AI infographics: five checks before presenting your numbers
AI infographics: five checks before presenting your numbers
Your team is preparing its sales review. The infographic says “sales up 20%”, with a large green arrow. It looks ready for the meeting. But incoming enquiries also rose by 20%. Did the team convert more effectively, or simply handle more prospects? This is an illustrative scenario, not a client case. A polished chart cannot answer the question on its own.
Before approving an AI-assisted infographic with numerical claims, check its source, denominator, scale, conclusion and access to the underlying detail. AI can help organise the presentation. Approval needs to cover what the reader will understand and decide, not just how the layout looks.
1. Do the numbers describe the same activity?
Before discussing colours, ask where the figures came from and when they were extracted. “Sales” might mean signed orders, delivered orders or contracts that customers can still cancel. Two teams can use the same label while counting different events.
Put the observation period, scope and metric definition on the presentation. Comparing a complete month with a month still in progress needs a clear qualification or a new, like-for-like comparison. If the source has been corrected since the export, yesterday’s finished graphic is no longer the version to present.
A connected dashboard still needs this check. Microsoft’s Power BI refresh documentation distinguishes data refresh from refreshing report visuals. An open screen, by itself, does not prove that every value reflects the latest state of the source.
2. What is the denominator behind the percentage?
In our fictional example, the team moves from 100 enquiries and 20 sales to 120 enquiries and 24 sales. Sales volume has indeed increased by 20%. The conversion rate remains 20%.
| Metric | Period A | Period B |
|---|---|---|
| Enquiries received | 100 | 120 |
| Sales | 20 | 24 |
| Conversion rate | 20% | 20% |
The headline changes the decision. “Our new sales method converts better” is not established by these figures. “More enquiries, stable conversion” describes them without overclaiming. Leadership can then examine sales capacity rather than prematurely rolling out a supposedly superior method.
For every percentage, ask for the numerator and denominator. For every average, ask what has been grouped together: an overall average may hide deteriorating performance in a particular channel or customer segment.
3. Does the chart give an honest impression of the difference?
Once the calculation checks out, inspect the geometry. A bar chart with an axis starting just below the values can make a small change look dramatic. The Office for National Statistics recommends a zero baseline for bar and area charts, where length or area represents the value.
That does not mean every axis must always start at zero. A narrower scale can make variation easier to read in a line chart or scatter plot; the scale still needs to be explicit. Also check whether adjacent charts use different scales without making the difference clear.
When creating an AI-assisted infographic, keep the numbers and chart in editable, checkable elements. If the output is a flattened image, reconcile every displayed value against the approved table. Readers should not have to guess whether the length of a bar encodes a quantity or is merely decorative.
4. Does the headline describe a result or invent its cause?
“Sales rose after the redesign” describes a sequence. “The redesign increased sales” assigns an effect. The second statement needs evidence that the first does not provide. An additional campaign, seasonality, a price change or improved product availability might also explain the increase.
Separate three things in the presentation: the observed fact, the proposed explanation and the recommended decision. For example: “Conversion is stable despite more enquiries. Hypothesis: the sales team is under greater pressure. Check: first-response time and unhandled enquiries.” Leadership can see what is known and what still needs investigation.
If this summary is produced regularly, our approach to measurable visual content helps organise review, distribution and measurement. The check here is narrower: the headline and caption must remain accurate even when someone reads them without the rest of the presentation.
5. Can someone check the message without deciphering the image?
A tiny source note is not enough. For complex images such as charts, the W3C recommends a short alternative and an appropriate detailed description. Depending on the format, provide a nearby explanation of the trends and an accessible table of the relevant values.
Test the version people will actually receive: a page on a phone, a PDF or a projected slide. Are the labels still readable? Is colour the only way to distinguish the series? Can recipients find the period and source without chasing a missing document? A clear table also helps the person checking the calculations before the meeting.
How do you repeat the checks without starting from scratch?
For a one-off presentation, a template and a named approver may be enough. Creation tools such as Adobe Express, with its customisable infographic templates, help with composition. That capability does not certify your data or conclusions.
When the same presentation is needed every week, the Say Digital Framework helps scope the production chain: identify the decision, prototype the visual using an approved extract, verify the values and labels, then build repeatable production. Testing and CI/CD — checks accompanying changes before release — become relevant when a connected component is developed. Roll it out gradually, tracking errors, review time and actual use.
The budget for a UI designer should therefore cover a template that prevents misreading, not just an attractive first version. A monthly meeting using a fixed set of numbers may only need a dated presentation. Managers who need to filter by channel and act daily have a stronger case for a dashboard.
Before circulation: approve, qualify or send back?
- Approve when the source, period, calculation, visual and conclusion agree.
- Qualify when the numbers are correct but the explanation is still a hypothesis: revise the headline and make the limitation visible.
- Send back when the source is missing, the populations are incompatible or the chart distorts the difference: correct it before circulation.
Need reporting visuals your team can trust? Say Digital can scope a first template covering reference data, visual hierarchy, checks and approval responsibility. Tell us what your team needs to produce and which decision it needs to support.
Sources and scope
Sources consulted on 21 October 2026: ONS axis guidance, the W3C complex-images tutorial, Microsoft refresh documentation and Adobe Express’s official product page, linked above. The five checks are Say Digital’s editorial synthesis, not an official certification. The sales example is fictional; no client performance gain is claimed.
