<?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 Theo Acquah on Medium]]></title>
        <description><![CDATA[Stories by Theo Acquah on Medium]]></description>
        <link>https://medium.com/@theobee.acquah?source=rss-9081f25673b2------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*m3lV-XewngrVyhbNTz6bJQ.png</url>
            <title>Stories by Theo Acquah on Medium</title>
            <link>https://medium.com/@theobee.acquah?source=rss-9081f25673b2------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 22:16:54 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@theobee.acquah/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[When the Data Tells a Story Nobody Wants to Hear]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@theobee.acquah/when-the-data-tells-a-story-nobody-wants-to-hear-537e4e5889b9?source=rss-9081f25673b2------2"><img src="https://cdn-images-1.medium.com/max/2172/1*4UZ-WP1TNbJLNdWYL4YalA.png" width="2172"></a></p><p class="medium-feed-snippet">Finding the truth is only half the analyst&#x2019;s job. What is the other half?</p><p class="medium-feed-link"><a href="https://medium.com/@theobee.acquah/when-the-data-tells-a-story-nobody-wants-to-hear-537e4e5889b9?source=rss-9081f25673b2------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@theobee.acquah/when-the-data-tells-a-story-nobody-wants-to-hear-537e4e5889b9?source=rss-9081f25673b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/537e4e5889b9</guid>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[storytelling]]></category>
            <category><![CDATA[data-analytics]]></category>
            <category><![CDATA[critical-thinking]]></category>
            <category><![CDATA[leadership]]></category>
            <dc:creator><![CDATA[Theo Acquah]]></dc:creator>
            <pubDate>Wed, 18 Nov 2026 17:03:39 GMT</pubDate>
            <atom:updated>2026-11-18T17:03:39.660Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[More Data Won’t Always Give You a Better Answer]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@theobee.acquah/more-data-wont-always-give-you-a-better-answer-3c380c3a47b6?source=rss-9081f25673b2------2"><img src="https://cdn-images-1.medium.com/max/2172/1*ruAFgMtTvxiSd-1wDzsXWg.png" width="2172"></a></p><p class="medium-feed-snippet">Knowing when to stop analysing and decide</p><p class="medium-feed-link"><a href="https://medium.com/@theobee.acquah/more-data-wont-always-give-you-a-better-answer-3c380c3a47b6?source=rss-9081f25673b2------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@theobee.acquah/more-data-wont-always-give-you-a-better-answer-3c380c3a47b6?source=rss-9081f25673b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/3c380c3a47b6</guid>
            <category><![CDATA[critical-thinking]]></category>
            <category><![CDATA[data-analytics]]></category>
            <category><![CDATA[decision-making]]></category>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[leadership]]></category>
            <dc:creator><![CDATA[Theo Acquah]]></dc:creator>
            <pubDate>Tue, 10 Nov 2026 17:54:51 GMT</pubDate>
            <atom:updated>2026-11-10T17:54:51.293Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Your predictive model was right. Your decision was wrong.]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@theobee.acquah/your-predictive-model-was-right-your-decision-was-wrong-c478b8825ff2?source=rss-9081f25673b2------2"><img src="https://cdn-images-1.medium.com/max/1774/1*x5zy2aN-HZSKYLoBhWlPtA.png" width="1774"></a></p><p class="medium-feed-snippet">Why accurate predictions don&#x2019;t necessarily lead to good business decisions.</p><p class="medium-feed-link"><a href="https://medium.com/@theobee.acquah/your-predictive-model-was-right-your-decision-was-wrong-c478b8825ff2?source=rss-9081f25673b2------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@theobee.acquah/your-predictive-model-was-right-your-decision-was-wrong-c478b8825ff2?source=rss-9081f25673b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/c478b8825ff2</guid>
            <category><![CDATA[decision-making]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[data-analytics]]></category>
            <category><![CDATA[predictive-analytics]]></category>
            <dc:creator><![CDATA[Theo Acquah]]></dc:creator>
            <pubDate>Tue, 03 Nov 2026 18:26:17 GMT</pubDate>
            <atom:updated>2026-11-03T18:26:17.563Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[When a Good KPI Becomes a Bad Target]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@theobee.acquah/when-a-good-kpi-becomes-a-bad-target-99cf479f68a8?source=rss-9081f25673b2------2"><img src="https://cdn-images-1.medium.com/max/1536/1*QlCCGlwucptx88JJVUwKdw.png" width="1536"></a></p><p class="medium-feed-snippet">The moment people are rewarded for hitting a number, they may start improving the number instead of the business.</p><p class="medium-feed-link"><a href="https://medium.com/@theobee.acquah/when-a-good-kpi-becomes-a-bad-target-99cf479f68a8?source=rss-9081f25673b2------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@theobee.acquah/when-a-good-kpi-becomes-a-bad-target-99cf479f68a8?source=rss-9081f25673b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/99cf479f68a8</guid>
            <category><![CDATA[data-analytics]]></category>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[kpi]]></category>
            <category><![CDATA[decision-making]]></category>
            <category><![CDATA[storytelling]]></category>
            <dc:creator><![CDATA[Theo Acquah]]></dc:creator>
            <pubDate>Tue, 27 Oct 2026 17:18:37 GMT</pubDate>
            <atom:updated>2026-10-27T17:18:37.700Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[How to Make Your Dashboard Extraordinary]]></title>
            <link>https://medium.com/@theobee.acquah/how-to-make-your-dashboard-extraordinary-8e15ca19b371?source=rss-9081f25673b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/8e15ca19b371</guid>
            <category><![CDATA[dashboard]]></category>
            <category><![CDATA[data-analytics]]></category>
            <category><![CDATA[data-visualization]]></category>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[storytelling]]></category>
            <dc:creator><![CDATA[Theo Acquah]]></dc:creator>
            <pubDate>Wed, 21 Oct 2026 03:19:45 GMT</pubDate>
