how many were opened last quarter and you get a pause. The pause is the finding.</p><p>The measurement takes an afternoon, works on the tool you already have, and routinely turns out to be the largest single input to a migration plan — larger than the licence analysis, larger than the fidelity report.</p><p><em>A worked example</em></p><h4>512 dashboards, counted</h4><p>A 1,200-person insurer, nine years of BI. The estate was quoted as “about five hundred dashboards” with pride. One afternoon with the usage logs produced this.</p><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*b31hiO4ynbDLRrXWEFF4aQ.png" /></figure><p>Illustrative composite. The figures are internally consistent and every step is one the product supports today.</p><p>Eighty-eight dashboards had never been opened by anyone including their author. The migration had been scoped, and budgeted, against 512.</p><p>Three other things fell out of the same query. Forty-one of the dormant dashboards were still driving scheduled extract refreshes, which were still querying the warehouse nightly. Twenty-two seats belonged to people whose only dashboards were in the dormant set. And the three teams who had each built their own revenue definition had done so because a search for “revenue” returned 140 results.</p><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*b31hiO4ynbDLRrXWEFF4aQ.png" /></figure><h3>How an estate gets this way</h3><p>Not carelessness. Four mechanisms, each individually reasonable, compounding over a decade.</p><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*b31hiO4ynbDLRrXWEFF4aQ.png" /></figure><blockquote><em>Every BI estate is a ratchet. Things can be added by anyone in a minute and removed by nobody without a meeting.</em></blockquote><h3>Take the measurement</h3><p>Three numbers, from usage logs you already have. Ninety days is the right window — long enough to catch quarterly reporting, short enough that “someone might need it” stops being an argument.</p><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*b31hiO4ynbDLRrXWEFF4aQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*b31hiO4ynbDLRrXWEFF4aQ.png" /></figure><p>The middle slice deserves attention because it is where the argument happens. A dashboard one person opens weekly is not shelfware to that person. It is also not a shared asset, and it should probably be a saved exploration rather than a maintained page.</p><h4>The number changes the migration scope, not just the tidying</h4><p>Most BI migrations are scoped from the file list, which means budgeting to rebuild and reconcile several hundred dashboards that nobody would notice the absence of. This is the single largest cost overrun available in a BI programme, and it is entirely avoidable by running one query first.</p><h3>What it costs to keep them</h3><p>The instinct is that an unopened dashboard costs nothing. It costs four things, none of which appear on an invoice.</p><ul><li><strong>Search noise.</strong> The reason the third team built their own was that they could not find the first two among four hundred results.</li><li><strong>Metric drift.</strong> Every unopened dashboard contains a definition. Those definitions are in circulation whether or not anyone looks at them, and they surface when someone finds one and trusts it.<a href="https://insightaiviz.com/blog-metric-drift.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-shelfware">How that compounds →</a></li><li><strong>Refresh load.</strong> Extracts feeding dashboards nobody opens still refresh, still query the warehouse, still cost money on a schedule.</li><li><strong>Licences.</strong> Seats are often allocated to the owners of unopened dashboards, and renewals are negotiated on headcount that includes them.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*b31hiO4ynbDLRrXWEFF4aQ.png" /></figure><h3>Retire before you migrate</h3><p>Retiring is faster than migrating, and it improves the estate more. The sequence that works, and the one order that matters:</p><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*b31hiO4ynbDLRrXWEFF4aQ.png" /></figure><ol><li><strong>Archive, do not delete.</strong> Move unopened dashboards somewhere they are still recoverable and stop maintaining them. The distinction matters politically and costs nothing technically.</li><li><strong>Announce it with the list and a date.</strong> “These 287 were not opened in ninety days and will be archived on the 30th.” You will get a handful of objections, and each one is useful information.</li><li><strong>Keep the objections.</strong> A dashboard someone rescues is a dashboard with a confirmed owner, which is more than most of the estate has.</li><li><strong>Migrate what is left.</strong> Now the fidelity report is about content people actually use, and the review list is short enough to work through.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*b31hiO4ynbDLRrXWEFF4aQ.png" /></figure><p><em>A worked example</em></p><h4>What happened when they announced the archive</h4><p>365 dashboards not opened in 90 days. The BI lead published the list with a date three weeks out: on the 30th, these are archived — recoverable, but unmaintained. This is what came back.