Practical guide

AI customer success plan verification checklist

A customer success plan is not verified because it has a polished goal statement, a long task list, or an AI-generated executive summary. Verification means a reviewer can trace the proposed outcome to the customer’s stated need, reproduce the baseline and target, see who owns each dependency, and tell exactly what evidence will trigger the next decision. Use this checklist after a plan has been.

By Two Prune

A customer success plan is not verified because it has a polished goal statement, a long task list, or an AI-generated executive summary. Verification means a reviewer can trace the proposed outcome to the customer’s stated need, reproduce the baseline and target, see who owns each dependency, and tell exactly what evidence will trigger the next decision. Use this checklist after a plan has been drafted and before it becomes the operating record for a customer review. It tests the artifact; it does not predict retention, expansion, adoption, or customer outcomes.

Freeze the plan you are checking

Record these fields before reviewing the content:

If the version, outcome, or decision is undefined, return the plan. Checking more boxes cannot repair an unstable review boundary.

Gainsight describes success plans as a way to capture goals, objectives, tasks, owners, timelines, and progress. Salesforce’s customer-success guidance similarly starts with a shared definition of success, customer goals, measures, roles, and a review rhythm. Those product guides support the structure below; they do not establish that a particular plan will produce a business result. Gainsight Success Plan overview Salesforce customer success journey

  • [ ] Customer or account identifier and plan version
  • [ ] Review period and exact workflow in scope
  • [ ] Customer-stated outcome and the source where it was recorded
  • [ ] Baseline date, target date, and measurement owner
  • [ ] Customer owner, provider owner, approvers, and dependency owners
  • [ ] Data sources, refresh dates, known gaps, and access restrictions
  • [ ] Next decision and the date or threshold that triggers it

Sources: [1] [2]

Apply the PROVE test

PROVE is a five-part verification pass for the finished plan.

### P — Pin the outcome, scope, and source

A statement such as “improve adoption” is too loose to pass. A reviewable statement names the work that should change and what observable condition would count as progress.

### R — Reconcile the baseline, target, and calculation

For example, “80% adoption” is not reproducible until the plan defines eligible users, the activity that counts, the observation window, and how new or inactive accounts are treated.

### O — Observe the evidence chain

Do not accept “confirmed in meeting” without the date, participants, decision, and accountable owner. A screenshot without the underlying definition or period is also weak evidence.

### V — Verify owners, dependencies, and authority

Shared ownership often means no ownership. Contributors can be listed, but one person must be responsible for closing each decision or escalating it.

### E — Examine the decision and review cadence

  • [ ] The outcome is written in the customer’s language, not inferred from product activity.
  • [ ] A meeting note, approved brief, or other dated record supports the outcome statement.
  • [ ] The plan identifies the people, workflow, location, product surface, and time period in scope.
  • [ ] Adjacent ambitions are labelled out of scope instead of quietly folded into the target.
  • [ ] The plan distinguishes a customer commitment, a provider commitment, and an assumption.
  • [ ] The baseline has a value, unit, population, date range, and source.
  • [ ] The target uses the same definition and denominator as the baseline.
  • [ ] Any percentage, rate, or time saving can be recalculated from retained inputs.
  • [ ] Product activity is not substituted for an operational or business outcome.
  • [ ] Missing, excluded, duplicated, or late records have an explicit treatment.
  • [ ] A change in instrumentation or process is recorded before two periods are compared.
  • [ ] Every milestone names the evidence that will show completion.
  • [ ] Evidence links to a stable system record, document version, or dated approval.
  • [ ] An AI-generated interpretation is distinguishable from source facts and human judgment.
  • [ ] Contradictory evidence is preserved and resolved, not omitted from the summary.
  • [ ] Sensitive material follows the approved access and retention process.
  • [ ] The plan states what remains unknown after the current evidence is reviewed.
  • [ ] Each action has one accountable owner and a dated next step.
  • [ ] Customer-owned and provider-owned actions are visibly separate.
  • [ ] Dependencies name the upstream owner, required input, and escalation path.
  • [ ] The listed owner has authority to make the stated commitment.
  • [ ] Risks have an owner, response, review date, and remaining exposure.
  • [ ] An overdue task changes the forecast or decision rather than only changing a colour.
  • [ ] Every review has a decision purpose, required evidence, attendees, and output.
  • [ ] Continue, revise, pause, escalate, and close conditions are explicit.
  • [ ] A failed threshold leads to a named action rather than an optimistic narrative.
  • [ ] Exceptions record who approved them, why, until when, and what compensates for them.
  • [ ] The next review date reflects the evidence cycle, not an arbitrary calendar habit.
  • [ ] The plan’s status is pass, return for revision, or hold—never merely “looks good.”

