CSCustomer Success Skillsfor AI agents

customer-advocacy

Who is safe to ask to speak publicly, privately, or to a prospect — what to ask them for, how to word it, and how not to burn them.

v1.0.0CSMAMCustomer MarketingVP CSCCOper-request · quarterly (pool review)MITSKILL.md ↗

When it fires

Skills match on what you actually type, not on a command name. Phrases the skill declares, verbatim from its description:

“reference customers”“ask for a case study”“who can we ask”“who can be a reference”“need a reference call”

What ships in the folder

SKILL.md37.9 KB — the runbook the agent reads
referencesadvocacy-ladder.mdask-templates.mdpool-management.mdreadiness-rubric.md
scriptsadvocacy_score.pysample_pool.json
assetsadvocacy-register.md
evalsevals.json

Read at runtime, on demand — the agent loads only what the task needs.

Install just this skill

npx skills add gaintrace/customer-success-skills --skill customer-advocacy

Or take the whole library: npx skills add gaintrace/customer-success-skills — all install options.

You are running the advocacy desk for a company where the same four customers get asked for everything, and the fifth request is the one that ends a relationship. The job is not to find someone willing. It is to find someone ready, spend that readiness once on the highest-value ask they can carry, and know exactly when they become eligible again.

The rookie version reads an NPS score, picks the promoters and mails them a reference request. It asks customers whose survey score belongs to someone who has since left; it sends a delighted admin into an enterprise call about procurement and multi-region failover; and it returns to the same three accounts every quarter until one stops replying, converting the best relationship in the book into a silent one and filing it as a scheduling problem. Advocacy readiness is behavioural, not attitudinal — a survey score is one of seven inputs and the weakest.

The elite version treats the advocate pool as a finite, depleting, replenishable asset with a capacity per customer per period, a rotation, a rest interval and a register of who was asked what and when. It reads a declined reference as a retention signal rather than a scheduling failure: an advocate who agreed in March and declines in September has told you something no health score has caught. Read ../cs-context/references/evidence-standard.md first — every claim carries provenance, a tier and a confidence level, and every gap is written UNKNOWN — requires X.

Before Starting

  1. Read .agents/cs-context.md (fallback .claude/cs-context.md); if absent, run cs-context. Never ask what that file answers — ARR, renewal date, notice period, segment boundaries, owning CSM, source inventory. Asking one tells the user you did not read the file.

  2. Take the data in whatever shape it arrives. CSV, TSV, XLSX, JSON, NDJSON, warehouse results, a CRM export with three title rows above the header, a pasted survey dump, a call transcript, an email thread — or no file at all, just the answers below.

    • Run ../cs-context/scripts/ingest.py first on every supplied file. It sniffs encoding and delimiter, finds the real header row beneath export preamble, maps columns onto the canonical schema with a confidence per column, normalises dates, money-as-text and booleans, resolves accounts across files, and reports the join rate.
    • Confirm every mapping below 0.80 confidence before using its numbers. last_reference_date → last_activity_date routes a customer asked three weeks ago into another call, and the artifact will not look wrong.
    • Degrade, never refuse. A partial export gives a partial pool with a coverage figure and a confidence cap; the only stop is under-40% coverage of the seven families. And never assume an export is complete or current — ask for the as-of date and print it, because a pool built on a 30-day-old support extract routes an open Sev-1 into a prospect call.

  3. Ask up to four questions, once, tappably — then run unattended. AskUserQuestion, every applicable question in a single batch. Skip any the prompt or cs-context answers.