</p><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*b31hiO4ynbDLRrXWEFF4aQ.png" /></figure><p>Illustrative composite. The figures are internally consistent and every step is one the product supports today.</p><p>Twenty-two objections out of 365, and every one produced something useful: sixteen dashboards that now have an owner, and two embedded surfaces nobody had documented. The two embedded ones are the reason to announce rather than simply archive.</p><h3>Why the new estate stays smaller</h3><p>Retiring once does nothing if the ratchet still works. Two properties change the equilibrium, and neither is a policy:</p><ul><li><strong>A question does not have to become a dashboard.</strong> Most one-off requests exist as pages because asking was not possible. When someone can ask in plain words and keep the answer only if it turns out to matter, the marginal dashboard stops being created.<a href="https://insightaiviz.com/blog-ask-the-dashboard.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-shelfware">What changes when anyone can ask →</a></li><li><strong>A variant does not have to be a copy.</strong> Filters, cross-filtering and drill mean the “same chart but by segment” request is a click rather than a duplicate.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*b31hiO4ynbDLRrXWEFF4aQ.png" /></figure><h3>Two honest caveats</h3><ul><li><strong>Usage logs undercount embedded views.</strong> If dashboards are surfaced inside another application, check that path separately before archiving anything. This is the one way to get the measurement badly wrong.</li><li><strong>Annual dashboards exist.</strong> A ninety-day window misses genuine yearly reporting. Run a twelve-month count alongside and treat the difference as the “check with a human” list rather than as noise.</li></ul><h3>The one paragraph to act on</h3><p>Before you scope a migration, before you negotiate a renewal, before you evaluate anything: count the dashboards nobody opened in ninety days. If it is more than half — and it will be — every other number in your business case is being calculated against the wrong denominator.</p><p><a href="https://insightaiviz.com/blog-measuring-it.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-shelfware">The other numbers worth taking a baseline on →</a> · <a href="https://insightaiviz.com/blog-first-ninety-days.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-shelfware">Where this sits in the sequence →</a></p><h3>About — InsightViz</h3><p><strong>InsightViz</strong> is AI-native business intelligence where the dashboard explains itself — press Why on any chart and it works out what drove the change from your own rows. Part of the <a href="https://www.insightlake.com/">InsightLake</a> enterprise AI suite by VerticalServe.</p><p><strong>VerticalServe Inc</strong> — Niche Cloud, Data &amp; AI/ML Premier Consulting Company, Partnered with Google Cloud, Confluent, AWS, Azure…50+ Customers and many success stories..</p><p><strong>Product:</strong> <a href="https://insightaiviz.com/?utm_source=medium&amp;utm_medium=syndication">https://insightaiviz.com</a></p><p><strong>Website:</strong> <a href="http://www.verticalserve.com/">http://www.VerticalServe.com</a></p><p><strong>Contact:</strong> contact@verticalserve.com</p><p><em>Originally published at </em><a href="https://insightaiviz.com/medium/blog-shelfware.html?v=1"><em>https://insightaiviz.com</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=237dd7439479" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Query Took 34 ms. The Dashboard Took 2.27 Seconds.]]></title>
            <link>https://verticalserve.medium.com/the-query-took-34-ms-the-dashboard-took-2-27-seconds-7fec3c808c98?source=rss-79851bb07a23------2</link>
            <guid isPermaLink="false">https://medium.com/p/7fec3c808c98</guid>
            <category><![CDATA[data-analytics]]></category>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[VerticalServe Blogs]]></dc:creator>
            <pubDate>Wed, 07 Oct 2026 22:57:33 GMT</pubDate>
            <atom:updated>2026-10-07T22:57:33.077Z</atom:updated>
            <content:encoded><![CDATA[<h4>Where the time in a slow dashboard actually goes, and whose problem it is.</h4><p><em>A dashboard that takes eleven seconds is a dashboard people screenshot once and stop opening. The seconds are rarely where teams look for them, and two of the biggest are not in the query at all.</em></p><p>Ask why a dashboard is slow and the answer is almost always “the warehouse”. Sometimes that is true. Often the query itself is a small part of the wait, and the rest is structural — decisions about how a page asks for its data that nobody revisits because they are invisible.</p><p>Here is where the time actually goes on a page with eight panels, and what can be done about each part.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*WoBXZCwt8zwg9_9p.png" /></figure><p><em>A worked example</em></p><h4>Eight panels, profiled</h4><p>Our own test dashboard, measured rather than estimated: eight panels over a 600-row local extract — deliberately the least favourable case for any caching, because there is almost nothing to cache away. The ratio is the finding, not the absolute numbers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*t272FGaytCHZzxGl.png" /></figure><p>Illustrative composite. The figures are internally consistent and every step is one the product supports today.