Verify Calculations and Charts in an AI Dashboard

Updated 5 November 2026 · Standard formula, shown below

Checking calculations and charts in an AI-generated dashboard helps catch wrong totals, misleading comparisons, and mismatched labels before anyone relies on the results. You will need access to the dashboard, its underlying data or a trusted source, and enough context to understand what each metric is meant to show.

What You’ll Need

Permission to view the dashboard and, if possible, its source data, query, or calculation definitions.
A trustworthy reference for comparison, such as a source report, spreadsheet, database export, or independently calculated sample.
The dashboard’s date range, filters, units, refresh time, and definitions for key metrics.
A calculator or spreadsheet for checking a few values independently.

Step-by-Step

Clarify what the dashboard is supposed to answer. Write down the decision or question it supports and identify the key measures, such as revenue, count of orders, average response time, or conversion rate. Find out how each measure is defined. For example, “customers” could mean unique people, accounts, or transactions; these are not interchangeable. If definitions are missing, ask the dashboard owner or data team before judging a result.
Record the dashboard’s scope and settings. Note the selected dates, time zone, filters, segments, and any excluded records. Check whether dates refer to the event date, creation date, or another field. Look for a last-refreshed timestamp and confirm the source data covers the same period. A correct calculation over the wrong date range is still the wrong answer for your purpose.
Trace each important number to its source. Open the metric details, query, dataset, or export if available. Identify the source columns and the operations applied, including joins, filters, grouping, and deduplication. Check that joins do not multiply rows and that excluded or missing values are handled intentionally. If the dashboard only shows a final figure, request its definition rather than assuming what it includes.
Recalculate representative values independently. Choose a small sample, including ordinary records and edge cases such as a zero value, a blank, a duplicate, or a boundary date. Reproduce the calculation outside the dashboard from source rows. For a rate, verify numerator and denominator separately before dividing; for an average, confirm which records are included. Compare at full precision where possible, then account for display rounding.
Check totals, subtotals, and relationships. Add visible categories and compare their sum with the displayed total. Determine whether categories should be mutually exclusive and exhaustive; overlapping categories can legitimately add up to more than the total. Check basic relationships, too: a percentage should use the stated denominator, and a cumulative total should not unexpectedly fall unless the definition or data permits it.
Verify every chart’s encoding. Read the title, subtitle, axis labels, units, legend, and tooltip. Confirm the chart type fits the data: a line chart usually communicates ordered time trends, while a bar chart compares categories. Check that axes begin and end appropriately, category order is sensible, and color meanings are consistent. A truncated axis can exaggerate differences; a dual axis can make unrelated series appear to move together.
Compare chart points with the underlying values. Select several points or bars and compare their tooltips or exported values with the independently checked figures. For time series, inspect the first and last dates, a peak, a low point, and a point around a gap. Ensure labels and tooltips use the same units and rounding as the metric definition. Watch for missing dates being treated as zero or connected across gaps without explanation.
Test filters and interactions. Change one filter at a time and observe which cards and charts update. Confirm that totals, percentages, and labels respond consistently, and that a cleared filter restores the expected overall view. Test combinations that users are likely to choose. If a chart ignores a filter or a card changes unexpectedly, record the exact settings and report the issue.
Check against a second trusted reference. Compare at least one important total and one chart trend with a separate report or source-system view using the same definitions and period. Differences may result from refresh timing, time zones, exclusions, or rounding, but they should be explainable. Do not treat an AI-generated explanation as evidence that a number is correct; verify the underlying calculation and records.
Document results and escalate discrepancies. Keep a brief verification log with the metric, settings, expected value, displayed value, source, and difference. Include a screenshot or a safe sample of relevant rows when useful, avoiding unnecessary sensitive data. Mark the dashboard as unverified or flag affected metrics until the owner resolves a material discrepancy. Recheck after a data or formula change.

Common Mistakes

Comparing figures with different date ranges, time zones, filters, or refresh times.
Assuming a label such as “total,” “active,” or “average” has an obvious definition.
Checking only the displayed rounded number instead of the underlying precision.
Adding categories that overlap, or assuming blanks and duplicates are handled the same way everywhere.
Trusting a visually polished chart without checking its axis, units, missing values, or source figures.
Changing several filters at once, which makes it hard to identify what caused a difference.

Tips

Prioritize metrics that affect decisions, then verify a representative sample rather than manually checking every row.
Use the same filters and definitions in the dashboard and reference calculation; write them down so another person can repeat the check.
Keep unrounded values for arithmetic, and round only when presenting the result.
Ask the owner to document metric definitions, data sources, refresh timing, and known exclusions.
Recheck after changes to source data, formulas, joins, chart settings, or filters.

Troubleshooting

The dashboard and reference total differ: Align the date range, time zone, filters, refresh time, and record inclusion rules first. Then compare source rows and inspect joins, duplicates, and null handling. If the gap remains, provide the metric definition and a reproducible example to the dashboard owner.

A chart does not match its headline metric: Check whether the chart uses a different aggregation, date field, filter, or unit. Inspect tooltips and chart-level settings. If these agree but the displayed values still differ, capture the view and report the issue rather than guessing.

A value changes when a filter is applied: Confirm whether the filter is intended to affect that card or chart and whether its denominator should change. Test one filter at a time, then reset the dashboard. Report controls that appear selected but do not affect the expected visuals.

There is no source data or calculation detail: Ask for a read-only export, metric definition, or a documented query from the owner. Until you can reproduce a high-impact figure, treat it as unverified and avoid presenting it as confirmed.

Frequently Asked Questions

How many values should I verify?
Start with the highest-impact metrics and check several representative points, including edge cases. Expand the sample if you find an error or if the data is inconsistent.

Should I compare rounded or exact values?
Compare exact values when available, then check whether the dashboard rounds them correctly for display. Small display differences may be harmless; differences in the underlying calculation need explanation.

What if the source data updates during my check?
Record the dashboard and source refresh times, then compare snapshots from the same period if possible. Otherwise, repeat the comparison after both have refreshed.

Can a chart be misleading even when its data is correct?
Yes. Axis ranges, inconsistent units, color choices, missing periods, or overlapping categories can create a misleading impression despite accurate values.

What should I do if I cannot reproduce a metric?
Request its definition, source fields, filters, and aggregation from the dashboard owner. Document what you could verify and mark the metric unconfirmed until the calculation can be traced.

Method: the standard formula for this calculation, rounded as each result states. See how the calculators work or report an error.