Your AI Chart Needs to Show Its Math

Key takeaways

• Conversational analytics makes it easier to follow one question with another.
• Metric definitions and filters can change the story a chart tells.
• Charts need visible data sources and calculation rules so people can check them.
• Useful discoveries can become dashboard metrics once their definitions are reviewed.

Ask whether revenue grew last month, and 10% and 0% could both be correct answers. Putting the chart in an AI conversation makes asking easier, but the answer still depends on what “revenue” means.

One question, two perfectly valid charts

Consider a hypothetical online store. In July, it collects ₩100 million and issues no refunds. In August, it collects ₩110 million but refunds ₩10 million.

Measured by payments collected, revenue grew 10%. After refunds, growth was 0%.

Both calculations work. A chart labeled “Monthly Revenue” leaves the reader guessing which one they are looking at.

Even a simple request carries a surprising amount of unfinished business. Does revenue include refunds? Which exchange rate applies to overseas orders? Does a comparison between this month and last month cover the same number of days?

Those choices can change the chart’s shape and the decision someone makes from it.

A useful assistant should clarify the assumptions that materially affect the answer. If the company already has an agreed revenue definition, it can use that definition and display it beside the chart. Nobody needs an interrogation before every bar chart, but consequential choices should be visible.

The real payoff is the next question

Conversational analytics lets people query data in ordinary language. Its appeal becomes clearer once the first chart prompts a follow-up.

An investigation might run like this:

“Show last month’s revenue by week.”
“Break down the second week by sales channel.”
“Show order counts for the channels where revenue fell.”

Each answer helps shape the next question. Someone can investigate an unexpected dip without planning the entire analysis upfront.

For that to work, the assistant needs to carry context forward. The third request should still refer to the second week of last month, unless the user changes the scope.

It also needs to show that it has done so.

A visible summary such as “Last month · Week 2 · Channels with declining revenue” lets the user check the current conditions. Otherwise, remembering context becomes another assumption hidden inside the interface.

Put the calculation within reach

A fluent explanation can make a chart easier to understand. It provides no guarantee that the underlying calculation is correct.

Readers need access to a few basic details:

Start with plain language: “Canceled orders excluded. Grouped by payment date. Daily totals.”

For people who need more detail, the database query should be available to expand and inspect.

Even then, showing SQL does not settle the question. Joining tables can accidentally count the same order more than once. Delayed data can make recent sales look lower than they really are.

Two checks matter: Can someone reproduce the calculation? And does it answer the intended question?

That context also needs to survive sharing. A colleague receiving a chart in a slide deck should be able to understand its scope without reading the conversation that produced it.

Give recurring metrics a permanent home

Dashboards are useful for checking established metrics consistently. Conversation is useful for exploring the questions those metrics raise.

The two can form a practical workflow. Spot a decline in the weekly revenue dashboard. Investigate customer types and sales channels through conversation. If the exploration produces a metric worth tracking, review its definition before adding it to the dashboard.

That review matters. An improvised definition of “active customer” can easily reach the next meeting dressed as an official company metric.

Record the definition and who reviewed it. Save the calculation rules alongside the chart so future updates use the same basis.

Otherwise, a familiar-looking chart can quietly begin answering a different question.

Make the numbers easy to question

Conversational charts can help more people investigate data for themselves. For those answers to become useful at work, checking the definitions and calculations needs to be just as straightforward as asking the question.

Before sharing the next chart, make sure its recipient can find the answer to one more question: “How did you get that number?”