</p><p><strong>34 milliseconds of engine time on a page that took 2.27 seconds.</strong> Any amount of query tuning would have found nothing, because the query was never the problem. The single biggest win was noticing that each panel was opening its own database transaction to write an audit row.</p><p>That is the general lesson and it generalises beyond this product: profile the page, not the query. On a page of small aggregates the query is frequently a rounding error, and the instinct to optimise it is the instinct to look where the light is.</p><h3>1. Stop asking the same question repeatedly</h3><p>A row of KPI cards is one SELECT. Total sales, total profit and profit ratio over the same period, with the same filters, differ only in which measure they name. Run them as three panels and you ask the warehouse the same question three times and pay for it three times.</p><p>InsightViz pools them. Panels whose queries differ <em>only</em> in their metrics — same model, same dimensions, same filters, same ordering — are merged into one query, and each panel is handed back only its own columns. Nothing else in the system knows it happened: the chart builder, the export and the cache all behave as before.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*nE2-iLMzQNhRp0Hj.png" /></figure><h4>What deliberately does not get pooled</h4><p>A panel with its own drill state, because drilling changes the query for that panel alone. A panel targeted by a cross-filter click its neighbours did not get, because half a WHERE clause is not a smaller answer — it is a wrong one. And a top-ten ordered by a metric, because top ten by revenue and top ten by margin are different ten rows. Each of those costs one query and changes no number, which is the right trade.</p><h3>2. Show the last known numbers while you fetch new ones</h3><p>A dashboard that shows nothing for two seconds and then everything feels slower than one that shows last night’s figures immediately and swaps them a moment later. Perceived time is the only time the reader experiences.</p><p>So each panel’s last result is kept, and the page asks for two things at once: whatever is already stored, and the truth. Whichever lands first gets painted.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*MYRwdVFUSNdBhbuw.png" /></figure><p>The honesty matters more than the speed. Anything served from the store is marked stale and carries the time it was computed, and the page says so above the panels. An old number that looks current is worse than a blank panel — that is the failure mode this feature could easily have introduced, so it is designed against explicitly.</p><h4>How the stored result stays safe</h4><p>The key covers which source it was, the exact SQL, and the bound values — and a person’s row and column rules are already part of that SQL. Two people with different access therefore compute different keys and can never be served each other’s numbers. A new extract version also changes the key, so a refresh cannot serve yesterday’s rows.</p><h3>3. The overhead nobody profiles</h3><p>Here is the finding that is worth more than either feature above, and it is slightly embarrassing. On a test dashboard with eight panels, the query engine accounted for <strong>34 milliseconds in total</strong>. The page took 2.27 seconds.</p><p>The gap was almost entirely per-panel bookkeeping — in particular a separate database transaction to write the audit log for each panel. Batching those into one transaction per page took it to 0.67 seconds without touching a single query.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*_BbTOzrJc0XywHKm.png" /></figure><p>Measured on an eight-panel dashboard over a 600-row local extract — deliberately the least favourable case for caching, because there is almost nothing to cache away. The point is the ratio, not the absolute numbers: on this page the engine was never the problem, and any amount of query tuning would have found nothing.</p><blockquote><em>Before optimising the query, measure how much of the wait is the query. On a page of small aggregates, it is often a rounding error.</em></blockquote><h3>Why the extract is doing most of the work</h3><p>None of the above would matter much if every panel hit the source warehouse directly. The reason engine time is measured in milliseconds is that a dashboard reads a governed, aggregated copy in your own cloud storage — columnar Parquet, read by an embedded engine, with only the needed columns and row groups touched.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*IoMAAAzu6_4-7Wv1.jpg" /></figure><h3>The limits, plainly</h3><ul><li><strong>None of this speeds up a slow source.</strong> On a live connection, the connection decides. Pooling still cuts the number of round trips, and painting from the store still fills the page immediately, but a query that takes nine seconds in your warehouse takes nine seconds here.</li><li><strong>An extract is a copy, with a staleness budget.