            <atom:updated>2026-10-21T03:19:45.637Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="A three-panel cartoon strip. In Panel 1, Bob faces a crowded dashboard and says, “It has twelve charts, so it must be impressive.” In Panel 2, Priti removes clutter until one urgent insight becomes visible and says, “Now it has something to say.” In Panel 3, Jojo holds his coffee mug securely and concludes, “A great dashboard doesn’t display everything — it reveals what matters.”" src="https://cdn-images-1.medium.com/max/1024/1*s6CgfOxo6qfzGdUy16I64w.png" /><figcaption>From Dashboard Clutter to Clarity — TheoToon™ by Theo Acquah · Created with AI assistance</figcaption></figure><blockquote>“The greatest value of a picture is when it forces us to notice what we never expected to see.” — John Tukey</blockquote><p>A standard dashboard provides information, but an outstanding one captures attention, clarifies meaning, and supports decision-making. Rather than trying to impress the audience by emphasizing how much data the analyst has collected, it shows how well the analyst understands the business problem.</p><p>Here are twelve ways to make your dashboard stand out.</p><h3>1. Begin with a decision, not a collection of data.</h3><p>Before you open Tableau, Power BI, or another visualization tool, identify the decision the dashboard must support.</p><p>Should a sales manager change his strategy for each region? Should the finance department investigate unexpected costs?</p><p>The decision guides the dashboard. All charts, KPIs, filters, and annotations should aim to answer it.</p><p>Even if it took several hours to develop, delete a visual if it doesn’t align with the decision.</p><h3>2. Give the dashboard a clear point of view.</h3><p>A really good dashboard doesn’t just tell people what happened. It also helps them understand what the information means.</p><p>Consider the difference between these two titles.</p><ul><li>“Portfolio Performance Dashboard”</li><li>“71 Projects Require Immediate Intervention”</li></ul><p>The first is a label; the second is a message.</p><p>A strong title can communicate the main finding before the user examines any charts. If the main insight changes over time, a dynamic title can make the dashboard even more useful.</p><h3>3. A dashboard needs a strong visual hierarchy.</h3><p>Not every number or visualization deserves equal prominence. Highlight the most important information by placing the main KPIs and findings at the top, where users typically begin their scan. Create a clear reading sequence using size, position, color, and spacing.</p><p>A good example of visual hierarchy is:</p><ol><li>What is the current situation?</li><li>Where is the problem?</li><li>Why is it happening?</li><li>What requires action?</li></ol><p>If the charts all make statements, then the dashboard becomes noise. Instead, let the key insight speak first.</p><h3>4. Pick KPIs that measure outcomes.</h3><p>Dashboards often get crowded because analysts tend to include every available metric. The most impressive dashboards are very selective.</p><p>Choose KPIs that reflect progress toward a meaningful outcome. A project portfolio dashboard, for example, may highlight projects requiring intervention, forecast cost overrun, schedule exposure, open high-severity risks, and resource-capacity coverage.</p><p>Avoid performance measures that look impressive but don’t drive decisions. Include a key performance indicator only if it helps the user evaluate performance, risk, or the action needed.</p><h3>5. Be careful to use the right chart for the question.</h3><p>For categories such as Risk Types or Price Plans, use bar charts.</p><p>If you need to show trends over time (for example, “Monthly Revenue in USD”), a line graph is ideal.</p><p>To show the relationship between two numerical variables (for example, Schedule Delay in Days versus Forecast Overrun in USD), use a scatterplot.</p><p>And when the dashboard user needs to find specific records, use detailed tables.</p><p>Don&#39;t select a chart merely because it looks “cool” or fancy, or because you want to impress. For example, an unfamiliar visual that requires a lengthy explanation may be less effective than a simple bar chart.</p><p>The most effective chart is the one that makes the intended comparison easiest to see.</p><h3>6. You have to use color strategically.</h3><p>Use color to communicate meaning, not to decorate your design.</p><p>Reserve strong colors for information that needs attention. For example, Grey for context, Blue for normal performance, Amber for warning, and Red for critical exceptions.</p><p>Use the same color scheme throughout. If red means ‘high risk’ in one chart, it shouldn&#39;t mean ‘top performer’ in another.</p><h3>7. Remove visual clutter.</h3><p>Gridlines, borders, shadows, extra labels, redundant legends, decorative icons, and unnecessary decimal places all compete for the viewer’s attention. Most of these elements are software artifacts or defaults that are usually not useful. For every visual element, ask yourself, “Does this help the user to understand the data?” If not, delete it.</p><p>Take advantage of whitespace. Whitespace is not blank space. It separates related elements and highlights key information. Clean dashboards often look more professional than cluttered ones.</p><h3>8. Make comparisons effortless.</h3><p>A number is usually meaningless without context.</p><p>Whether a forecast overrun of $33 million is good or bad depends on whether you have a target, a previous period, a benchmark, or a threshold.</p><p>Useful comparisons include:</p><ul><li>Actual versus target</li><li>Current period versus previous period</li><li>Forecast versus budget</li><li>One region versus the organizational average</li><li>Current risk level versus tolerance</li></ul><p>Whenever you refer to a figure, compare it, since this gives the figure importance and lets the audience quickly tell whether performance is improving, deteriorating, or flat.</p><h3>9. You must design filters around real questions.</h3><p>Filters should help users investigate rather than forcing them to redo the analysis. Include only filters that support realistic business questions, such as:</p><p><em>Which regions are underperforming?</em></p><p><em>Which projects are at the highest level of risk?</em></p><p><em>What accounting period am I interested in?</em></p><p>Use clear names and reasonable defaults for filters. And show the actively selected options so users always know what they are looking at.</p><p>Use as few filters as possible because a dashboard with numerous filters usually places too much analytical work on the user</p><h3>10. Surface exceptions and actions.</h3><p>Business leaders don’t need a dashboard to confirm what they already know. They only need a dashboard to know where to focus their attention.</p><p>A great dashboard makes it easy to identify issues such as projects exceeding cost thresholds, customers showing churn signals, or regions missing targets.</p><p>It doesn’t end there. The dashboard also links each issue to an action by showing who is responsible, the severity level, how long it’s been unresolved, and what to do next.</p><p>It’s not just for monitoring. It is also used for management.</p><h3>11. Build trust through validation and transparency.</h3><p>A beautiful dashboard is useless if users cast doubt on its figures.</p><p>Review the totals, verify the filters, calculated fields, joins, date logic, and aggregation levels, and ensure the key figures match those in the source systems.</p><p>Also, test filter combinations, look for duplicate records, and check for issues in the calculated fields.</p><p>Explain the key definitions, particularly those that might be interpreted differently by people using the metrics. Although expressions such as ‘revenue’, ‘active customer’, and ‘at-risk project’ may seem straightforward at first glance, they actually include several business rules.</p><p>Exceptional dashboards don’t just look authoritative. They can also withstand scrutiny.</p><h3>12. Finish with a clear takeaway.</h3><p>After exploring the dashboard, the audience should be able to complete this sentence:</p><p>“On the basis of what I have observed, we should…”</p><p>You can achieve that objective with an executive summary, a brief annotation, a callout box, or a prioritized list of recommended actions.</p><p>Don’t leave the user with twelve charts and expect them to make up their own minds. Data visualization isn’t complete when the information appears on the screen. It’s finished only when the audience knows what to focus on and why.</p><h3>Extraordinary means useful.</h3><p>A dashboard doesn’t need graphics, animation, or a variety of colors to be good. Careful thought gives it its special quality.</p><p>It provides a genuine answer, sets priorities, simplifies comparisons, spotlights exceptions, builds trust, and, most importantly, helps a person make better decisions.</p><p>In Tukey’s view, a picture’s main benefit is helping us notice what is unexpected. It applies to dashboards too.</p><p>The dashboard should go beyond simply showing information people already know and instead help them notice things they might miss.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8e15ca19b371" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Good Data, Bad Charts: 12 Habits That Bury the Story]]></title>