Header Question Options — recommended first
Job What do you need from this? Fill a specific request (Recommended) · someone has asked for one advocate; returns a ranked shortlist with the ask written — Build the pool · score the book for readiness and produce the register and coverage cells — Pool health check · fatigue, rotation, over-asked accounts, and the declines worth investigating — Repair a burned advocate · one named customer who was over-asked, mismatched or exposed
Ask What is the ask? Sales reference call (Recommended) · most requested and most depleting rung — Case study or public story · 6–12 weeks of calendar and two approval chains — Review, testimonial or logo · low effort; FTC and platform rules apply — Advisory board or co-development · top of the ladder; a two-year commitment, not a favour
Window When must this happen? Inside 14 days (Recommended) · restricts the shortlist to warm advocates inside their rest interval — This quarter · allows a milestone to be reached before the ask — No deadline, building supply · optimises for coverage cells rather than speed
Data What can I work from? Connected sources plus anything I upload (Recommended) · ingest.py runs over the files — I'll paste exports or a transcript · same path, pasted — I'll answer questions instead · from cs-context and your answers; confidence capped at Low — Context file only · fastest and thinnest
  1. Never block, and never guess. Every missing input resolves one of three ways — read it (derive it, show the derivation), ask it (step 3, only where two likely answers change the shortlist), or mark it (UNKNOWN — requires <source> plus a confidence cap). Acting on a substituted value here means putting a customer in front of a prospect. If the batch goes unanswered, run the defaults, state them at the top, and give each a row in the Assumptions table. If support, sentiment and relationship are all missing, name the gap and stop.

How This Skill Works

Output mode — Brief by default

Brief (default, ≤20 lines) unless the user asks for depth. Full — the complete Output Template — when they ask, when the register goes to customer marketing or a QBR, or when a request is being declined and someone will challenge the reasoning. Brief is the answer written first: the named advocate, the rung, the readiness call, the window, who asks by when, confidence in three words, and the falsifier. Then: Full analysis, coverage ledger and workings on request. Brief drops the display of the reasoning, never the reasoning.

The rules this skill enforces

From ../cs-context/references/operating-rules.md, enforced in the output — a deviation states its rule number, the circumstance, and what will be watched.

Rule Enforced how
R1 · The Opt-Out Calendar · R8 · The Health Gate Timing runs against renewal_date − notice_period_days, and no public ask lands inside an unresolved opt-out window. Below the Watch band there is no ask, and the withholding is printed with the band that caused it
R11 · Value First · R14 · The Written Skip · R17 · One Play No ask in a conversation carrying an apology, a miss, a price rise or a commercial ask, and 14 days' separation from any expansion ask either way. Every disqualified candidate is listed with the reason and the date it clears. One open ask per customer at a time
R18 · The Firewall Readiness score, band, ARR, pool tier and fatigue index stay internal. The customer sees the copy block and nothing else
R19 · No Date You Do Not Own No publication date, event slot or approval timeline promised without the named owner agreeing it
R21 · The Stop-Loss Two declines removes an advocate from the pool for two quarters. A third ask costs more than the reference is worth
R22/R23 · Ordering, and the Coverage Cap Readiness is a ranking, not a probability of yes — no conversion rate without a backtest. Confidence never exceeds coverage, and all seven families print

Know the business model before scoring

Read ../cs-context/references/business-model-profiles.md first — it decides which rungs exist here at all, and recommending a keynote to a self-serve business is the most recognisable form of generic output. Product-led: often no champion and no exec sponsor, so the ladder collapses to rungs 1–3; relationship weight falls to 8 and usage and sentiment carry the score. Consumption: readiness is commitment pacing, and a customer above commitment is a strong advocate and a poor case study — the story reads as a bill. Enterprise and regulated: the full ladder, with legal, comms and procurement each holding a veto and a vendor-mention policy adding a 6–16 week chain that often returns an anonymised story. Partner-led or public: not yours to publish; blackouts are absolute.

Seven signal families — read for advocacy readiness

Every family is checked and reported, every time — one with nothing to report prints "checked, clear" and is never dropped.

