Case study · B2B SaaS · reconciliation

Satisfaction said 4.0. The session data said nine minutes a batch.

An invoicing platform for accounting teams, roughly 30,000 monthly active users. Renewal conversations were fine and satisfaction was steady, yet finance-team churn had crept up two quarters running with no complaint trail explaining it. The silent-friction signature, textbook.

Before

2.9@0.81

After · 9 weeks

3.6@0.87

Delta

+0.7

What the measurement said

The Task surface scored 2.4 behaviour against 4.0 perception: the silent quadrant, concentrated in the monthly reconciliation flow. Session data showed a median of 9 minutes and 40 interactions to reconcile a batch that the matching engine resolved in seconds: users were manually confirming matches the system had already made, one row at a time, and describing the routine in interviews as “how reconciliation works”.

Baseline confidence was 0.81: the behavioural side fully instrumented, perception from eight interviews and two quarters of tickets.

Zero tickets in eighteen months, and forty thousand sessions a month paying nine minutes each. That is what silent friction looks like in production.

The four surfaces at baseline

SurfaceBehaviourPerceptionQuadrant
Task2.44.0Silent
Commitment3.43.5Flow
Access3.13.2Watch
Belief3.63.7Flow

The treatment

Per the silent-friction playbook: subtract, do not explain. Confirmed matches became auto-accepted with a review list, replacing per-row confirmation; the step was removed outright. Mismatch errors were rewritten to name the specific row and offer the two legal fixes inline, closing the error-recovery gap that doubled batch abandonment after any error event.

What did not work

The failed first attempt

Before removing the confirmation step, a tooltip shipped explaining that matches could be batch-accepted (the explain-it option the playbook warns against, kept because it was cheaper). Four weeks of data showed no movement: users who had normalised the routine did not read guidance about the routine. The step had to go. That failed month is inside the nine weeks, and it is the most instructive part of the engagement.

9m → 70s
Median time to reconcile a batch
+18pts
Batch completion
2.4 → 3.9
Task surface, behaviour

Composite, nine weeks apart

2.9@0.81 3.6@0.87

Perception stayed at 4.0 throughout, as the framework predicts for silent friction. Nobody praised the change, and nobody needed to. The delta lives in the behavioural numbers, which is why the numbers exist.

Next case · Homeware brand · phantom friction Find your gap