            <link>https://medium.com/@theobee.acquah/good-data-bad-charts-12-habits-that-bury-the-story-77f5784ab769?source=rss-9081f25673b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/77f5784ab769</guid>
            <category><![CDATA[data-visualization]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[storytelling]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[business-intelligence]]></category>
            <dc:creator><![CDATA[Theo Acquah]]></dc:creator>
            <pubDate>Thu, 08 Oct 2026 03:30:47 GMT</pubDate>
            <atom:updated>2026-10-08T03:30:47.387Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Bob proudly presents a crowded, colourful dashboard: “Everything is on it!” Priti asks, “Where should we look?” Jojo highlights one bar labelled “Needs attention” and replies, “Start with what needs a decision.”" src="https://cdn-images-1.medium.com/max/1024/1*e1jadcz1d1cswdVS_tcwAQ.png" /><figcaption>Too Much Data, Too Little Focus — TheoToon™ by Theo Acquah · Created with AI assistance</figcaption></figure><p>You can spot a mediocre dashboard at once. After months spent perfecting the data, teams often implement a cluttered, run-of-the-mill dashboard that fails to impress. Executives glance at the result, feign interest, and promptly return to asking for spreadsheets.</p><p>Here’s how to recognize a dashboard that misses the mark.</p><h3>1. The Lonely Number</h3><p>A chart conspicuously displays a message such as “Total Users: 54,302”</p><p>However, it is not clear whether this number is heart-warming or alarming. Nor is it clear whether it’s an increase or decrease, and you can’t guess the basis for comparison or the target.</p><p>A figure by itself is just trivia. It just makes your brain work hard. It isn’t useful because it isn’t compared with last month’s figure, for example, or a target. And there is no plain-language explanation to make it more meaningful.</p><p>A number alone, without context, provides no insight.</p><h3>2. The “Rainbow Coalition”</h3><p>You have used too many colors in the dashboard design. Every chart uses a different color scheme; your top 10 analysis bar chart has ten different colors; and your legend looks like a Home Depot paint swatch catalog. As a result, nothing stands out.</p><p>What is more, you haven’t used color consistently or purposefully.</p><p>Also, your color choices go against accepted conventions, like red symbolizing danger and green symbolizing growth.</p><p>In other words, the dashboard seeks to decorate rather than educate.</p><h3>3. The Buffet</h3><p>Your dashboard qualifies for the accolade “Information Overload.” It has far too many charts and widgets. If all aspects of your analytics are important, then none of them actually are.</p><p>The most useful dashboards have only five to six charts and a few carefully selected KPIs neatly arranged to fit into one screen without scrolling.</p><p>But your dashboard isn’t appealing because the elements seem to compete for attention and screen real estate.</p><p>Use discretion and leave certain things off the dashboard, just like you don’t have to sample every dish at the end-of-year buffet dinner.</p><h3>4. Time travel</h3><p>For example, one chart shows data over the last quarter. The second shows month-to-date, while a third shows the last 90 days.</p><p>As it is, you can’t compare one thing to another because the reference periods are different.</p><p>As you navigate across the dashboard, you keep jumping from one period to another, like time travel in a sci-fi movie.</p><h3>5. Nothing Jumps Out at You</h3><p>The charts are all the same size, forming a grid of equal parts. The fonts are also all the same size.</p><p>Imagine a newspaper front page. It starts with a large headline showing the main news item, followed by other news items, articles, and ads, deliberately organized to make it easy to scan and read. The font type, size, and color create an impactful reading experience.</p><p>Unlike a newspaper, your dashboard can’t be scanned in five seconds. Your eye has nowhere to land. No wonder even your colleagues do not find it attractive.</p><h3>6. Mystery Labels</h3><p>Your charts have titles that require deciphering. Titles such as ‘Performance’ or ‘Growth Over Time’ don’t mean much. What kind of performance or growth are we talking about?</p><p>You have axes labeled ‘Amount,’ but what units is the amount measured in: dollars, liters, kilograms, or something else?</p><p>The charts have titles no one really understands; the axis labels are vague or ambiguous. For instance, one axis is labeled “New MRR ($).” This immediately raises two questions.</p><p>First, what is MRR? Are you sure every member of your audience understands what it means?</p><p>Secondly, how do you define “New”?</p><p>A better label would be something like this: “New Monthly Recurring Revenue, Last 30 Days ($).”</p><h3>7. The Default Skin</h3><p>Your dashboard looks mediocre because it has “chart junk” begging to be cleaned. I am referring to default elements such as:</p><ul><li>fonts</li><li>colors that do not reflect the analytics</li><li>grid lines and zero lines</li><li>marks (ticks) and data labels</li><li>legends</li><li>background, spacing, padding, borders, etc.</li></ul><p>Your dashboard is a turn-off if it has titles and axis labels drawn directly from the data source, “customer_id,” or “business_unit,” for example. Unreasonable number formats, especially too many decimal places, are another serious defect.</p><p>The point I’m making is that your dashboard isn’t customized for the user because all the features are default.</p><h3>8. Half-Baked Information</h3><p>Your dashboard includes information that isn&#39;t ready for consumption. For example, a user first has to download information into a spreadsheet before they can access it.</p><p>Dashboards are intended for making decisions, not for storing information. So, if something has to be exported to make sense, don’t be surprised if users hate it.</p><h3>9. The Fancy Blunder</h3><p>Your charts are fancy but unfamiliar. Some of them require a learning curve. Above all, they are not appropriate for the visualization you want to convey.</p><p>I’m talking about three-dimensional bar charts, Chord Diagrams, Sunburst Charts, Donut Charts, and so on.