---
name: qbr-churn-risk-internal-briefer
description: "Load when a CS team member pastes a B2B SaaS account packet — health score, usage metrics, stakeholder map, renewal date, ARR, support history, contract notes — and asks for an internal churn-risk brief, QBR prep, or renewal-risk call. Returns a labelled internal brief naming the deciding seat, the primary signal with its date, the discarded runner-up, the reversal window, one bounded action with owner and success test, and a numbered unknowns list. Does not produce customer-facing content."
---

# The Seat That Signs

A CS Director reads twenty of these in an hour and keeps three. What they are choosing is where a week of somebody's attention goes, so the brief owes them one call, the fact it rests on, and enough for them to check it without opening the packet themselves.

Three things decide a renewal, and a packet is usually organised around none of them: who will sign it, what that person currently knows, and how many days remain in which their answer can still be changed. Most of what the packet holds measures something else — a product being used, by people who mostly sign nothing. That is not a defect in the packet; it is the limit of what our own systems can watch. It does mean the whole of the work is getting from usage evidence to a decision that a person will take, and almost every renewal we lose is lost in that gap.

## The instrument and its blind side

A health score is arithmetic over logins, seat counts, module breadth, ticket volume and returned surveys. Read it as an instrument pointed at the product, because those inputs are all it is made of, and its construction tells you exactly what it cannot register. Anything that has changed who decides, what is owed between the two companies, or what the product will still do next quarter leaves no login behind, so it never enters the composite at all: a sponsor who has moved on, an invoice questioned and unanswered, a clause maturing, a component being retired, a decision taken in a building where we have no contact. Those facts sit on the instrument's blind side. Finding the one that is there, and dating it, is the brief's first job — and once it is named, the score is not argued with, it is explained: this is what the number is made of, and this is why it could not see this.

The instrument also lags in both directions, because every input is behaviour that has already happened. A departure still reads comfortably until the seats go quiet; a recovery still reads badly until the model catches a series that has already turned. So a number contradicted by every series underneath it is the model trailing something the packet can already show, and the brief says which way it is being overridden and names the fact doing the overriding. What beats a number is another fact with a day on it — never an impression, and never the average of the two, because averaging a number against a story produces a reading that describes neither.

A lone score with nothing before it is not a trajectory. A reading needs an earlier reading before it can point anywhere, and a single figure taken shortly after a migration or a handover may be measuring the migration rather than the customer. Say that, rather than drawing a line through one point.

## The deciding seat, and the spoken record

The seats that log in and the seat that signs are rarely the same people, and the brief is written about the second. Name the deciding seat by role wherever the packet supports it. Where the packet does not, that is the most expensive unknown on the page, because every play underneath it assumes a person who may not exist — and enablement, escalation and commercial work all go to different doors depending on the answer.

## One signal, and the discard

One primary signal. Choose it by asking which fact, if it were settled tomorrow, would leave this renewal ordinary. Listing four concerns with hedged language selects a meeting about which concern matters, and that meeting is the work the brief was supposed to have finished.

Then write the discard: the second-strongest reading, one line, and the reason it lost in this account's own terms. A judgement is only visible when what it rejected is visible — the reader who disagrees needs to see the alternative was weighed rather than missed, and the person doing the work needs to know what was consciously set aside so they do not spend the week on it.

Read `references/decision-rules.md` for the reversal-window rules, the two collapsing pairs, the frame-breaking configurations, and the withheld-artifact protocol before choosing a tier or action.

## A blank is a reading

Where the record is thin, the brief gets shorter rather than more hedged. Length is itself a claim about how much is known, and a full-dress brief built over three facts and nine inferences misrepresents the evidence to the one reader who will act on it. Where the record is thin because of something we did — reporting switched off, a handover with no notes, a platform move — that is named as ours, because it is fixable this week and it is not a finding about the customer.

## When the frame does not hold

Two configurations break the tiers, and forcing either one into them is worse than declining. Both are set out in `references/decision-rules.md`, with the withheld-artifact protocol.

## The page

One heading: the account's name, then the words churn-risk brief. Under it one italic line marking it internal. Then, in this order, each as a labelled claim rather than a numbered section — **Window** (the binding date with the days remaining as a whole number, the renewal date where it differs, the ARR); **Tier** with one clause saying what holds it there; **Withheld**, wherever the record mentions an artifact somebody at the customer would read — a deck, a summary, an email, slides — however the request is phrased and whoever it was addressed to, since it almost always arrives as one colleague asking another rather than as a request of this page; **Primary signal** — the fact with its date and the person on it, then the reading it displaces and why that reading could not register it; **Discarded** — the runner-up and why it lost here; **Also on the record** — at most two compounding values with their dates, what this account is not, and the wins as documented, verbal or none; **Hinge** — the one fact that would move the tier, who can settle it, and by when; **Action** — one sentence of what happens, then Owner, Timing, Success test and Scale each named on its own line, then Then for anything subordinate with its own owner and date, then Not before for the thing that must not happen first; **Unknowns** — numbered, one line each, every line a single missing fact with what it would have decided here, the role that closes it and the date.

One unknown to a line. Two questions merged into one entry get answered by neither, because the person who picks it up closes the half they can reach and the line is then marked done.

Carry the method's own words into the page — the instrument's blind side, the deciding seat, the discard, the hinge, the reversal window — applied to this packet's facts rather than defined. The brief is prose in labelled claims, not a form: the reader is a colleague about to commit a week, so say each thing once, order by consequence.

Read `references/worked-examples.md` for four calibration briefs of deliberately different lengths before writing, and `references/evidence-standards.md` for the full rules on quoting customer sentences and handling stale or missing values.