</strong> That is the right default for reporting and the wrong one for an operational screen that must be current to the second. Live connections exist for that case, and they are honest about being slower.</li><li><strong>Pooling helps rows of similar panels most.</strong> A dashboard of eight genuinely different questions pools nothing, and should not.</li></ul><h3>What to measure on your own estate</h3><ol><li><strong>Time the page, not the query.</strong> The number that matters is from click to last panel painted, and it is usually much larger than the sum of the query times.</li><li><strong>Count the round trips.</strong> Open the network panel. If a page of twelve tiles makes twelve requests, several of them are asking the same question.</li><li><strong>Check what happens in the first 500 ms.</strong> If it is a blank rectangle, that is the cheapest second you will ever recover.</li></ol><p><a href="https://insightaiviz.com/docs.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-why-slow#reading">How a dashboard behaves for a reader →</a></p><h3>About — InsightViz</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*bU8GumY__tl-Q-tm.jpg" /></figure><p><strong>InsightViz</strong> is AI-native business intelligence where the dashboard explains itself — press Why on any chart and it works out what drove the change from your own rows. Part of the <a href="https://www.insightlake.com/">InsightLake</a> enterprise AI suite by VerticalServe.</p><p><strong>VerticalServe Inc</strong> — Niche Cloud, Data &amp; AI/ML Premier Consulting Company, Partnered with Google Cloud, Confluent, AWS, Azure…50+ Customers and many success stories..</p><p><strong>Product:</strong> <a href="https://insightaiviz.com/?utm_source=medium&amp;utm_medium=syndication">https://insightaiviz.com</a></p><p><strong>Website:</strong> <a href="http://www.verticalserve.com/">http://www.VerticalServe.com</a></p><p><strong>Contact:</strong> contact@verticalserve.com</p><p><em>Originally published at </em><a href="https://insightaiviz.com/medium/blog-why-slow.html?v=1"><em>https://insightaiviz.com</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7fec3c808c98" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Sales Fell. The Dashboard Explained Why — Without Guessing]]></title>
            <link>https://verticalserve.medium.com/sales-fell-the-dashboard-explained-why-without-guessing-8e2c4951939c?source=rss-79851bb07a23------2</link>
            <guid isPermaLink="false">https://medium.com/p/8e2c4951939c</guid>
            <category><![CDATA[data-analytics]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[business-intelligence]]></category>
            <dc:creator><![CDATA[VerticalServe Blogs]]></dc:creator>
            <pubDate>Tue, 06 Oct 2026 11:33:07 GMT</pubDate>
            <atom:updated>2026-10-06T11:33:07.338Z</atom:updated>
            <content:encoded><![CDATA[<h4>What AI-native BI looks like when every sentence is traced back to a row of your data.</h4><p><em>Every chart in every BI tool can show you that a number fell. Not one of them tells you why. That question goes to an analyst, comes back on Thursday, and the meeting has moved on. Here is what it looks like when the dashboard answers it in one click — worked out from the rows, not guessed by a model.</em></p><p>A regional director opens the Monday dashboard. Sales are down. The chart is perfectly clear about <em>that</em>. It is silent on everything she actually needs: was it one region or all of them, one product line or the whole range, one bad week or a slide that started in June, and is this the lowest it has been. Those four questions are the meeting. The chart answers none of them.</p><p>In InsightViz the chart has a button on it that says <strong>Why?</strong>. Here is what it does on the sample workbook that ships with the product:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*8-joJmtKAeHnvMzA.jpg" /></figure><p>Nothing on that panel was written by a language model. The two periods, the contribution of every region, category, segment and sales band, the ranking, the member that moved the other way and the sentences themselves are all computed. A model is asked, optionally, to write a three-sentence summary over the ranked tables — and that summary is then checked figure by figure against those tables before anyone sees it. If none of its figures trace, it is not shown; if only some do, it is shown with a warning mark beside it, so the reader knows which sentences to lean on.</p><h3>What “worked out” means</h3><p>The explanation is a short, fixed procedure that runs through the same query compiler every dashboard uses, with the reader’s own access rules inside every query. It never writes SQL of its own and never sees a table the reader may not.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*JW0j_jKwcfUvHwUD.png" /></figure><p>Three properties fall out of that order, and they are the reason a finance director will trust the panel where they would not trust a chat window:</p><ul><li><strong>It cannot invent a cause.</strong> A sentence only ever says what a row supports. “West fell because of two lost accounts” can be said if the rows show it; “because of the weather” cannot be said at all.