Copyable verification record

Use one row for each material outcome, milestone, or exception.

| ID | Customer outcome and source | Baseline / target | Evidence and refresh date | Accountable owner | Dependency | Decision threshold | Disposition | Residual uncertainty | | --- | --- | --- | --- | --- | --- | --- | --- | --- | | CSP-01 | exact statement plus dated source | value, unit, population, period | stable record and owner | one named person | input, owner, due date | continue / revise / hold rule | pass / return / hold | what is still unknown |

Attach the frozen plan, source manifest, calculation sheet, decision log, and exception register. Keep rejected AI suggestions or superseded calculations identifiable when policy permits; otherwise a later reviewer cannot reconstruct why the plan changed.

Worked test: invoice-exception workflow

This synthetic plan covers a customer’s AI-assisted invoice-exception workflow. The draft says the goal is to “increase adoption to 80% and reduce handling time by 30%.” It assigns the customer success manager to every task and schedules a monthly review.

PROVE first checks the customer source. The discovery note says the customer wants analysts to resolve routine exceptions faster without increasing correction work. The plan’s adoption target is therefore an activity indicator, not the customer outcome. The reviewer rewrites the outcome as a measurable decision question: can the defined analyst group reduce median active handling time for eligible routine exceptions while keeping the correction rate within the customer’s approved limit?

The reconciliation pass then finds that the baseline uses all exceptions from July, while the target excludes complex cases and uses only logged-in users as its denominator. That comparison fails. The plan is returned until both periods use the same eligible-case rule, analyst population, and handling-time definition. The 80% figure is retained as a secondary usage measure with its own denominator, not presented as proof of impact.

The ownership pass exposes another problem. The customer success manager cannot change the customer’s exception taxonomy or approve access to correction data. The revised plan assigns taxonomy approval to the customer operations owner, data access to the security owner, measurement preparation to the analyst, and the review recommendation to the programme owner.

Finally, the review record defines a continue threshold, a revise condition for missing or changed data, and a hold condition if correction records cannot be obtained. At the next meeting, participants will decide whether to expand the workflow, revise it, or pause—not merely review a dashboard. The resulting artifact passes because the evidence and decision are inspectable, not because the example proves the workflow caused an outcome.

Return or hold conditions

Return the plan when the customer outcome is replaced by a vendor activity metric, baseline and target definitions differ, calculations cannot be repeated, dependencies lack owners, evidence has no stable source, or review meetings have no decision rule.

Hold it when the customer has not approved the outcome, essential data cannot be accessed under the agreed process, the accountable decision maker is absent, a material contradiction remains unresolved, or the plan makes a causal claim that the available design cannot support.

Two Prune can use a customer-success-plan task in a realistic AI-enabled work assessment and evaluate how a participant frames the customer need, navigates evidence, uses AI, verifies calculations and claims, exercises judgment, and communicates a plan. It provides assessment evidence, not an employment decision, customer-success service, or proof that a plan or control is effective.

Sources

Gainsight: Success Plan overview

  • Salesforce Trailhead: Customer success journey

Sources: [1] [2]

Sources

  1. 1.Gainsight Success Plan overview · support.gainsight.com
  2. 2.Salesforce customer success journey · trailhead.salesforce.com