</p><p>In the worst case, users ask why you didn’t use a simple bar chart instead of a speedometer dial used to indicate a metric that has nothing to do with speed.</p><h3>10. Dashboard Malfunction</h3><p>Some user interactions generate a nasty error message or a blank page.</p><p>Also, the dashboard “panics” and does outlandish things when it can’t find data, instead of showing a friendly message like “There were no sales in this region for this period.”</p><p>It is frustrating for dashboard users to encounter situations like these, where errors or missing data aren’t handled efficiently.</p><h3>11. The Lying Filter</h3><p>A rogue filter is on the loose. It operates unpredictably and erratically.</p><p>When you apply the filter, three charts update, two don’t, and one updates only partly. That’s a perfect way to destroy trust.</p><p>If this happens while presenting to the business, you can be sure that the show will surely come to an abrupt end.</p><h3>12. The Dead End</h3><p>This is the most serious demerit. Your dashboard only states what happened. Just that.</p><p>For example, the dashboard is screaming “Revenue decreased by 18% in Texas.” Well then? Why did it decrease? Where is it originating from? What am I supposed to do? How does that help me make a business decision?</p><h3>Recap and Key Takeaway</h3><p>It’s no wonder business leaders are disinclined to use your dashboard. Here’s why your dashboard didn’t fly.</p><ul><li>Numbers without context</li><li>Too many colors</li><li>Clutter and information overload</li><li>No uniform time period across charts</li><li>No visual hierarchy</li><li>Labels and titles are not meaningful.</li><li>No customization, only default features</li><li>Information repository, not a decision-making tool</li><li>Fancy but inappropriate charts</li><li>Errors not handled gracefully.</li><li>Defective filters or parameters</li><li>Providing information without actionable insights</li></ul><p>Fix these defects, and your dashboard will shine like a star.</p><p>Happy Dashboarding!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=77f5784ab769" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Needs Rules, Not Rebranding]]></title>
            <link>https://medium.com/@theobee.acquah/ai-needs-rules-not-rebranding-49f5f7171a9f?source=rss-9081f25673b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/49f5f7171a9f</guid>
            <category><![CDATA[responsible-ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai-ethics]]></category>
            <category><![CDATA[technology-policy]]></category>
            <category><![CDATA[regulation]]></category>
            <dc:creator><![CDATA[Theo Acquah]]></dc:creator>
            <pubDate>Wed, 30 Sep 2026 20:25:41 GMT</pubDate>
            <atom:updated>2026-09-30T20:25:41.832Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Bob relabels an AI machine “SI” while the machine performs its own safety audit. Priti watches skeptically as Jojo installs transparent guardrails and says, “A new name doesn’t make it accountable.”" src="https://cdn-images-1.medium.com/max/1024/1*Nddj2_vszzo8oOJnK0Krng.png" /><figcaption>New Label But No Guardrails? — TheoToon™ by Theo Acquah · Created with AI assistance</figcaption></figure><p>On September 29th, President Donald Trump met at the White House with some of the technology sector’s most influential figures. Executives from Google, Meta, OpenAI, Anthropic, Nvidia, and xAI signed what Trump called an AI ‘constitution’.</p><p>The document appears reassuring, as AI companies promise internal controls, safety teams, external audits, and board-level oversight. However, the agreement is voluntary, and Trump referred to it as “morally binding”, stating that the industry would “police” itself.</p><p>It isn’t a constitution; it’s just a promise.</p><p>At that event, Trump declared that the federal government would substitute ‘artificial intelligence’, or AI, with ‘Super Intelligence’, or SI. An executive order requires agencies to use the new term in their official communications and policy documents immediately.</p><p>Changing a technology’s name doesn’t make it any safer. “Super intelligence” sounds like an all-knowing authority rather than a category of flawed computer systems.</p><p>But that is the wrong mental model because we know AI often hallucinates, discriminates, manipulates, exposes private information, and fails in unpredictable ways.</p><p>What I’m trying to say is “Superintelligence” is a ridiculous misnomer, at least for now.</p><h3>No Guardrails” is Not a Serious Policy</h3><p>The idea that there should be ‘no guardrails’ is frequently presented as a justification for innovation. The reason given is that regulation will hinder American companies while China advances.</p><p>Leadership is not merely about speed; a country can also lead by establishing systems people can trust.</p><p>We do not oppose medical research when we regulate medicines, nor do we oppose air travel when we set aviation standards, nor do we oppose investment when we require financial disclosures. These regulations exist because some failures have serious consequences for people, including those who did not choose to take the risk.</p><p>Artificial intelligence already affects employment, education, credit, insurance, healthcare, policing, and people’s access to public services. If an automated system incorrectly denies someone a job, marks a student as having cheated, or causes an innocent person to be suspected, the injustice can be very hard to reverse.</p><h3>The Inherent Conflict in Self-Policing.</h3><p>Voluntary standards have value because companies understand their own systems and can respond faster than legislation can. It is worthwhile to keep internal safety teams, independent testing, and board supervision.</p><p>Yet self-policing means the same organization produces the product, sets the level of acceptable risk, chooses the evaluator, and determines what the public should learn.</p><p>Even the most careful companies face pressure to launch their products before their competitors. If a safety team can give advice but cannot delay a product release, it does not act as a safeguard. A person selected and paid by the company, with no assured access to the evidence and no obligation to make the findings public, is not properly independent.</p><p>A board committee reviewing confidential reports might protect the corporation without protecting the public.</p><p>The White House agreement does recognize the actual risks involved. Its flaw lies not in the requirements it places on companies, but in what happens when they fail to meet them. Nothing!</p><p>No obligation to disclose, no regulatory body with the power to investigate, no penalty proportionate to the harm caused, and no clear way for the individual whose life is disrupted by a defective system to seek redress.</p><p>Being morally binding does not mean it is legally enforceable.</p><h3>“Super Intelligence” Only Worsens the Debate</h3><p>The attempt to call AI “Super Intelligence” goes beyond a harmless branding exercise.</p><p>Present-day AI does not have super-intelligent capabilities. It is only a powerful pattern-matching technology that can carry out certain tasks very well while failing at others. It has no wisdom, moral judgment, or general understanding of the world.