# Family Weight What it answers for advocacy
1 Product usage & adoption 20 Are they still doing the thing they would be vouching for?
2 Commercial & contract 15 Have they renewed at least once, and does the paper permit publicity?
3 Relationship & engagement 22 Is there a champion with tenure, standing, and a second thread behind them?
4 Support & reliability 15 Would the last 90 days come up unprompted on the call?
5 Sentiment & VoC 13 Have they said something positive, recently, in their own words?
6 Billing & payment 5 Is the money clean? A customer in collections is not a reference
7 Firmographic & external 10 Does their organisation permit this, and is this their moment?

Relationship carries the most weight because advocacy is a person, not an account — a 90-readiness account whose champion started six weeks ago has no advocate in it. Rubrics, thresholds and the PLG profile: references/readiness-rubric.md. Run sequence: demand → sweep seven families → score → disqualifiers → ladder ceiling → pool capacity → timing → ask → log.

Step 1 — Resolve the demand before scoring anyone

Four jobs arrive wearing the same words, and the Job answer from Before Starting fixes which one. A named request asks who is the closest fit, inside their rest interval, that legal has already cleared — get it wrong and your most enthusiastic customer describes a use case they do not run. Building supply asks which coverage cells are empty and which accounts are 90 days from filling them. Pool health asks who is over-asked and which declines are risk signals — get it wrong and you meet the fatigue on the call, in front of a prospect. Repair asks what the burn was and what stops until it is fixed. Job runbooks: references/pool-management.md.

For a named request, capture the fit cell first: segment · industry · use case · buyer persona · deployment shape · the objection the prospect actually raised. A reference matched on enthusiasm and mismatched on cell converts worse than none.

Step 2 — Sweep the seven families and score readiness

Walk references/readiness-rubric.md in full. Record signals fired, checked and clear and not checkable per family; all three reach the output. Readiness = Σ (family_score × weight) / Σ (weights of families with data) — renormalising over available families stops a data gap from manufacturing a false negative; say when it happens.

Score Band Means
80–100 Ready Eligible to the top of the ladder, subject to each rung's own gates
60–79 Ready with limits Rungs 1–5. No public speaking, no advisory board, no co-development
40–59 Not yet Rungs 1–2, zero disqualifiers. Name what would move them up
0–39 Not a candidate No ask. Name the two things that would change it, and the re-check date

The three checks that catch the advocate everyone else misses:

Check Why it matters
Champion tenure and standing, not account tenure A five-year account with a six-week champion has no advocate. Require ≥180 days in role plus one observed internal advocacy event — a meeting they ran without us in the room
A quantified outcome they stated — not a usage stat "43% faster close", produced by their system, in their words. Ours is marketing; theirs is a reference. No agreed number, no rung above 3
Support silence, not support absence · the survey respondent's identity and date Zero tickets is not health — check the last Sev-1, the last reopen, and whether the root cause was closed or merely stopped being reported. A promoter score is one person on one day; a six-month-old 10 from a departed admin is history, not disposition

Step 3 — Apply the disqualifiers

Decisions, not indicators — a high readiness score does not neutralise them. Apply every one that matches; the most restrictive ceiling wins.

Trigger Effect Clears when
Risk band At Risk or worse (R8) No ask, any rung Band back to Watch or better for 30 days
Open P1/Sev-1, or one closed inside 30 days No rung above 1 30 days after closure, root cause documented
Invoice >60 days overdue, credit hold or an open billing dispute No ask, any rung Balance cleared or dispute resolved
Inside the opt-out window, renewal unresolved (R1) No rung above 2 Renewal signed, then 14 days
Champion <180 days in role, or a departure inside 90 days · no quantified outcome the customer stated No rung above 3 180 days in role plus one internal advocacy event · an outcome in their own numbers
No-publicity, no-logo or blanket-NDA clause in the contract No rung above 8 (private only) Legal confirms an exception in writing
Their earnings quiet period or blackout · pool cap reached this period Defer everything dated · rotate, do not ask The window they name · the next-eligible date in the register
The prospect is a direct competitor of the customer Never route Never. Find a different cell
Declined inside 90 days · two declines in 12 months (R21) No re-ask · out of pool two quarters 90 days after diagnosis · two quarters plus a repair conversation with no ask in it

Print every disqualifier that fired, its evidence and its clearing date — one excluded without a written reason and a revisit date is indistinguishable from an oversight (R14).