</li><li><strong>It respects who is asking.</strong> A regional manager who may only see their region gets an explanation of their region. The rules are in the queries, so there is no version of the answer that leaks.</li><li><strong>It is fast enough to click.</strong> A handful of grouped queries on a governed extract takes a few hundred milliseconds. The sample dashboard explains itself in under a second.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*r4z21AhkjYF28rVo.png" /></figure><h3>The worked example, in numbers</h3><p>The unfiltered version of the same panel — every region and every segment — gives the figures below. This is the sample workbook every trial starts with, so you can reproduce it exactly.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*F0nTH_Q-WLoMVA0h.png" /></figure><p>Notice what the ranking did. Three dimensions each have a member that “explains” about half the decline. That is not three causes; it is what an even split looks like when a dimension has two or three members. The panel scores each dimension by how far its leader is <em>above</em> an even split, so a genuinely concentrated driver ranks above a large member of a small dimension. When nothing is concentrated it says so: <em>the decline is broad; no single region explains more than 39% of it.</em></p><p><em>A worked example</em></p><h4>The Monday meeting</h4><p>The director has the panel open. The chart said sales fell; the panel said the High sales band carried most of it and that August was the lowest month in twenty. What she does next is the part no demo shows, because it is the follow-ups that make or break the tool.</p><p><strong>Click — “Look at High on its own”.</strong> The whole dashboard narrows to the High band. Every other chart follows: the trend, the category split, the headline number.</p><p><strong>Click — “Break High down by category”.</strong> The same explanation, now for High only: Technology inside the High band fell hardest. Two clicks, and the question has gone from a company-wide number to a product line in a price band.</p><p><strong>Ask — “Is this unusual for a month?”.</strong> The answer uses the twenty months on the panel: it is the largest single-month fall in the series and the lowest level; the previous low was February 2024. Nothing outside the panel was consulted, and the answer says so.</p><p><strong>Ask — “Was it fewer orders or smaller ones?”.</strong> The panel does not carry order counts, so one further governed query is run — sales and order count for the two months — and the answer is drawn from that. The query it ran is shown under the answer.</p><p><strong>Click — “Alert if Sales falls below $11,485”.</strong> An alert is set on the panel, to her, Monday mornings. If it happens again she hears before the meeting, not in it.</p><p>Illustrative composite. The figures are internally consistent and every step is one the product supports today.</p><p>Every step is either a click on something the panel already computed or a question answered from the panel’s own evidence. The one time the evidence was not enough, the assistant ran one more governed query and showed it — it did not guess, and it did not go and read the whole warehouse.</p><h3>What a chat window does instead</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*bY5gug_nDa5kw1EF.png" /></figure><p>This is the same rule the product applies to every question — a query may only use the approved definitions, an answer may only use the rows it was given — carried one step further, to the sentence. It is why the panel can sit in front of a CFO. The chat box has to sit in front of an analyst who checks it.</p><h3>What changed, before you ask</h3><p>The same procedure runs across every panel of a dashboard, latest period against the one before, and ranks the moves. It is on the dashboard as <strong>What changed</strong>, on the Home page across every dashboard you can see, and in the body of the scheduled email above the attachment — so the person who reads the Monday PDF sees the two lines that matter before they open it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*lcSTQ_WVeTw2TPRu.jpg" /></figure><h4>What it will not do</h4><p>The panel explains what the data shows. It does not know what happened in the world. “High-band sales fell because two Technology orders in August were half the size of July’s” is what it can say; <em>why those customers ordered less</em> is a question for a person, and the honest answer to it is a phone call, not a chart. The product says exactly that, in one sentence, and names what would answer it.</p><h3>Keep the question</h3><p>At the foot of every explanation is <strong>Make this an insight</strong>. It keeps the question rather than the answer: the trend, the breakdowns that drove it and a written synthesis become a story that re-runs whenever the data does, at whatever filters the reader has on, with every figure traced the same way. Published, it can sit on the dashboard as a panel and go out in the Monday email. <a href="https://insightaiviz.com/blog-before-the-meeting.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-why-did-sales-drop">The next post</a> is about that.