Proving It in Eighteen Months

The core problem: NPS-to-revenue correlation is not a causal model, and Finance is right to reject it. The ROI case has to be built bottom-up from a specific mechanism in a specific population.

Measured (not assumed):

• Baseline churn and expansion rates by segment (ARR band, tenure, implementation cohort), using 24+ months of history.
• The size of the target population the intervention actually touches - e.g., accounts that go through the redesigned implementation. If implementation covers 600 accounts a year, your ROI must be computed on 600, not 1,800.
• Cost per unit: fully loaded program cost of $2.1M over 18 months, plus the run-rate after.

Assumed (stated with a range and a source):

• The expected reduction in first-year churn for accounts that complete implementation on the new path vs. the matched old-path cohort. This must come from an observed effect - pilot vs. comparison group, or historical cohort analysis - with a conservative and a base case. Any assumption without an observed anchor should be labeled as a hypothesis and excluded from the base case.

Excluded explicitly:

• Any value attributed to NPS movement.
• Cross-sell revenue not yet observed.
• Brand or advocacy value.
• Savings from headcount that has not been removed.

The presentation that survives Finance: one page, three lines - accounts affected, observed effect with confidence range, revenue retained. Then the payback period. If the base case doesn't clear the hurdle, say so and propose the smaller program that does.

Do not explain the churn rise away, and do not let the NPS rise do the explaining either. Diagnose it as a composition effect and prove it.

The stated cause - discounted deals in the sub-$15k band - is a testable claim, not a conclusion. Test it: does the churn increase concentrate in the specific acquisition cohorts sold during the discounting period, controlling for ARR band and tenure? If yes, you have a mix and vintage story, and the honest headline is 'churn rose because we acquired a cohort with structurally higher churn; the CX program neither caused nor prevented it.' If the increase is spread across all sub-$15k accounts including older vintages, the discounting explanation fails and the program's effect is genuinely ambiguous.

Presentation to the board in nine months:

• Show segment-level churn, not blended churn. The blended number is what makes the program look like it failed.
• Show the program's target population separately, with its own before/after, and a comparison group if one exists.
• State plainly that NPS moved and that NPS is not the ROI metric. Explaining the difference between a relationship indicator and a financial outcome is a credibility-building move, not an admission.

On the sub-$15k segment: the strategic question is whether it is worth serving with the same model at all. Options: a lower-touch, digitally-served tier with a different promise; a price/service-level change; or exit. This is a segmentation decision, not a CX program decision, and the program should supply the evidence (churn, cost-to-serve, support volume per account) rather than absorb the blame. Raising the price floor or tiering the service is likely the higher-value lever than anything in the 18-month plan.

Top of the scorecard: net revenue retention for the target segment, with customer-reported effort or success as the leading indicator beneath it.

Why not NPS at the top: it is a relationship summary that moves for reasons unrelated to the program (product releases, market sentiment, survey timing, sample shifts), it does not decompose into action, and Finance has already rejected it as evidence. Putting it at the top of the scorecard sets up the program to be judged on a number it cannot defend causally. It is still worth tracking - it is cheap, comparable, and useful for the board as context - but as a supporting line, not the headline.

Why net revenue retention for the target segment: it is the outcome the program exists to move, it is auditable from the billing system, it decomposes into churn, contraction, and expansion, and it is comparable to whatever benchmark Finance already believes. Segmenting it (by ARR band and implementation cohort) is what prevents the composition effect from swamping the signal again.

The leading indicator beneath it: a measured effort or success metric tied to the specific journeys being redesigned - implementation completion time, first-value milestone attainment, or self-reported effort on the top three tasks. These are the things the program actually controls quarter to quarter; net revenue retention is what they are supposed to add up to.

Practical rule: every line on the scorecard needs an owner, a source system, and a defined response if it moves the wrong way. A scorecard where nobody knows what they would do differently if a metric fell is a reporting exercise, and it will not survive its second review.