</p><p>To call all such systems ‘super intelligent’ overstates what they can do and obscures their limitations. It leads people to rely on machines when they should be asking questions such as: What data trained this system? How frequently does it make mistakes? Who gains from the recommendations it gives? And can a human override its decision?</p><p>Public policy requires precise language; otherwise, officials will be unable to distinguish between a school chatbot and a frontier model that can assist with cyberattacks or biological research. They thus will not be able to regulate either effectively.</p><p>We have to keep using the well-known term “artificial intelligence” and define systems by their capabilities and applications. There is AI for self-driving vehicles, but that is all it is. How can we call this superintelligence when this type of AI cannot perform any other human task?</p><h3>What Reasonable AI Regulation Could Look Like</h3><p>Washington does not need to approve every software update for regulation to be effective. However, it needs clear obligations when failure could seriously harm people.</p><p><strong>1. Start by dividing up AI according to the level of risk involved and its purpose.</strong></p><p>An AI system that recommends music should be subject to different rules than one that decides who gets a mortgage, who receives medical treatment, or who faces police scrutiny. Higher-risk applications should carry greater responsibilities.</p><p><strong>2. Independent testing should be required before any deployment.</strong></p><p>AI models should undergo standardized evaluations in cybersecurity, biological misuse, deception, and loss of control. High-impact decision systems should be tested for accuracy, fairness, and accessibility.</p><p><strong>3. Serious incidents should be reported by legal obligation.</strong></p><p>When AI causes a major data breach, dangerous model behavior, physical harm, or widespread denial of rights, companies should quickly inform the relevant regulator.</p><p>A confidential reporting system can protect information security while enabling regulators to deal with serious incidents.</p><p><strong>4. We must ensure that people who are the subject of automated decisions are protected.</strong></p><p>Individuals should be made aware when consequential AI is used, provided with a clear explanation, and given the opportunity to appeal to a qualified human. Certain decisions, such as criminal sentencing or whether a child is entitled to disability benefits, must never be fully automated.</p><p><strong>5. Introduce real accountability.</strong></p><p>Regulators should be empowered to examine evidence, require corrective action, impose substantial fines, and suspend any systems that pose an imminent danger.</p><p><strong>6. We should safeguard research and promote competition.</strong></p><p>The rules must guarantee access for independent researchers, protect whistleblowers, and avoid setting up compliance measures that only the biggest companies can afford.</p><p><strong>Above all, the rules should be reviewed periodically. </strong>Congress must establish firm regulatory legislation. Sunset reviews should continually fine-tune AI regulations, reinforcing effective ones and repealing ineffective ones.</p><h3>Guardrails Make Progress Possible</h3><p>The situation is not a matter of regulation versus innovation. It is a choice between accountable innovation and carrying out a large-scale experiment on the public.</p><p>The agreement reached at the White House marks the start of a regulatory checklist, and it is important not to eliminate internal controls, safety teams, external audits, or board oversight.</p><p>Instead, these should be made credible through legal obligations, true independence, public reporting, and enforcement.</p><p>America does not need to rename AI. It needs the confidence to govern it before the harm arrives, and the will to do so.</p><p>The choice is still available, but not for long. Act before it disappears.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=49f5f7171a9f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Your Analysis Found a Pattern, But Don’t Celebrate Yet.]]></title>
            <link>https://medium.com/@theobee.acquah/your-analysis-found-a-pattern-but-dont-celebrate-yet-240cb3692446?source=rss-9081f25673b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/240cb3692446</guid>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[statistics]]></category>
            <category><![CDATA[critical-thinking]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Theo Acquah]]></dc:creator>
            <pubDate>Sat, 26 Sep 2026 03:21:17 GMT</pubDate>
            <atom:updated>2026-09-26T03:21:17.696Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Priti tells Bob that employees who exercise more get promoted more often, prompting him to pedal harder on an exercise bike. Jojo challenges the conclusion with alternative explanations, such as ambition and flexible schedules. Bob jokes that Jojo is terrible for motivation; Jojo replies, “I’m excellent for methodology.” The Priti Moment: “Finding the pattern gives you a clue. Testing the explanation gives you the insight.”" src="https://cdn-images-1.medium.com/max/1024/1*S_M5acSh8Z4VD5OzMjDAYg.png" /><figcaption>What is the Right Intepretation for the Pattern? — TheoToon™ by Theo Acquah · Created with AI assistance</figcaption></figure><blockquote>“Correlation is not causation.”<em><br></em><strong><em> — A principle worth remembering</em></strong></blockquote><p>A wonderful moment in data analysis comes when you discover a pattern.</p><p>Sales rise with advertising spend. Customers who use a particular feature stay longer. Projects with larger teams finish later. Employees who complete training perform better.</p><p>You lean toward the screen. <em>“Aha!” </em>You’ve found something.</p><p>And that’s exactly when you have to be careful: discovering that two things go together is not the same as discovering that one caused the other.</p><p>The pattern may be real, but your explanation may not be.</p><h3>A Pair of Feet Plus a Pair of Wings May Not Make a Duck</h3><p>Imagine analyzing ice cream sales and drownings.</p><p>During some months, both increase dramatically. Plot ice cream sales in dollars and number of drownings on a chart, and you might discover a surprisingly strong relationship. Should we ban ice cream near swimming pools?</p><p>Probably not. Another factor is at work: <strong>Summer.</strong></p><p>Hot weather increases ice cream consumption. Hot weather also brings more people into the water. Ice cream didn’t cause the drownings. Another factor did. A third factor influenced both.</p><p>This example is deliberately trivial, just to help me illustrate a point.</p><p>In business data, though, <strong>the hidden factor isn’t that easy to find.</strong></p><h3>Business Data Is Full of Traps</h3><p>Suppose customers who use your company’s mobile app spend 30% more than customers who don’t. That seems to be an interesting insight, and you might conclude, “The app increases customer spending.” Maybe.</p><p>However, your most loyal customers are more likely to download the app. In that case, loyalty may affect both app usage and spending.</p><p>Now imagine the company spends heavily promoting the app because of your analysis. Downloads increase, but revenue stays flat.