Step 4 — Set the ladder ceiling

Ten rungs, lowest to highest. Effort and risk run on both sides — which is why this is a ladder, not a menu: the top rungs cost the customer more than they cost you, so they are earned. Nobody's first ask is rung 8; Lincoln Murphy's rule is to start with small asks and increase the level of the ask over time, pegged to their success [P · Lincoln Murphy, Sixteen Ventures].

# Rung Their effort Their risk Our risk Min readiness
1 Survey / NPS response 2 min None None 40
2 Testimonial quote 15 min Low — their name on a sentence The quote drifts from what they meant 40
3 Review-site review 20–30 min Low, but public and permanent Platform and FTC rules (Step 7) 55
4 Logo use 0 min, all legal Their brand next to yours Using it past the term 55
5 Case study / public story 2–4 hrs + approvals Medium — their numbers in public 6–12 weeks of calendar [P] 65
6 Webinar / joint content 3–5 hrs Medium — live and recorded They are asked something we did not prepare them for 70
7 Conference speaking 8–20 hrs + travel High — reputational, public The slot moves after they book flights 75
8 Sales reference call 30–45 min, recurring Medium — repeated, and invisible to their exec The depleting rung; the one that burns people 75
9 Advisory board 6–10 hrs/yr, multi-year Medium — NDA, time, exec exposure Running it as a roadmap focus group 85
10 Co-development / design partner Ongoing, contractual High — their roadmap bets on ours Their production depends on something we may not ship 85

Per-rung gates, lead times, legal paths, per-model rung tables and the advisory-board charter: references/advocacy-ladder.md.

Step 5 — Check the pool before you check the person

Advocacy capacity is a budget. Spend it like one. These are library conventions [P], not measured benchmarks — replace them with your own observed decline rates once you have them.

Control Default Why
Asks per customer per rolling 12 months 3 across all rungs Past three, declines rise and replies slow before anyone calls it fatigue
Sales reference calls per customer per year 4, max 1 per quarter Highest repeat demand, lowest visible reward
Minimum rest between any two asks · separation from any commercial ask 45 days · 14 days either side (R11) Recovery, and it forces rotation. An ask beside an upsell reads as leverage, permanently
Open asks per customer at once · advocates per coverage cell 1 (R17) · ≥3 Two open asks is how a yes becomes a silence. Below three advocates, one holiday empties the cell and the same person gets called

Rotation. Rank by readiness × fit − recency penalty, never by willingness — the most willing advocate is the one you are about to burn. Fatigue is measured, not felt: four observables per advocate per quarter — reply latency against their own baseline · declines in the trailing 12 months · enthusiasm delta between first and latest participation · whether their exec has started routing our asks to someone junior. Two of four wrong means rest them. Thresholds and the burned-advocate repair: references/pool-management.md.

Step 6 — Time the ask

Anchor The window Why
A verified value milestone Within 14 days of a customer milestone — a business outcome they state, not a feature they clicked Value is perishable, and the ask lands as a shared moment rather than a favour [P · Murphy]
A renewal Never inside an unresolved opt-out window. 14–30 days after signature is the strongest window on the calendar Before signature the ask is leverage; after it, the decision is fresh and self-justifying
A QBR Only where the review produced a quantified outcome and nothing is open. Put it on the mutual action plan as a named commitment with the champion and their comms owner The QBR is where their exec hears the number — the audience an advocacy ask needs
A promoter response · a support recovery 7–14 days after the response, to a live contact with standing · 30 days after the Sev-1 closes with a documented root cause Disposition decays — past 90 days a survey score is a historical fact. A recovery story is powerful; asking during the recovery is extraction

Never ask, in any wording: during an open escalation · in the same conversation as an apology, a miss, a price increase or a commercial ask (R11) · inside 14 days of an expansion ask either way · during their quiet period · while their champion is under 180 days in role · when the honest answer would embarrass them · when you have not delivered what you promised last time.

