rulebase-co Attributing revenue retention to support
Support and CS get asked to justify themselves in revenue terms, usually before a budget cycle. The request is legitimate. The usual answer — sum the ARR of accounts that support touched and renewed — is not, and it is transparently not, which is why these numbers rarely survive their first serious audience.
This skill is mostly about what you cannot claim, because the credible version of the claim is much smaller and much more durable than the one people reach for.
Why the obvious method fails
Almost every retained account was touched by support. In most businesses, support contact reaches the large majority of customers, so "accounts support touched that renewed" approximates total renewed revenue. The metric has no discriminating power: it would look identical if support were excellent or useless.
Three further problems:
- Selection. Customers who contact support are more engaged, and engaged customers renew more. The correlation exists without support doing anything.
- Multiple claimants. Product, CS, sales and support all touch the same retained account. If each claims it, the company's attributed retention exceeds its actual retention — a real and common outcome that discredits everyone's numbers at once.
- Reverse causation. Growing accounts generate more contacts. Contact volume is partly an effect of expansion, not a cause.
The honest ladder
Four claim strengths. Use the strongest you have evidence for, and name which one you are making.
1. Contribution, unquantified. "Support resolved 14,000 issues, including 340 on accounts that renewed at over £100k." True, useful context, no revenue claim. Perfectly respectable, and better than an inflated number.
2. Association, with a comparison. "Accounts with an unresolved escalation churned at X% versus Y% for matched accounts without one." Requires matching on segment, size, tenure and usage trend. Supports a directional claim and a defensible order of magnitude.
3. Influence, with a defined mechanism. For specific, traceable events — a save following an escalation, an expansion following a request-shipped notification, a renewal where the customer named a support recovery. Countable, small, and real. This is usually the strongest honest claim available, and its smallness is the point: it is the part you can actually defend line by line.
4. Causal, with a holdout. Withhold the motion from a random subset and compare. The only claim that survives a sceptical audience, and it is available far more often than people assume — request loop-closing, proactive outreach, adoption nudges and save plays can all be randomised at low cost.
If you are at rung 1 or 2, say so. A stated association is more persuasive to a serious reader than an overclaimed causal number, because the reader can tell the difference and will discount the whole analysis if they catch one.
Fix the double-counting before publishing
Agree the rule with the other claimants before the number goes anywhere:
- Single-touch by convention — one team gets the account. Simple, arbitrary, and it makes the totals add up.
- Explicitly non-additive — each team reports its own influenced figure, clearly labelled as overlapping and never summed. Honest and requires discipline in every deck that quotes it.
- Split by agreed weights. Feels rigorous, is arbitrary, and produces an argument every quarter. Avoid unless finance owns the weights.
The failure to avoid is each team quietly using method two while leadership adds them up. Say in the deliverable whether the figure is additive with other teams' claims. That sentence is the most valuable thing in the analysis.
What support can measure that is genuinely its own
Rather than competing for credit on the revenue line, measure things support uniquely controls and that plausibly precede revenue:
- Repeat contact and unresolved rate — the strongest support-side predictors of account risk.
- Unkept commitments — countable, clearly support-owned, and clearly bad.
- Service attainment against what each account was sold, per account rather than in aggregate.
- Time to resolution from the customer's first contact, including escalation paths.
- Recovery outcome — for accounts that had a serious failure, what share were still there a year later, against matched accounts.
These are defensible, they are not contested by other teams, and they move before revenue does — which makes them more useful to manage by than an attributed revenue figure anyway.
Framing the negative case
Sometimes the honest finding is that support's revenue effect cannot be isolated. That is not a failure of the analysis and it is not an argument against funding support — plenty of essential functions cannot be attributed a revenue share, and nobody asks security to prove its NRR contribution.
Say it directly, offer the strongest available alternative (rung 1 or 2 plus the support-owned metrics above), and propose the holdout that would settle it. A team that declines to invent a number and instead proposes the experiment tends to be trusted with the next question too.
Guardrails
- Do not sum influenced figures across teams, and say so explicitly in the artifact.
- Do not present rung 2 language as rung 4. "Support drove £2m of retention" is a causal claim; "accounts with resolved escalations retained at a higher rate" is not. The wording is the claim.
- Do not build the number backwards from the budget you want to defend. It will be caught, and the cost is every future analysis.
- Do not use attribution to deprioritise low-revenue accounts. In regulated sectors, differential service on that basis is a conduct issue.
- Joining support, CRM and billing data creates a personal-data record none of the sources held. Flag purpose and retention.
- Cite ids and aggregates, never transcripts. These figures go to boards and investors.
Present results to the user
- Which rung of the ladder the claim sits on, named, at the top.
- Why the naive method was rejected, briefly — it pre-empts the first question.
- The comparison for any association claim: what was matched on, the margin, the interval.
- The traceable influence cases, counted individually, with evidence ids.
- Whether the figure is additive with other teams' claims, stated in one sentence.
- The support-owned leading metrics, as the thing to manage by.
- The holdout design that would upgrade the claim, with its cost — so the choice between a weak number now and a strong one next quarter is the reader's.