</p><p>The correlation (app usage moves with spending) was genuine. But the causal story (app usage is the reason for higher spending) was wrong.</p><p>That’s a noteworthy distinction.</p><h3>Ask: “What Else Could Explain This?”</h3><p>Whenever I see an interesting relationship in data, I like to ask: <strong>“What else could explain what I’m seeing?”</strong></p><p>Suppose projects with more people tend to finish later. One explanation is that <strong>larger teams cause delays.</strong></p><p>But another is that <strong>more difficult projects require larger teams.</strong></p><p>Or <strong>troubled projects get additional people after they start falling behind.</strong></p><p>It’s the same correlation at play, but three significantly different explanations. And each story suggests a completely different management response.</p><p>That’s why the analyst shouldn’t stop when the pattern appears. Instead, that’s when the investigation begins.</p><h3>Beware of the Story Your Brain Wants</h3><p>Humans are remarkably good at jumping to conclusions.</p><p>For instance, we see ‘Advertising’ and ‘Sales’ on the dashboard decorated by a beautiful, bold, green, upward-pointing arrow, and our brain immediately registers: <strong>Advertising caused sales to increase.</strong></p><p>It seems obvious, or it feels intuitive. It may be correct.</p><p>But what if advertising increased simply because it was the holiday season? In that case, seasonality drove both advertising and sales, as it always does.</p><p>Or did a major product launch happen at the same time?</p><p>Did competitors raise their prices, prompting corresponding price increases to keep up?</p><p>The problem is not that we tell stories. The problem is that the first plausible story commonly feels like the only one.</p><p>Good analysis requires us to resist that temptation.</p><h3>Segment Before You Conclude</h3><p>One useful technique is to break the relationship apart.</p><p>If customer satisfaction and retention move together overall, ask:</p><p>Does the relationship hold for new customers, long-term customers, large accounts, different regions, different products, etc.?</p><p>Sometimes a strong overall pattern behaves differently when you segment the data. Sometimes it gets stronger. Other times it reverses. Each outcome teaches you something.</p><p>The purpose is not to complicate the analysis. It’s to confirm that your explanation not only makes sense but is also defensible.</p><h3>Ask What Happened First</h3><p>Time matters.</p><p>If A supposedly causes B, A generally needs to occur before B. That is simple logic. Surprisingly, business analysis habitually overlooks this simple logic.</p><p>Suppose employees who receive promotions are more engaged. Can we say that engagement leads to promotion, or that promotion increases engagement? Or are high performers both more engaged and more likely to be promoted?</p><p>A single snapshot may not tell you. So sometimes you need longitudinal data, an experiment, or more evidence.</p><p>And sometimes the most responsible conclusion is: <strong>“We don’t know yet.”</strong></p><p>That is not weak analysis. That’s intellectual discipline.</p><h3>AI Makes Plausible Explanations Cheap</h3><p>This matters even more in the age of AI, because if you give an AI system a correlation and ask it to explain the relationship, it can quickly generate several polished, convincing possibilities.</p><p>That is useful for brainstorming, but a beautifully written explanation is only a hypothesis until the data supports it. AI can help us generate answers, but we still need to test those answers.</p><h3>Key Takeaway</h3><p><strong>A pattern is an invitation to investigate — not a conclusion to celebrate.</strong></p><p>Correlation is useful because it helps us identify relationships worth exploring.</p><p>But the moment we turn <em>“these things move together”</em> into <em>“this caused that,”</em> we cross an important analytical line. We need evidence for that step.</p><h3>Try the Three-Explanation Test</h3><p>When you next find an exciting relationship in your data, do not explain it immediately. First, force yourself to write down <strong>three possible explanations</strong>.</p><p>For example:</p><p><strong>A causes B.</strong></p><p><strong>B causes A.</strong></p><p><strong>Something else causes both.</strong></p><p>Then ask: What evidence would distinguish between them?</p><p>This small exercise can keep you from getting too attached to your first interpretation.</p><p>And sometimes explanation number three will be far more interesting than number one or two.</p><h3>Final Thoughts</h3><p>Finding patterns is one of the joys of analytics. It is the moment the numbers begin to feel alive.</p><p>But good analysts know when to resist excitement.</p><p>They don’t ask only: <strong>“What did I find?”</strong></p><p>They also ask: <strong>“What else could explain it?”</strong></p><p>The difference may seem small, but it is not.</p><p>One produces an interesting chart while the other moves us nearer to understanding.</p><p>Celebrate the pattern, but just don’t settle on the explanation yet.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=240cb3692446" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Data Doesn’t Speak for Itself]]></title>
            <link>https://medium.com/@theobee.acquah/the-data-doesnt-speak-for-itself-72ac8fabe3d7?source=rss-9081f25673b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/72ac8fabe3d7</guid>
            <category><![CDATA[data-storytelling]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[critical-thinking]]></category>
            <category><![CDATA[data-analytics]]></category>
            <dc:creator><![CDATA[Theo Acquah]]></dc:creator>
            <pubDate>Mon, 21 Sep 2026 04:13:24 GMT</pubDate>
            <atom:updated>2026-09-21T04:13:24.373Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*O_rHX3kGklzJgWpD34VTAg.png" /><figcaption>Data Doesn’t Speak — AI-created image</figcaption></figure><blockquote>“The facts are meaningless if they are not interpreted.”<em><br></em><strong><em> — Unknown</em></strong></blockquote><p>Let us examine a cliché that appears in almost every analytics conversation: <strong>“The data speaks for itself.”</strong></p><p>I don’t think it does.</p><p>Data needs context, interpretation, skepticism — and someone willing to explain what it means.</p><p>It sounds sensible and objective. It may even sound scientific. However, there is one problem: <strong>Data doesn’t speak.</strong></p><p>Data just sits there, manifesting itself in rows, columns, numbers, categories, dashboards, spreadsheets, reports, etc.</p><p>Somebody must decide which data to look at, which metric to calculate, which comparison to make, which chart to produce, and which conclusion to focus on.</p><p>By the time a number reaches a presentation, humans have already made dozens of decisions about it.</p><p>That is why good analytics is not simply identifying facts. It is about interpreting them responsibly.</p><h3>One Number, Three Stories.</h3><p>Picture the revenue rising by 15% this quarter. Is that good news? Perhaps.</p><p>The Sales Director might confidently declare, “Our new sales strategy is working.”