How to Audit Your Sales Metrics in One Afternoon

What triggers a sales metrics audit?

Two reports disagreeing in public. The moment a board deck shows a 31% win rate and the sales dashboard shows 27%, every number in the stack becomes negotiable, and the meeting turns into a discussion about the report instead of the business.

The disagreement is almost never a bug. It is two defensible definitions computed by two teams who never compared notes. One report counts all closed opportunities, another excludes deals disqualified before stage two, a third counts only opportunities that reached proposal. Each is reasonable. Only one can be the company number.

Audit quarterly, and audit immediately after any CRM schema change, reporting tool migration, or sales process redesign. All three break definitions silently, and the break surfaces in front of an audience.

Step one: which metrics are actually published?

Inventory every metric that appears in a recurring report, not the metrics that exist in the warehouse. The list is longer than expected and that length is the first finding.

Walk every artifact that goes out on a schedule. The weekly forecast deck, the monthly business review, the board package, the rep scorecard, the marketing pipeline report, and the customer success health review. Write down every metric name that appears in any of them, along with which artifact carries it.

Metrics appearing in exactly one artifact are usually fine. Metrics appearing in three or more with slightly different names are where the disagreements live. "Win rate", "close rate", and "conversion rate" showing up in three decks usually means three calculations.

Step two: does each metric have a definition anyone can read?

Write the plain-language definition and the exact calculation side by side, then check they describe the same thing. They diverge more often than they should. The filters row catches the most defects. Silent filters, meaning conditions applied inside a report that nobody wrote down, are the leading cause of two reports disagreeing. Excluded record types, excluded deal sources, and date ranges anchored to different fields all qualify.

Step three: do the numbers reconcile across reports?

Pull the same metric from every artifact for the same period and put the values in one column. Any difference is a finding, and one traced difference usually exposes several more.

Reconcile the levels too. Sum the rep-level values and compare to the team figure, then sum the teams and compare to the company figure. Ratios do not aggregate, so a company win rate derived by averaging team win rates will not match a company win rate computed from raw counts, and the gap can run into double digits.

Close-date anchoring is worth an explicit check. A report filtered on the close date field and a report filtered on the fiscal period field will disagree whenever deals slip, and slippage is constant. In ORM customer data, a rep changing the close date is the strongest slippage signal, which means the field moves frequently by design.

Step four: is the sample large enough for the precision shown?

Check the denominator behind every ratio published at a granular level. Rep-month win rates on three closed deals are being reported to a precision the data cannot support.

Divide 100 by the closed-deal count to get what one deal is worth in points. At 10 deals, a single outcome moves the rate 10 points. Publishing that number next to a rep with 40 closes invites a comparison the sample cannot onlyify.

Add a suppression rule to the register during the audit. Below the minimum sample, show raw counts instead of a percentage. It costs one line of logic and removes an entire class of false conclusions from the weekly meeting.

Step five: does anyone act on it?

Every metric in the register needs a named owner and a threshold that triggers a specific action. Metrics without both are reporting overhead, and the audit is the moment to say so.

Ask the owner what value would cause them to do something and what they would do. If the answer is vague, the metric is decorative. If there is no owner at all, that absence is stronger evidence than any analysis, since a number nobody is accountable for has never changed a decision.

Assign or retire. Both outcomes are fine. What is not fine is leaving a metric in the board pack with no owner, because it will eventually generate a question that no one can answer.

Step six: what do you do about messy data?

Check consistency, not cleanliness, and resist the urge to pause reporting until the data is perfect. Every revenue team believes their data is uniquely bad and that this is why they cannot run the business properly. It is not true and it is not the constraint.

Imperfect data computed the same way every period still produces reliable trends. Garbage in does not have to mean garbage out, as long as the garbage is consistent. What actually breaks a metric is inconsistency, meaning the definition changed mid-year, a team started using a field differently, or a filter was added to one report and not another.

So the audit's data step is narrow. Verify that each metric was computed identically across the periods being compared and across the teams being compared. That check is fast, it is decisive, and it protects the trend lines that forecast accuracy work depends on. Standardizing what each stage means is the companion exercise, and how to create a sales forecast walks through where those definitions bind.