Measure churn and run save plays
Churn is two numbers, not one: logos lost and revenue lost. And a save play is a priced bet — offer the cheapest retention motion whose expected saved revenue beats its cost.
Procedure
Fix the cohort and window. Monthly logo churn over January is
logos_lost / logos_at_start; always state the month and the universe.Compute revenue churn separately:
revenue_churn = mrr_lost_minor / mrr_start_minor, in integer minor units.Distinguish voluntary (they cancel) from involuntary (the card fails); involuntary churn belongs to the retry ladder, not the save desk.
For each cancellation compute the value at risk:
var_minor = mrr_minor * expected_remaining_months. Use a survival estimate (about1 / monthly_churn_ratemonths) rather than the average tenure of everything.Rank save offers by expected value:
ev = accept_prob * var_minor - offer_cost_minor, and try them in descending EV order.Worked example: a customer at USD 49/mo with an expected remaining life of 12 months gives
var = 4900*12 = 58800minor. A 50%-off-for-3-months offer costs4900//2 * 3 = 7350minor; at 40% accept,ev = 0.40*58800 - 7350 = 16170minor. Worth offering.Cap the discount window; a permanent discount converts churn into a permanently lower MRR, which is contraction by another name.
Log a reason code on every cancellation and report churn split by reason so next quarter's fix targets the biggest slice.
Re-baseline on net revenue retention: save plays cut churn but also cut expansion, so measure NRR, not gross churn, to catch the trade.
Compute dollar-weighted churn alongside logo churn; one enterprise loss can dominate both.
Track a save rate with its denominator (saves / cancellation attempts) and trend it; a falling rate means the offers are stale.
Feed reason codes into the roadmap as a ranked list; the top cancellation reason is the cheapest churn fix.
Separate voluntary churn triggered by price from that triggered by product; they need different save offers.
Model the save-offer cost against a control group to confirm the save was incremental, not a discount to someone who would have stayed.
Re-forecast MRR after a save campaign, since a successful save at a discount lowers next quarter's expansion.
Pitfalls
Reporting only logo churn hides the accounts that downgrade, and revenue churn hides the accounts that leave.
Counting a save as a win when it was a permanent 60% discount: you saved the logo and destroyed the margin.
Using the average lifetime instead of a survival estimate overstates value at risk for a long-tenured customer already past the hazard peak.
Mixing involuntary churn into the save-desk denominator inflates the save rate.
No reason code means you cannot tell a product problem from a price problem, so you discount the wrong thing.
A save rate without offer cost hides that the saves were bought with margin, not won with product.
Dollar-weighted churn ignoring a single large account can flip a healthy logo number into a bad revenue quarter.
Reason codes collected as free text cannot be ranked; force a fixed taxonomy at cancellation.
Discounting a customer who would have stayed anyway is leakage, not a save; a control group tells the two apart.
Verification
python3 -c "var=4900*12; cost=4900//2*3; print(var, cost, 0.40*var-cost)"
# 58800 7350 16170.0 -> save offer EV is positive, rank it above any costlier save
Report: monthly logo churn, monthly revenue churn, and the top three save offers with their EV in minor units.