Step 7 — Write the ask

Frame it as value to them, then name what you give back. Four currencies, descending in value to a senior customer: peer access (two named introductions at their scale — the strongest; rungs 8–10) · internal visibility (the outcome analysis they forward to their own exec, their team named — rungs 5–6) · external profile (a speaking credential, a named public story) · their time back (we draft, we prepare them, we take the questions they do not want). Six rules, enforced in the copy block: (1) name what they achieved, with their number and their date, not ours · (2) name the ask precisely — what, how long, to whom, where it appears · (3) state the approval rights up front; they approve every word · (4) pre-build the graceful decline, "if the timing is wrong, say 'not this quarter'" · (5) name what they get concretely — "exposure" is not a benefit, two named introductions is · (6) one ask per message. Rules with real penalties:

Rule Detail
FTC Rule on the Use of Consumer Reviews and Testimonials, effective 21 Oct 2024 [A] An incentive may not be contingent on a review being positive or expressing a particular sentiment — express or implied — and the reviewer's own disclosure does not cure it. Insider reviews require clear disclosure. Civil penalties up to $51,744 per violation
Review-platform policy (G2 Community Guidelines) [V] Incentive value capped at $100; eligibility never based on the opinions expressed; confirmed incentivised reviews carry a permanent "Incentivized" tag; vendor employees and direct competitors may not review; reseller reviews are labelled "Business Partner" and excluded from the score
Approval chains [P] A written case study runs 6–12 weeks through customer review, legal sign-off and quote approval. Start the clock when you ask, and offer the anonymised version before their legal team invents it

Every customer-facing word follows ../cs-context/references/customer-voice.md: a fenced text block below the divider, email-client formatting, no unfilled placeholders. Five worked examples, the decline reply and the repair note: references/ask-templates.md.

Step 8 — Log it, and read the declines

Write every ask to the register (assets/advocacy-register.md) when it is sent, not when it is answered: account · contact · rung · asked by · asked on · outcome · next-eligible date · fit cells · approval owner. An unlogged ask is how one customer gets asked twice in a fortnight.

Observation Read it as Do this
Agreed to a reference, then withdrew Advocacy withdrawal — a P0-grade signal that something changed which no survey has caught Run churn-risk on the account this week. Do not re-ask
A previously willing advocate declines with no reason Champion standing has changed, or an internal issue exists we cannot see Ask the champion what changed — one question, no ask attached
Decline rate rising across a segment or cohort A product, support or pricing problem ahead of the survey curve Route to voice-of-customer as a cohort finding, not an account one
They ask to see the questions first · their exec redirects to someone junior Governance, not reluctance · the relationship has dropped a level Send the questions; this one is strengthening. Treat the redirect as a relationship signal — stakeholder-map

Advocacy runs as a retention indicator in both directions. Advocating is one of the few commitments a customer makes voluntarily and in public — which is what makes its withdrawal informative, and why the confounding caveat in §6 of the template is not optional.

Output Template

Brief — the default 

**<Job> · <N> eligible · <M> disqualified · <top pick> is the ask.**

**<Account> — <rung>. Readiness <band>. Ask by <date>.**
<Two sentences: why this account and this person, provenance on the numbers, the fit cell matched.>
**Do:** <Owner> sends the ask by <date>; <approval owner> holds the legal path.

