A conversion decline is a signal to investigate

Investigate a conversion decline by checking definitions, cohorts and timing before treating it as the cause of a revenue shortfall or choosing a corrective action.

This worked example uses sample data to make the management reasoning visible. Follow the inputs, interpretations and proposed test, then decide which investigation to authorize before the next operating review.

Read the two signals precisely

Assume a review pack contains only these illustrative figures:

Measure	Earlier period	Current period	Change
Revenue index	100	90	Down 10%
Conversion rate	22%	18%	Down 4 percentage points

The revenue index expresses relative movement without specifying an actual currency amount. Moving from 100 to 90 is a decline of 10 divided by 100, or 10%.

Conversion falls by 4 percentage points because 22 minus 18 equals 4. The relative decline is 4 divided by 22, approximately 18.2%. Calling it a “4% decline” confuses the absolute difference between rates with the change relative to the earlier rate.

Neither calculation establishes why the movement happened. The review pack does not yet say what converted, which population was eligible or whether the revenue belongs to those same opportunities.

Identify the missing bridge to revenue

Google’s HEART research, published at CHI 2010, connects goals with measures selected to represent them. In this example, our application is to demand an explicit definition before using conversion to explain a commercial outcome.

Request the evidence needed to connect the measures:

Missing evidence	Why it could change the decision
Conversion definition and denominator	The rate may describe a different stage or eligible population
Cohort dates and maturity	Recent opportunities may not have had time to convert
Opportunity volume and segment mix	Fewer opportunities or a changed mix may affect the aggregate
Deal value and discounting	Revenue per completed sale may have changed
Revenue timing and recognition	Current revenue may come from earlier selling periods
Reporting changes and data completeness	A cutoff or definition change may create an apparent decline

Give each request an owner and ask when the answer can be reliable. Do not fill an unknown field with an assumption simply to complete the narrative. An AI-assisted draft may organize these questions, but someone accountable for the source needs to validate the answers.

Test whether the simplest explanation is coherent

Consider a deliberately restrictive assumption: both measures describe the same mature cohort, while opportunity volume, average revenue per conversion and recognition timing remain unchanged. Under those conditions, revenue would scale with conversion.

The conversion ratio is 18 divided by 22, approximately 0.818. Applied to an index of 100, that would imply roughly 81.8, rather than the supplied revenue index of 90. The mismatch shows that the simple assumptions cannot all describe these synthetic inputs exactly. It does not tell us which assumption fails.

Now change one assumption. Suppose conversion measures newly created opportunities while current revenue mostly reflects earlier cohorts. The two signals may then describe different stages of a longer commercial process. Today’s conversion movement could matter for a later period, but it cannot by itself explain today’s revenue gap.

If matched cohorts later show stable conversion within each segment and a shift toward a lower-converting segment, the investigation should turn toward mix. If comparable mature cohorts show a decline within segments, examining the handoff or offer becomes more credible. These are conditional paths, not additional facts about the example.

Compare actions before choosing a test

One option is to reconcile reporting and cohorts first. It uses analytical effort and delays a commercial intervention, but it addresses whether the signal can guide action.

A second option is a narrow handoff test in a verified segment. It could examine a plausible process problem, but it needs a defined intervention, comparable observations and enough time for outcomes to mature.

A third option is a broad message or offer change. It creates more exposure and changes several explanations at once. The current evidence does not justify that commitment.

The KDD 2014 work by Kohavi and colleagues discusses learning from controlled experiments and the limits of generalized rules. A simple before-and-after movement should therefore not be presented as proof that an intervention caused a change.

Authorize the next piece of learning

For this example, propose a two-week evidence review owned by the commercial operations lead. Its scope is to reconcile definitions, map revenue to cohorts and check the segment pattern using authorized records. It does not authorize a broad rollout, external promises or new spending.

At the checkpoint, require a documented definition, a cohort comparison and an explanation of remaining gaps. Stop the proposed intervention if the data cannot support a comparable test. If the evidence supports a handoff hypothesis, return with a bounded test design and an outcome review date suited to the actual sales cycle.

The next step is an evidence review, with outcomes assessed at its checkpoint. Inspect the operating review example to see how this kind of uncertainty can sit beside a recommendation, then use the operating review solution to frame the decision your own evidence can support.