Cohort analysis for product management: activation

Cohort analysis for product management compares activation and repeat use among customers who entered under a shared condition, such as the release they first used. Give every cohort the same observation window and keep the denominator fixed. A higher rate identifies a difference worth investigating; it does not by itself show that a release caused the improvement.

Define the comparison before opening the workbook

Start with a prepared customer summary approved by product operations. This is a business review of supplied counts, not a guide to collecting product events. The product owner must define which completed activity counts as activation and which later activity counts as repeat use.

Choose one unit. A business product might count customer accounts; a different review might count individual users. Mixing accounts in the denominator with users in the numerator makes the rate meaningless.

For the worked example, a US business software company counts customer accounts. Release A contains January 2026 signups; Release B contains February 2026 signups. The release assignment comes from the supplied summary rather than an assumption about signup dates.

Use these rules:

• Activation means the agreed core activity occurred from signup through the following thirteen days.
• Repeat use means an activated account completed that activity again from fourteen through twenty-seven days after signup.
• Include only accounts whose full twenty-eight-day window has closed by March 31, 2026.
• Keep self-service and assisted onboarding separate so their mix can be checked.

For longer-term retention, use the separate guide to building a cohort retention table. This review concentrates on the initial activation journey.

Bring a prepared summary with explicit counts

Ask for cohort, onboarding route, eligible accounts, activated accounts, and repeat accounts. The summary owner should confirm that each account belongs to one cohort and one route, and that counts use the windows above. Retain a customer-level supporting file for investigating discrepancies.

Put the following approved summary in cells A1:E5. These are the complete inputs for the example.

Do not add immature accounts with blank outcomes to the eligible count. Blank means the observation is unavailable unless the summary owner explicitly defines it otherwise. Once a window has closed, an eligible account without the qualifying activity contributes a zero outcome.

If you receive customer rows instead, the approved cohort, route, eligibility, and outcome flags can be counted with COUNTIFS. Google's COUNTIFS documentation describes counting rows that meet multiple criteria. Deduplicate customer IDs before that step; the function counts matching rows, not distinct customers.

Calculate rates with the same denominator

Add activation rate in F2 and repeat-use rate in G2, then fill both down:

=IF(C2=0,"",D2/C2)
=IF(C2=0,"",E2/C2)

Both rates use eligible accounts. Format the results as percentages with one decimal place. This gives:

For release totals, put A and B in I2:I3. Eligible accounts in J2, activated accounts in K2, and repeat accounts in L2 are:

=SUMIFS($C$2:$C$5,$A$2:$A$5,I2)
=SUMIFS($D$2:$D$5,$A$2:$A$5,I2)
=SUMIFS($E$2:$E$5,$A$2:$A$5,I2)

Google's SUMIFS documentation explains summing values that meet specified criteria. Divide K2 and L2 by J2, then fill the formulas down.

Release B is higher by 74.0% - 68.0% = 6.0 percentage points on activation and 56.0% - 48.0% = 8.0 points on repeat use. These are differences in observed cohort rates.

Covirage is not affiliated with Microsoft or Google.

Check the totals and the mix

For each input row, repeat accounts must be no greater than activated accounts, and activated accounts no greater than eligible accounts. The supplied customer file should reconcile to the summary's distinct account counts. Release totals must equal the sum of their routes.

Then examine the mix. Release A is 60 / 100 = 60.0% self-service and 40.0% assisted. Release B is 80.0% self-service and 20.0% assisted. Its total therefore reflects a different route composition.

Hold the route weights at Release A's mix as a sensitivity check. Using Release B's route rates gives activation of 60.0% × 70.0% + 40.0% × 90.0% = 78.0%, and repeat use of 60.0% × 50.0% + 40.0% × 80.0% = 62.0%. Those standardized figures answer a different question from B's observed totals; label them separately.

Neither calculation isolates a release effect. NIST's explanation of completely randomized designs describes random assignment of factor levels to experimental units. A release-by-calendar comparison has no such assignment merely because its windows match.

Use marketing cohort analysis for acquisition questions and behavioral cohort analysis for grouping by observed actions. For reading longer time-based tables, see how to read a cohort retention table.

Where it goes wrong

• Unequal maturity. A newer cohort has had less opportunity to activate or return.
• Changed definitions. A revised core activity breaks comparison with earlier cohorts.
• Averages of subgroup rates. Sum qualifying counts and eligible counts before dividing.
• Changing customer mix. Show relevant subgroup counts beside the overall movement.
• Causal release claims. Route differences and calendar changes remain possible explanations.

Review your prepared cohort file in Covirage

Upload your approved cohort summary or Google Sheet to Covirage. Covirage maps the columns; its tools compute and check the comparisons, and the external AI model explains them with every figure traced to its rows. Create a free Covirage account and bring a sample of your own data.

See self-service analytics for the supporting business reporting context.