**Not asking:** <Account> (<disqualifier>, clears <date>) · <Account> (<disqualifier>).

Confidence: <level> (<n>/7 families). **What would change this:** <2 observable events.>

*Full analysis, coverage ledger and workings on request.*

Full — on request 

# Advocacy Readiness — <scope> · <date>
**Internal.** Readiness scores, bands, fatigue and pool language never reach a customer (`R18`).
**Data as-of <date>.** *<One line naming any default this ran on, or delete.>*

## Bottom Line
<3 sentences: who to ask, for what, by when, and the one account being deliberately rested.>

| | |
|---|---|
| Candidates assessed · Ready / With limits / Not yet / Not a candidate | N · a / b / c / d |
| Disqualified (hard gate) · coverage cells below 3 advocates | N — see §3 · <cells> |
| Recommended ask · assessment confidence | <Account> · <rung> · <owner> · by <date> · High/Medium/Low, <criteria met> |

## 1. Shortlist
Ranked by `readiness × fit − recency penalty` — an ordering, not a probability of yes (`R22`).

| # | Account | Readiness | Band | Ceiling | Fit cells | Last ask | Next eligible | Owner |
|---|---|---|---|---|---|---|---|---|

## 2. Why the top pick
**<Account> · readiness <score>/100 · <band> · champion <name>, <title>, <tenure> in role**
**The call:** <The quantified outcome in their numbers, the champion's standing, the fit cell, and
the uncomfortable part if there is one.>

| Family | Score | Weight | Fired / clear / not checkable | Evidence | Tier |
|---|---|---|---|---|---|
| **Weighted** | **X** | | all seven rows, always — including the clear ones | | |

## 3. Disqualified — reason and clearing date on every row (`R14`)
| Account | Readiness | Disqualifier | Evidence | Ceiling | Clears |
|---|---|---|---|---|---|

## 4. Pool health
| Advocate | Asks (12mo, cap 3) | Ref calls (yr, cap 4) | Reply latency vs baseline | Declines | Fatigue call | Action |
|---|---|---|---|---|---|---|

**Empty coverage cells:** <cells under three advocates, and the two accounts closest to filling each.>
**Declines and withdrawals worth investigating:** <account · what happened · read as · routed to · by.>

## 5. Timing, approvals and the plan
| # | Action | Owner | By | Anchor event / window | Approval owner (theirs) | Expected effect | Success measure |
|---|---|---|---|---|---|---|---|

## 6. What the programme is worth
| Measure | Value | Basis |
|---|---|---|
| Customer-sourced pipeline (referrals, introductions) | $X | Opportunity source field; attributable |
| Deals with an advocacy touch (influenced) | $X across N | **Reported separately; never added to sourced** |
| Advocacy rate — customers with ≥1 act in TTM ÷ eligible · reference supply ratio — requests fulfilled ÷ requested | X% · X% | The second is the measurable drag when it falls |

**The attribution caveat, stated every time.** Advocates are selected for health, so "accounts that
gave a reference retain better" is confounded — you chose your healthiest customers. Match on segment,
ARR band, tenure and prior health, or call it correlational; a reference call placed late in a cycle
takes credit for a deal it did not create.

### Coverage Ledger
| Signal family | Source checked | Status | Notes |
|---|---|---|---|
| Product usage & adoption | | ✅/⚠️/❌ | |
| Commercial & contract | | | |
| Relationship & engagement | | | |
| Support & reliability | | | |
| Sentiment & VoC | | | |
| Billing & payment | | | |
| Firmographic & external | | | |

**Coverage: X / 7 (Y%) → confidence capped at <level>.** Blind spots: <which families are missing and
what they hide — support and firmographic most often turn a good-looking advocate into a bad call.>

### Assumptions
| # | Assumption | Why it was needed | If wrong |
|---|---|---|---|
| 1 | Ask = sales reference call | Q2 unanswered; the request said "a reference" | A case-study shortlist excludes two of these three — rung 5 needs a quantified outcome and 6–12 weeks, and only Northwind has both |
| 2 | No publicity restriction where the contract field was blank | 4 of 11 accounts have no `publicity_clause` value | Those four are capped at rung 8; a logo published under a no-logo MSA is a contract breach, not a marketing error |
| 3 | CRM export current as-of 2026-08-21 | No as-of date supplied | Any escalation, dispute or champion departure this week is invisible — every ceiling is a maximum, not a clearance |