</p><p>Finance might say, “The main part of the increase came from higher prices,” whereas Operations quickly points out, “Order volume barely changed.”</p><p>It is possible for all three statements to be true. It’s the same data but different perspectives.</p><p>Now imagine that we discover a single high-profile customer made a very large purchase during that quarter. The story changes again at this point, but revenue still increased 15%. The number didn’t change, but <strong>our understanding did.</strong></p><p>That is the difference between data and insight.</p><h3>Every Chart Involves Decisions</h3><p>Imagine a simple line chart that displays monthly sales. No matter how objective it looks, someone decided:</p><ul><li>Where the timeline begins and ends.</li><li>Whether the vertical axis starts at zero.</li><li>Whether to publish the results on a monthly or quarterly basis.</li><li>Whether to include last year’s comparison.</li><li>Whether or not to allow for inflation.</li><li>Whether to separate products or combine them.</li></ul><p>Those decisions keep the chart from being misleading. Besides, they shape what the reader sees. Visualization is more than showing data. It frames<strong> information</strong>. And framing has to be done responsibly.</p><h3>Context Changes Everything</h3><p>Suppose a dashboard reports: <strong>Customer churn: 8%. </strong>Is that good or bad? We don’t know.</p><ul><li>What was it a month ago?</li><li>What was it last year?</li><li>What is considered normal within the industry?</li><li>Which customers are leaving?</li><li>Are high-value customers leaving?</li><li>Has the company recently altered its prices?</li></ul><p>8% is just a number without context.</p><p>Add a comparison, and the statement gains meaning. We could be in trouble if the churn rate was 5% last month. If the figure was 12 percent, we might be making progress. If competitors average 4 percent, we might be celebrating too soon.</p><p>The figure (8%) stays the same until the context changes the story.</p><h3>The Analyst Is Part of the Analysis</h3><p>We often think analysts find truths already contained in the database.</p><p>The analyst chooses questions, makes assumptions, defines metrics, selects comparisons, and decides what gets attention. Doing all that takes good judgment.</p><p>A good analyst should therefore be willing to ask:</p><ul><li><strong>What am I assuming?</strong></li><li><strong>What kind of context may be missing?</strong></li><li><strong>Could that be interpreted another way?</strong></li><li><strong>What kind of evidence would alter my conclusion?</strong></li></ul><p>Asking these questions is not a sign of uncertainty. It is a sign of professionalism.</p><h3>Storytelling Doesn’t Mean Spinning the Data</h3><p>The idea of <em>data storytelling</em> can sometimes make people feel uncomfortable. The reason is easy to see. Stories sound subjective while Data sounds objective.</p><p>Storytelling has nothing to do with adjusting the figures to match the story we want. It is the opposite. It’s about helping people grasp what the evidence really supports.</p><p>If the data doesn’t provide a clear answer, state that.</p><p>If an outlier matters, show it.</p><p>When two explanations are plausible, you should recognize both.</p><p>When the evidence contradicts your initial hypothesis, you should revise the story.</p><p>Resist the temptation to do otherwise in all these situations.</p><p><strong>The analyst’s loyalty must be to the evidence, not to the story it tells.</strong></p><h3>AI Makes This Even More Important</h3><p>AI can now summarize datasets, detect patterns, produce visualizations, and give remarkably confident explanations.</p><p>That is useful. But AI doesn’t eliminate interpretation. It only adds another interpreter.</p><p>Someone still needs to ask:</p><p><strong>Why did the AI highlight this pattern?</strong></p><p><strong>What did it ignore?</strong></p><p><strong>What assumptions underlie this conclusion?</strong></p><p><strong>Does it make sense from our business perspective?</strong></p><p>The simpler the story, the more essential it is to tell it responsibly.</p><h3>Key Takeaway</h3><p><strong>Data offers evidence. Context gives rise to meaning. Judgment links the two.</strong></p><p>That is why two capable analysts, looking at the same dataset, will notice different things. One may focus on growth, the other on risk.</p><p>The aim is not to pretend interpretation doesn’t exist. At the end of the day, different interpretations help make our interpretation transparent and justified.</p><h3>The “Other Story” Test</h3><p>The next time you finish an analysis, pause before presenting it. Ask yourself: <strong>Could the same figures plausibly account for something else?</strong></p><p>If so, try to build that case. If your original conclusion survives, your analysis is stronger. Then you’ll have discovered an important thing before your audience does.</p><p>Either way, you win.</p><h3>Final Thoughts</h3><p>The idea that data speaks for itself is reassuring because it takes us out of the situation. But we are never actually absent.</p><p>Someone has to do all of the following: ask the question, choose the metric, build the chart, and explain what it means.</p><p>And a decision-maker eventually decides.</p><p>Our duty as analysts is not to explain the data in a way that corresponds to what we already think.</p><p>Instead, we give the evidence enough context for people to understand what it says, because data does not tell its own story. <strong>We speak for it.</strong></p><p>And we’d better choose our words carefully.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=72ac8fabe3d7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Average Is Lying to You]]></title>
            <link>https://medium.com/@theobee.acquah/the-average-is-lying-to-you-ae508da9a9ab?source=rss-9081f25673b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/ae508da9a9ab</guid>
            <category><![CDATA[statistics]]></category>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[decision-making]]></category>
            <category><![CDATA[data-visualization]]></category>
            <category><![CDATA[data-analytics]]></category>
            <dc:creator><![CDATA[Theo Acquah]]></dc:creator>
            <pubDate>Tue, 15 Sep 2026 03:00:42 GMT</pubDate>
