Retention is the thinnest block in most agency go-to-market systems, and it fails the same way every time: nobody wrote down what "working" means, so the client decides alone, quietly, at renewal.
Two modes. Ask which; default to PLAN.
Write the output in the user's language. In Ukrainian use plain words: success criteria, check, continuation, expansion, no anglicisms.
Mode PLAN: the success plan per account
What it produces
A one-page agreement, confirmed on a call, that answers four things: what success is, how we will know, who cares, and what we do when it slips.
1. Success criteria, in the client's words
3 to 5 criteria. Each: what changes · how it is measured · who confirms it · by when.
- Written from what the client said at kick-off or the last review, in their vocabulary. The contract wording is a fallback, not the source.
- Each criterion must be checkable by someone other than you. "The client is happy" is not a criterion.
- Mix leading and lagging measures: a plan made only of lagging outcomes gives you nothing to steer with for three months.
- If the client cannot name a criterion, that IS the finding. Write "criterion not agreed" and make agreeing it the first checkpoint. Do not invent one to fill the row.
2. Stakeholder map
| Person | Role in the decision | What they need from us | Risk if they leave | Last real contact |
|---|
- Minimum: the economic buyer (who pays), the day-to-day contact (who works with you), the end user of the result (who feels it).
- Single-threaded accounts (one contact only) are the most common preventable churn. Flag it as a risk with a named action, not as an observation.
- "Last real contact" means a conversation. A status email does not count.
3. Checkpoint rhythm
| Rhythm | What happens | Who |
|---|---|---|
| Weekly | delivery status, blockers | PM and day-to-day contact |
| Monthly | criteria check: on track or slipping, with numbers | account manager and day-to-day contact |
| Quarterly | quarterly review (qbr-builder, if installed) |
account manager and economic buyer |
The monthly criteria check is the one that gets skipped, and it is the one that prevents surprises. Pick a date, name an owner, put it in the calendar as part of the definition of done.
4. Early signals and what we do
Signal, owner, action, deadline. Minimum four:
| Signal | Action |
|---|---|
| A criterion slipped two months in a row | escalate to the economic buyer with a cause and a fix (an apology alone is not a fix) |
| The economic buyer has been silent 6+ weeks | a direct meeting request, asked for by their own side of the table |
| Scope is shrinking | bring the next phase forward (from upsell-mapper, if installed) before the current one closes |
| The decision maker changed | treat it as a new sale: re-qualify (scope-qualifier, pack sales-engine-skills, if installed) and re-agree the criteria |
5. What we need from the client
Explicit: access, data, decisions, review turnaround. Most "agency underperformance" is a client input that never arrived. Writing it down early is what makes that conversation possible later.
Mode FEEDBACK: client feedback into revenue decisions
A feedback summary on its own leads nowhere. Every item exits as a decision or a discard, with an owner.
Process
- Collect the raw material: review calls, NPS answers, quarterly review notes, complaints, churn reasons, lost-deal reasons. Quote verbatim; never paraphrase into corporate tone.
- Classify each item on two axes:
- About what: service · process and communication · people · price · expectations (what was sold differs from what was delivered)
- What to do: fix · explain · sell (an unmet need is an expansion) · reject
- Root cause, not symptom. "They are slow to reply" is a symptom; the cause is capacity, routing, or a response time nobody agreed, and each has a different fix. Classify to the cause.
- Count. One angry client is a story; the same complaint from four is a process defect. Say which of the two you are looking at, with the count.
- Revenue read: for each cluster, which accounts it touches, what revenue sits behind them, and whether it threatens renewal or blocks expansion.
- Decision list: action · owner · date · which account is told about it. Closing the loop with the client who complained is where retention actually comes from.
- Expansion candidates found in feedback go to
upsell-mapper(if installed) with the quote attached. A client's own words are the strongest reason an expansion can carry. Strong positive quotes with numbers are raw material forcase-study-writer(if installed), with the client's permission.
NPS: use it properly or do not run it
The score alone is decoration. What matters is the follow-up question ("what would have to happen for this to be a 9?") and a human answering every detractor within 48 hours. If nobody will do the follow-up, do not send the survey: an ignored survey costs more trust than no survey. Never publish a score built on a handful of responses as a company metric.
Inputs from other skills (optional)
- Account health from
client-health-check(packaccount-management-skills). If it is not installed, ask the account owner for a short health read and label it self-reported. - Expansion options from
upsell-mapper(same pack). - Quarterly review from
qbr-builder(same pack).
Rules (both modes)
- Facts and quotes only. Never invent a criterion, a stakeholder's motive or a feedback theme.
- You write; the client-facing people speak and send.
- If
anticopywriting-ai(packgtm-skills) is installed, run the prose through it. - Save the plan as a Notion page under the client's page if the Notion connector is available;
otherwise as a markdown file
<client>-success-plan.md(or<client>-feedback-<date>.mdin FEEDBACK mode). - Definition of done, PLAN mode: criteria agreed on a call with the economic buyer, the monthly check dated, every signal assigned an owner. Until the economic buyer has agreed to it, the plan is only a document.
Credits
Method by Victor Shulga (victorshulga.com).