Customer Success

How do you build a customer success system?#

1. Write a success contract, not a goal#

For each segment, document: the business or risk outcome, the baseline and target, the owner and data source, the time to first value, the customer actions required, the product milestones, the review cadence, and the conditions outside your control.

"Increase adoption" is not an outcome. Adoption is a behaviour that is supposed to cause an outcome, and stating it as the goal quietly relieves everyone of proving that it did.

2. Segment coverage on three variables#

Coverage tiers run digital/self-service → pooled → named → strategic. Note that these three variables do not always point the same way: a small, technically simple, high-reference-value design partner may warrant strategic coverage that its ACV cannot justify. Make that a deliberate, named exception rather than a quiet one.

3. Build a health model you can explain#

Useful dimensions: time-to-value progress, depth and breadth and frequency of successful use, outcome achievement, unresolved severity-weighted incidents, champion and executive engagement, contract and payment risk, and organisational change at the customer.

Start with explicit rules and validate them against what actually happened at renewal. A black-box score that a CSM cannot explain to a customer will not change anyone's behaviour.

risk precision = flagged accounts that churned ÷ all flagged accounts

risk recall = churned accounts that were flagged ÷ all churned accounts

Both matter, and they trade off. Flagging every account yields 100% recall, a precision equal to your base churn rate, and no prioritisation whatsoever — which is functionally the same as having no health model, but more expensive.

4. Run a closed loop, not a status meeting#

At each review: verify the outcome against the baseline, name the blocker, assign an owner and a date, and decide which lever applies — product, enablement, services, commercial change, or a managed exit. Log product feedback by frequency and economic impact, because the loudest request and the most valuable one are rarely the same request.

5. Test whether the programme pays for itself#

CS return = incremental gross profit preserved or expanded − incremental CS cost

The subtraction is easy. The word incremental is where the analysis actually happens. Compare matched cohorts, phase the rollout, or use a coverage threshold as a natural experiment. Crediting customer success with every retained dollar is not measurement, it is accounting for a decision you have already made.

Worked example: does a segmented success programme pay?#

A cohort of 50 customers at $24,000 ACV starts the year at $1.2 million ARR. Baseline gross revenue retention is 82% and expansion is 14% of starting ARR.

baseline retained ARR = $1,200,000 × 82% = $984,000

baseline expansion = $1,200,000 × 14% = $168,000

baseline ending ARR = $1,152,000, so baseline NRR is 96%.

The company pilots a segmented success programme costing $72,000 a year. A comparable cohort under the programme shows GRR of 88% and expansion of 17%:

programme retained ARR = $1,200,000 × 88% = $1,056,000

programme expansion = $1,200,000 × 17% = $204,000

programme ending ARR = $1,260,000, so NRR is 105%.

incremental ending ARR = $1,260,000 − $1,152,000 = $108,000

At an 82% gross margin:

incremental annual gross profit = $108,000 × 82% = $88,560

first-year return = $88,560 − $72,000 = $16,560, or 23% on programme cost.

Three reasons not to believe that 23%#

It is not causal. If the programme cohort contained larger customers, or benefited from a better product release, the gap is not the programme's. A phased rollout or a matched-cohort analysis is the minimum standard before this number leaves the building.

It compares a run-rate to a cash cost. The $108,000 is an ARR delta at year end, not revenue recognised during the year. The $72,000 was spent across the year. If the benefit accrues roughly evenly, in-year recognised incremental revenue is nearer $54,000, gross profit $44,280, and the first-year cash result is:

$44,280 − $72,000 = −$27,720

The programme is not a 23% first-year return; it is a negative first-year cash result that pays back in year two, assuming retention holds. Those are different decisions, particularly on a constrained runway.

The cost line may be understated. The programme provides 18 hours per customer, or 900 hours across the cohort. At $72,000 that implies $80 per hour fully loaded. A customer success manager costing $150,000 fully loaded and delivering 1,300 productive hours a year costs about $115 per hour, so 900 hours is nearer $103,800. Re-run the run-rate calculation at the realistic cost:

$88,560 − $103,800 = −$15,240

The programme now loses money on its own optimistic basis. The formula did not change; an assumption buried inside a budget line did. Before defending a CS return figure, price the hours independently of the budget you were given.

And capacity is not an average#

Even at 900 hours, one CSM against 1,300 productive hours looks comfortable — 69% utilisation. It is not. Renewals cluster, onboarding arrives in waves, and escalations do not queue politely. Staff against the peak and the arrival pattern, not the annual mean, or the programme's worst month will be the one that decides the renewal.

What are the common mistakes?#

• Renaming support as customer success. Reactive case resolution and proactive outcome delivery need separate measures, separate staffing, and separate targets. Merging them means the urgent permanently displaces the important.
• Starting the retention conversation at renewal. Most churn risk is created during qualification and onboarding. By the renewal call the causes are historical.
• Treating logins as value. Activity may be necessary without being sufficient. A daily user completing the wrong workflow generates telemetry and no outcome.
• Building an opaque health score. If the team cannot see the drivers, they cannot act on the score, and they will quietly revert to their own instincts.
• Compensating on NRR alone. It rewards selling into weak accounts. Pair any expansion target with a gross-retention and outcome-achievement gate. See Net Revenue Retention.

When does customer success break?#

• It is compensating for weak product-market fit. If a large share of accounts require heroic intervention to succeed, the target market is too broad or the product is unfinished. Narrow the segment or fix the product — do not hire around it. See Product-Market Fit.
• The sale promised something the product does not do. No amount of proactive engagement resolves a mis-sold outcome. This is a qualification defect that shows up in a retention number.
• The customer lacks the capacity to change. Some outcomes require the customer to reassign staff, clean their data, or alter a process. If they cannot, the outcome will not occur regardless of coverage. Name this in the success contract as a condition outside your control.
• Health data is measuring the wrong surface. Usage that happens outside the product, account identities that change, and outcomes that lag by two quarters all corrupt telemetry. Combine signals with explicit customer confirmation, and record the uncertainty.
• Measurement turns into surveillance or pressure. Use only data the customer has authorised, avoid renewal tactics that trade on friction, and keep offboarding and data export workable. Retention earned by value and retention produced by captivity look identical in the metrics and behave very differently the moment a credible alternative appears. See Switching Costs. Data-handling obligations vary by jurisdiction and contract; this is general guidance, not legal advice.