            <atom:updated>2026-09-15T03:00:42.495Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Bob looks at his phone and says, “Good news, guys. Our average project is only five days late.” Priti smiles: “That’s not bad!” Jojo: “How many projects are on time, Bob?” Bob checks again. “Nine out of ten.” Priti: “See? Even better!” Jojo: “And the tenth?” The elevator doors open. Bob, staring at his phone: “…47 days late. Priti’s expression changes completely. Then Bob looks at Jojo: “You’re really not going to let me enjoy this elevator ride, are you?” Jojo, stepping out: “I was enjoying" src="https://cdn-images-1.medium.com/max/1024/1*-Ul8BiPWkX9VLV3GOM2hBA.png" /><figcaption>The Average Just Summarizes — AI-gernerated image</figcaption></figure><p>The average is one of the first statistical concepts most of us encounter. You calculate the total, then divide it by the number of observations. Simple and useful! Yet the average can be very misleading.</p><p>Imagine you are informed that customers wait an average of six minutes for service. That might seem rather impressive. However, when you go to the shop, you find that about half of the customers are attended to almost immediately, while the other half just stand there, wondering whether anyone even knows they are there.</p><p>Six minutes on average is mathematically accurate, but it refers to a waiting time that very few people have actually experienced. And this is one of the major pitfalls of analytics: <strong>an average can be correct even if it hides the real situation.</strong></p><h3>The Convenient Middle</h3><p>We like averages because they simplify things. For example, rather than looking at 10,000 customer transactions, we use the average dollar amount. Instead of considering hundreds of salaries, we take the average salary. Rather than checking each delivery time, we use the average delivery time.</p><p>A single number replaces thousands, which is useful, but representing a distribution with a single number has a cost. Each time we reduce a distribution to one number, we lose information.</p><p>More often than not, what we discard is exactly what matters.</p><h3>Tale of Two Companies</h3><p>Imagine Company A has five employees earning:</p><p>$60,000<br>$62,000<br>$64,000<br>$66,000<br>$68,000</p><p>The average salary is $64,000.</p><p>Now imagine Company B also has five workers:</p><p>$30,000<br>$35,000<br>$40,000<br>$45,000<br>$170,000</p><p>The average amount is also $64,000.</p><p>These are two distinct workplaces (at least in terms of salary distribution), but the average salary is the same in both cases. If you consider only the average salary, you would overlook almost all the interesting aspects of either company.</p><p>That is why experienced analysts don’t settle for simply asking, “What is the average?” Instead, they ask, “What is beneath it?”</p><h3>Meet the Median</h3><p>In some cases, the median is a more appropriate measure because it is the middle value when the values are arranged from smallest to largest. The median salary at Company B is $40,000.</p><p>This median salary presents a completely different picture and raises at least two important questions. What is the minimum salary? And what is the maximum?</p><p>No single measure — mean or median — is automatically the correct way to represent the data, since each answers a different question, which is not surprising given how they are calculated. The average includes all salaries, while the median is just the middle value when salaries are ranked.</p><p>The median provides a more accurate indication of the data’s disparity. When extreme values are present, it better represents the data.</p><p>If you look only at the average, these two companies seem similar; but when you consider their medians, you can tell them apart at once.</p><h3>But Even the Median Can Hide the Story</h3><p>This is where things get even more interesting.</p><p>Suppose that a company’s average customer satisfaction score is 8 out of 10 and the median is also 8; this seems healthy in every respect.</p><p>Yet different parts of the business tell different stories. For instance, if new customers average 9.4, long-term customers average 6.1, enterprise customers are declining rapidly, and in one case the figure has even fallen below 5, the overall rating of 8 out of 10 still seems reassuring, though it may be masking a serious business problem.</p><p>The valuable insight doesn’t always lie in the mean or the median; it can be found in the differences between groups.</p><p>That is why it is important to segment. You should always examine each data segment for insights.</p><h3>Ask One More Question</h3><p>Whenever someone presents an average, I’ve developed a mental habit<strong>.</strong> I ask<strong>, “Average of what — and how is it distributed?”</strong> Here are some examples.</p><p>Average revenue per customer: <em>Which customers?</em></p><p>Average delivery time: <em>Across which locations?</em></p><p>Average employee satisfaction: <em>Which departments?</em></p><p>Average project margin: <em>Across which project types?</em></p><p>Asking one more question usually starts a whole new line of conversation and yields further insights.</p><h3>The Executive Dashboard Problem</h3><p>Averages are especially appealing on dashboards.</p><p>Managers value simplicity, and we therefore don’t include too many details. For this reason, we produce elegant KPI cards such as ‘<strong>Average Resolution Time: 4.2 Hours&#39; </strong>with a bold, fancy green arrow alongside.</p><p>All appears to be in order; in about 80% of cases, the problem is resolved within an hour, whereas a small number of serious cases take three days. Maybe the end of the distribution is where the company’s biggest customer-service risk lies!</p><p>The KPI wasn’t misleading, but it still failed to tell the whole truth.</p><p>Good dashboards simplify information. But excellent dashboards simplify information without concealing what matters<strong>.</strong></p><h3>Don’t Replace Averages. Interrogate Them.</h3><p>I am not arguing that we should give up on averages, since that would be absurd; they are among the most useful tools in statistics and analytics.</p><p>Rather, use the average as a starting point for your investigation, not as the final answer. When this is important, consider the following:</p><p><strong>Median</strong>: What information does the middle observation give us?</p><p><strong>Range: </strong>What is the distance between the two extremes?</p><p><strong>Distribution: </strong>Are the values clustered, skewed, or grouped?</p><p><strong>Segments: </strong>Do the important groups behave differently?</p><p><strong>Outliers: </strong>Are a few extreme observations changing the overall picture?</p><p>Each of these can put the average in the right perspective.</p><h3>Key Takeaway</h3><p><strong>An average summarizes the data. It doesn’t necessarily tell the whole story.</strong></p><p>In a business context, this distinction matters because businesses don’t deal in averages. Individual customers leave, individual projects fail, individual stores underperform, and individual transactions become exceptions.</p><p>But the average helps us see the forest. The treasure is sometimes concealed behind a specific group of trees.</p><h3>Try This With Your Next Analysis</h3><p>The next time you work out an average, don’t leave it at that — ask yourself three further questions instead.</p><p><strong>What’s the median?</strong></p><p><strong>What does the distribution look like?</strong></p><p><strong>What happens when I divide the data?</strong></p><p>It might still take five more minutes, but during those five minutes your conclusion could change entirely.</p><h3>Final Thoughts</h3><p>An average can be true yet still mislead. It may be incomplete, which is one reason analytics needs more than calculation; it also demands curiosity.</p><p>The average could be entirely accurate, and the dashboard will display exactly what you asked it to display. However, accuracy alone is not sufficient. A figure may be correct without conveying the most important information.</p><p>The most important discovery in the data could still be hidden in the rest of it, ready to be noticed when someone asks, “But what’<strong>s really going on here?”</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ae508da9a9ab" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>