</p><h3>Try it on your own numbers</h3><p>Every InsightViz trial starts with the sample workbook above, so the first thing your team clicks is the panel in this post. Import one of your own workbooks and the same button is on every chart it produces — the explanation runs on your definitions, under your access rules, in your cloud.</p><h3>About — InsightViz</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*NdJUmDMdpbp1PBs8.jpg" /></figure><p><strong>InsightViz</strong> is AI-native business intelligence where the dashboard explains itself — press Why on any chart and it works out what drove the change from your own rows. Part of the <a href="https://www.insightlake.com/">InsightLake</a> enterprise AI suite by VerticalServe.</p><p><strong>VerticalServe Inc</strong> — Niche Cloud, Data &amp; AI/ML Premier Consulting Company, Partnered with Google Cloud, Confluent, AWS, Azure…50+ Customers and many success stories..</p><p><strong>Product:</strong> <a href="https://insightaiviz.com/?utm_source=medium&amp;utm_medium=syndication">https://insightaiviz.com</a></p><p><strong>Website:</strong> <a href="http://www.verticalserve.com/">http://www.VerticalServe.com</a></p><p><strong>Contact:</strong> contact@verticalserve.com</p><p><em>Originally published at </em><a href="https://insightaiviz.com/medium/blog-why-did-sales-drop.html?v=1"><em>https://insightaiviz.com</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8e2c4951939c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Your AI Will Inherit Your Three Definitions of Revenue]]></title>
            <link>https://verticalserve.medium.com/your-ai-will-inherit-your-three-definitions-of-revenue-f4adf963a194?source=rss-79851bb07a23------2</link>
            <guid isPermaLink="false">https://medium.com/p/f4adf963a194</guid>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[tableau]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[data-visualization]]></category>
            <dc:creator><![CDATA[VerticalServe Blogs]]></dc:creator>
            <pubDate>Mon, 05 Oct 2026 13:38:01 GMT</pubDate>
            <atom:updated>2026-10-05T13:38:01.975Z</atom:updated>
            <content:encoded><![CDATA[<h4>Metric drift is the first thing to fix before you put an AI assistant on top of your BI.</h4><p>It starts reasonably. Someone builds a workbook with a calculated field called <em>Net Revenue</em>. A second team needs almost the same thing, so they copy the workbook and adjust it — their business excludes intercompany transfers. A year later a third team builds their own from scratch because they could not find the first two.</p><p>Now there are three definitions of net revenue, all correct in their original context, all called the same thing, and none of them visible to anyone who has not opened the file.</p><p><em>A worked example</em></p><h4>Three definitions of net revenue at a £240m distributor</h4><p>Nobody wrote a wrong formula. Each was correct for the question its author was answering, and each was copied forward for years. This is what they were.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*HV0M2-cFZHjIGIOs.png" /></figure><p>Illustrative composite. The figures are internally consistent and every step is one the product supports today.</p><p>A £4.3m spread on the same quarter, and every one of the three reconciles perfectly against itself month over month. That is why nothing looks wrong until two of them meet.</p><p><strong>Five years, in four steps</strong></p><p><strong>2019 — One definition.</strong> Sales builds a workbook with a calculated field. It is correct for sales commission, which is what it was for.</p><p><strong>2021 — A second, for a different purpose.</strong> Finance needs a statutory-aligned figure and cannot see how the sales one is defined, so they build their own. Also correct — for their purpose.</p><p><strong>2024 — A third, by accident.</strong> A new ops analyst searches the estate, finds four hundred dashboards, cannot tell which figure is authoritative, and builds a fresh one. Also correct.</p><p><strong>The board meeting — All three in one room.</strong> Sales presents £62.4m growth against target. Finance’s pack says £58.1m. Forty minutes of the meeting goes on establishing that both are right, and the agenda item they were there for is carried.</p><p><strong>Afterwards — The real damage.</strong> Nobody trusts a number they did not produce. Three teams start keeping their own, which is how a fourth definition gets created.</p><h3>Why it is invisible until it is expensive</h3><p>Drift has no symptoms until two of the copies meet. Each team’s number is internally consistent and reconciles with itself month over month, so every dashboard looks healthy. The divergence only surfaces when two teams present to the same audience — usually a board pack, usually at the worst time.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*RSXkbDoueVttD-ML.png" /></figure><p>The cost is not the reconciliation work, painful as that is. The cost is that after the second or third occurrence, people stop trusting <em>any</em> number they did not produce themselves — and start keeping their own. That is the point at which a BI estate has failed at its actual job, regardless of how many dashboards it contains.