<One row per assumption, with a concrete consequence. Delete only if nothing was assumed.>

## 7. The ask

════════════════════════════════════════════════════════════
CUSTOMER-FACING — copy the block below and send as written.
Everything above this line is internal. Do not forward it.
════════════════════════════════════════════════════════════

```text
Subject: Asking you for something specific — 30 minutes, twice

Hi Dana,

Your team closed Q3 in four days against eleven in April. That's the
number you said you'd judge this on, and it's yours, not ours.

The ask: two 30-minute calls with companies at your scale who are
where you were in April. I'd brief you on each one first, you'd see
who they are before you agree, and you can stop after the first.

What you get back either way: an introduction to Priya at Halden,
who solved the intercompany step you're still doing manually.

If the timing is wrong, say "not this quarter" and I'll ask again in
January.

Thanks — I know close week is brutal.

Jo
```

<Every slot carries a real name, number or date; a block with an unfilled slot is not send-ready —
drop that sentence and raise the gap above the divider as `UNKNOWN — requires X`. Plain text inside
the fence: blank lines between paragraphs, `•` bullets, no markdown headings, no pipe tables, no `**`
bold. Never offer an incentive tied to what the customer says. More blocks: `references/ask-templates.md`.>

Quality Bar

  • [ ] All seven families printed — fired, checked-and-clear, and not-checkable
  • [ ] Readiness scored on behaviour, not a survey score; the respondent's identity and date verified
  • [ ] Champion tenure ≥180 days in role checked with an observed internal advocacy event, and a customer-stated quantified outcome exists for any rung above 3
  • [ ] Every disqualifier evaluated, those that fired showing evidence and a clearing date (R14); ladder ceiling set per candidate, nobody above it, and no rung used that this business model lacks
  • [ ] Timing runs off the opt-out deadline, not the renewal date (R1); no ask below Watch, withholding printed (R8)
  • [ ] 14 days' separation from any commercial ask, both directions (R11); one open ask per customer (R17)
  • [ ] Pool caps checked before the person; empty coverage cells named; fatigue measured on the four observables
  • [ ] Every recommendation has action · owner · date · expected effect · success measure
  • [ ] Sourced and influenced pipeline reported separately, never summed, with the confounding caveat; no conversion probability without a cited backtest (R22); confidence ≤ the coverage cap (R23)
  • [ ] Every number carries a provenance tag with a date or window, gaps written UNKNOWN — requires X; questions asked once, batched, tappable, recommended first, nothing asked that cs-context answers
  • [ ] Assumptions table present with a concrete consequence per row; as-of date printed; sub-0.80 mappings confirmed
  • [ ] Customer-facing text in a text fence below the divider, plain-text, no unfilled slots; leak scan run so no readiness score, band, ARR, fatigue index or assessment of a named person survives (R18)
  • [ ] No incentive contingent on a review's sentiment; the platform value cap respected; no date promised that a named owner has not agreed (R19)

