Best Prompts for Customer Success

Customer risk rarely appears in one clean signal. It shows up as lower usage, missed meetings, unresolved support issues, a changed sponsor, vague outcomes, or several small changes at once. Customer Success teams must decide which accounts need attention and what action could actually help.

A prompt can organize the evidence and propose a next-best action, but only if it avoids turning incomplete account data into a prediction. The practical goal is a review that helps a person decide what to do next and why.

Why ordinary prompting fails

“Is this customer at risk?” encourages a yes-or-no answer from mixed evidence. The model may overvalue a single metric, confuse correlation with cause, or recommend a generic check-in email. It can also overlook healthy explanations, such as seasonal usage or a completed rollout phase.

A stronger prompt defines the intended outcome, asks for competing explanations, and ties each proposed action to a specific signal. It should separate observed facts from account interpretation and make missing evidence visible.

Reusable prompt

Copy
ROLE
You are a Customer Success review partner. Assess account risk from supplied evidence and recommend the smallest useful next action. Do not predict churn or assign motives without evidence.

REQUIRED INPUTS
1. Account name and lifecycle stage
2. Customer goals, success plan, and agreed measures
3. Product usage data with metric definitions and comparison periods
4. Meeting notes, emails, support history, and open commitments
5. Stakeholder map and known role changes
6. Contract dates and relevant commercial facts
7. Known seasonality, rollout phases, or data-quality limits
8. Review date and available action options

STEPS
1. Check input completeness and data freshness.
2. Summarize the customer’s stated goals and current evidence of progress.
3. Identify risk signals and health signals. Cite the supplied record and date for each.
4. For every risk signal, list at least one plausible alternative explanation.
5. Group signals under adoption, outcomes, relationship, support, and commercial context.
6. Rate each category Red, Amber, Green, or Unknown, with a short rationale.
7. Propose up to three next actions. Connect each action to evidence, a desired learning or outcome, an owner, and timing.
8. Select one next-best action based on urgency, reversibility, and likely information gained.
9. Draft a concise internal note and, only if appropriate, a customer message.

OUTPUT FORMAT
A. Account summary
B. Goal progress table: goal | evidence | status | confidence
C. Risk and health signals: signal | category | source/date | alternative explanation
D. Category assessment with rationale
E. Next-action options: action | evidence | purpose | owner | timing | risk
F. Recommended next-best action
G. Internal CRM note
H. Optional customer message
I. Missing evidence and review date

EVIDENCE AND UNCERTAINTY RULES
- Use only supplied account information.
- Label facts, interpretations, and unknowns.
- Do not calculate a churn probability.
- Do not treat silence, low usage, or a stakeholder change as proof of dissatisfaction.
- Mark stale or incomplete data.
- Preserve conflicting signals.
- If no action is justified, recommend monitoring and specify what to watch.

What to provide

Start with the customer’s own goals and the success plan. Usage data is more meaningful when the prompt knows what the customer intended to accomplish and which phase they are in. Define every metric, its time window, and the expected comparison. Add dated meeting notes, support history, commitments, stakeholder changes, and contract context.

Include healthy signals as well as worrying ones. A risk-only source pack will produce a risk-only story. Note known seasonality, planned pauses, implementation milestones, and data gaps. If communications are included, provide only what your team is permitted to process and remove irrelevant personal information.

How to review the output

First confirm that the account summary reflects the customer’s stated outcome rather than your internal adoption target. Check each Red or Amber assessment against the cited evidence and ask whether the alternative explanation is credible. Make sure old notes are not being treated as current facts.

Then examine the proposed actions. A good next action has a clear purpose, such as validating a changed priority, resolving a blocked workflow, or confirming ownership. It should not create unnecessary escalation. Review any customer message for tone, accuracy, and context before sending it. The account owner should make the final decision.

Where it fails

The review will be weak when success criteria were never agreed, usage definitions are unclear, or relationship notes are stale. It cannot observe unrecorded conversations or organizational changes. It may also miss account-specific context known by the Customer Success manager but absent from the inputs.

A structured review is not a churn model, and its category labels should not become automatic commercial decisions. Legal, contractual, or sensitive relationship issues require the appropriate internal review. Sometimes the best response is to collect better evidence rather than contact the customer immediately.

Practical takeaway

The best customer-risk prompt does two jobs: it organizes mixed signals and slows down premature conclusions. By requiring health signals, alternative explanations, and a purpose for each action, it supports a more proportionate response.

Try the prompt in Agentic Workers on one account review, then have the account owner confirm every signal and approve the next action.

<!-- X derivative: The best customer-risk prompt weighs risk and health signals, tests alternative explanations, and links one next action to evidence instead of guessing churn. -->