</p><h4>The tell</h4><p>If your finance function routinely rebuilds numbers that already exist on a dashboard before putting them in a pack, you already have this problem. The rebuild is not diligence. It is a rational response to not being able to see how the dashboard figure was defined.</p><h3>Four things that make drift inevitable</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*9YHtoAvBv8Pdl-Ww.png" /></figure><p>None of these are bugs. They are consequences of the file being the unit of work. Any tool built that way produces the same outcome given enough time and enough teams.</p><h3>Defining it once, where everyone can see it</h3><p>In InsightViz the definitions are not inside the dashboards. They are a separate, versioned object — the business model — that dashboards, explorations and AI answers all draw on. A metric has a business name, the expression behind it, a description, and the other words people use for it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*VtLSRLwI13WR0t-7.jpg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*OljaeCbLSn_l4QGj.png" /></figure><h3>The part that is actually hard</h3><p>Here is the concession, and it is a large one: <strong>no tool settles your definitions for you.</strong> Putting them in one place makes the disagreements visible; it does not resolve them. When three teams’ definitions land in one model, somebody has to decide which one net revenue means — and that is a business decision with consequences for people’s targets.</p><p>What changes is that the argument happens <em>once</em>, deliberately, in week three of a migration, instead of repeatedly and by accident in board meetings for the next five years.</p><blockquote><em>The migration does not fix your metric definitions. It forces you to have the conversation you have been avoiding, at a moment when you have budget and attention for it.</em></blockquote><p>In practice this is the most valuable phase of a BI migration and the one most likely to slip. Plan for it: two or three weeks, the people who own the numbers rather than the people who own the tool, and a decision log. The import gives you a first draft — every field the workbooks already named — so the conversation starts from what exists rather than from a blank page.</p><h3>Why this also decides whether AI is useful</h3><p>An assistant answering questions about your data is only as trustworthy as the vocabulary it is limited to. Point one at three conflicting definitions of net revenue and it will confidently pick one — and you will have industrialised the drift rather than fixed it.</p><p>This is the part most AI-for-BI pitches skip. The model is not the hard problem. The hard problem is that most organisations cannot state, in one place, what their own metrics mean — and until they can, every answer is a guess with good grammar.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*WyKDkRBVtLUHgKha.jpg" /></figure><h3>Where to start, if this is familiar</h3><ol><li><strong>Count them.</strong> Search your estate for the three or four metrics that appear in board packs. How many distinct definitions are in circulation? The number is usually worse than the guess.</li><li><strong>Import a few real workbooks.</strong> The field names and calculations come across as a first draft of the model, which is a far more productive starting point than a workshop with a whiteboard.</li><li><strong>Settle the top ten, not all of them.</strong> A small number of metrics carry almost all the decisions. The long tail can drift for another year without anyone noticing.</li><li><strong>Write down who owns each one.</strong> Drift resumes the moment a definition has no owner, whatever tool it lives in.</li></ol><p><a href="https://insightaiviz.com/why-switch.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-metric-drift">The full business case →</a> · <a href="https://insightaiviz.com/blog-ask-the-dashboard.html?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=blog-metric-drift">Why the vocabulary decides whether AI is useful →</a></p><h3>About — InsightViz</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*3hnxpoxKS5x7lPRU.jpg" /></figure><p><strong>InsightViz</strong> is AI-native business intelligence where the dashboard explains itself — press Why on any chart and it works out what drove the change from your own rows. Part of the <a href="https://www.insightlake.com/">InsightLake</a> enterprise AI suite by VerticalServe.</p><p><strong>VerticalServe Inc</strong> — Niche Cloud, Data &amp; AI/ML Premier Consulting Company, Partnered with Google Cloud, Confluent, AWS, Azure…50+ Customers and many success stories..</p><p><strong>Product:</strong> <a href="https://insightaiviz.com/?utm_source=medium&amp;utm_medium=syndication">https://insightaiviz.com</a></p><p><strong>Website:</strong> <a href="http://www.verticalserve.com/">http://www.VerticalServe.com</a></p><p><strong>Contact:</strong> contact@verticalserve.com</p><p><em>Originally published at </em><a href="https://insightaiviz.com/medium/blog-metric-drift.html?v=3"><em>https://insightaiviz.com</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f4adf963a194" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>