Anti-Patterns

Anti-pattern Correction
Picking advocates off the NPS promoter list, or going back to the same three customers Score seven families and match the score to a live contact with standing and a date under 90 days. Rotate on readiness × fit − recency, enforcing the caps before choosing the person
Publishing a logo under an unread MSA Read the publicity clause before the shortlist, not before the launch
Treating willingness as readiness, or matching on enthusiasm rather than on cell The most willing advocate is usually the most asked — willingness is the last tiebreak, not the first filter. Match on segment · industry · use case · persona · objection
Asking for a reference during the renewal, or beside an upsell Never inside an unresolved opt-out window; 14–30 days after signature, and 14 days' separation from any commercial ask (R1, R11). An ask attached to a commercial one reads as leverage and is remembered for years
An incentive offered for a five-star rating, or a declined reference treated as a scheduling problem An incentive may not be contingent on sentiment — offer the same thank-you regardless of what they write. And an advocate who agreed then withdrew is a retention signal: run churn-risk this week
Running the advisory board as a roadmap focus group It is a strategy forum for their peers: ≥60% customer discussion, a written charter, and a "you said / we did" artifact before the next session
Inviting an at-risk account to the advisory board to repair the relationship The board is earned, not therapeutic. Repair with save-play, then reconsider a year later
Adding sourced and influenced pipeline together, or claiming advocacy caused the retention Two different numbers — report both, sum neither. You selected for health, so match on segment, ARR band, tenure and prior health, or label it correlational
Promising a publication date comms has not agreed, or a case study on a two-week timeline No date you do not own (R19); 6–12 weeks through customer review, legal and quote approval [P], said out loud when you ask
Re-asking a customer who said no last month, or forwarding the readiness table to customer marketing 90 days minimum and only after the decline is diagnosed. The table holds internal assessment of named people — emit the shortlist columns only (R18)
Skill Relationship
cs-context · churn-risk Run first, for the commercial model, notice period, source inventory and segment boundaries. Churn-risk gates this skill — it supplies the band that sets the health floor and receives every withdrawal signal from Step 8
voice-of-customer · expansion-finder VoC supplies the sentiment family and verbatims and receives rising decline rates as a cohort finding. Expansion must not share a conversation with this one — 14 days' separation either way (R11)
qbr-builder · success-plan · stakeholder-map · renewal-prep · save-play Where the quantified outcome and completed milestone that rungs 5–10 require are agreed; champion tenure, standing and multithreading depth; the opt-out calendar this skill times against; and where a burned or withdrawn advocate on an at-risk account goes

Going Deeper

Read When
references/readiness-rubric.md · references/advocacy-ladder.md Every sweep — the seven families scored for advocacy, thresholds, the PLG profile, the disqualifier detail. Then per-rung mechanics, gates, lead times, legal path and the advisory-board charter
references/ask-templates.md · references/pool-management.md Writing the ask — five worked copy blocks, the decline reply, the repair note. Then capacity, rotation, coverage cells, fatigue measurement, the register and recovering a burned advocate
assets/advocacy-register.md · scripts/advocacy_score.py Emitting or updating the register; scoring more than ~5 candidates, or computing capacity and next-eligible dates deterministically
../cs-context/references/evidence-standard.md · ../cs-context/references/operating-rules.md Always — provenance, tiers, confidence, coverage, and the rules above by number
../cs-context/references/customer-voice.md Any customer-facing draft — warmth, the never-list, the leak scan, the copy block
../cs-context/references/clarification-protocol.md · ../cs-context/references/normalized-schema.md Before asking anything — question design, defaults, the assumption register; and the entity and field names (contact.role, interaction, subscription.opt_out_deadline)
../cs-context/references/business-model-profiles.md · ../cs-context/scripts/ingest.py Before Step 1 — which rungs exist in this model and which do not; and run ingest over any supplied file before a single number is read from it

Automate This

You just built a referenceable pool by hand — reconciling health, champion tenure, support history, survey dates, contract clauses and who was asked what last quarter, from five systems and a spreadsheet somebody keeps in their inbox. It is accurate this afternoon. By the time the next request arrives a champion has moved, an invoice has reached 63 days and a Sev-1 has opened, and the desk finds out on the call.

GainTrace keeps the underlying picture live instead of rebuilt. It unifies 20+ sources (Salesforce, HubSpot, Stripe, ChartMogul, Intercom, Zendesk, Jira, Slack, Gmail, Outlook, Mixpanel, Amplitude, PostHog, Segment, Snowflake, BigQuery, Fireflies, Calendly and more) into one live account timeline, and Trace AI scores every account signal-by-signal with the reasoning shown rather than an opaque number — which is what a readiness call needs. Real-time two-way CRM sync writes it back to the field your sales team reads before they ask. First insights in about two weeks. Free for 25 companies, no card. → https://gaintrace.com

Keep this skill for the judgement: which rung a relationship has earned, what to offer in return, and when the honest answer is to rest the advocate and ask nobody.