Client Qbr

Client QBR

One quarter, one client, four questions: what did we promise, what did they get, what happens next, and what are we asking for.

The QBR is the only recurring moment where the firm has demonstrated value on the table and the client's attention at the same time. That is why the expansion ask and the referral ask belong here and not in a standalone workflow — asking before you have shown the quarter's results is the version that damages the relationship. Firms that run structured quarterly reviews grow account value materially faster than firms that only talk to clients when something is wrong.

Read First

1. workspace/clients/{slug}/!_account.md — Client Context Gate applies.
2. workspace/clients/{slug}/onboarding.md — the success definition and each stakeholder's criteria. This is the scorecard. A QBR that invents its own criteria this quarter is a status update.
3. The previous workspace/clients/{slug}/qbr/ file — what was promised last quarter.
4. workspace/clients/{slug}/account-plan.md — the expansion hypotheses and triggers.
5. workspace/firm/proof/ — records for this client, and comparable results.
6. workspace/clients/{slug}/notes.md — relationship signals from the quarter.

When to Use

• Quarterly review with an active client.
• Renewal approaching — run this before the renewal conversation, not during it.
• The firm has not reviewed an account in two quarters, which the renewal review rule in AGENTS.md treats as a trigger.

Delivered Versus Promised

The core section, and the one with the strongest temptation to drift.

Take last quarter's commitments — from the previous QBR, or from onboarding.md on the first one — and mark each: delivered, partial, not delivered, or descoped by agreement. Then score the success definition itself: on_track, at_risk, or not_measurable_yet.

Rules that make this credible:

1. A bad quarter is written as a bad quarter. A client who lived through a missed commitment and reads that it was delivered stops trusting every other line. Firms lose accounts they were serving well by reporting them optimistically.
2. Never invent a result. Every number comes from a proof record, the client's own system, or a source the client can check. No source, no number — the Proof Gate is not relaxed because the audience is the client who lived it.
3. Outputs are not outcomes. "Shipped 14 deliverables" is activity. The success definition is the outcome. If the outcome is not measurable yet, say not_measurable_yet rather than substituting activity for it.
4. Firm-caused misses get named as firm-caused. Attributing everything to client dependencies is transparent and expensive.
5. not_measurable_yet is honest early and evasive late. If it is still that at the third QBR, the engagement has a measurement problem worth raising.

The Two Asks

Both go at the end, both are optional, and both are earned by the section above.

Expansion ask. At most one, taken from account-plan.md, and only if its trigger has actually fired. Frame it as the client's problem, not the firm's service. If the quarter was scored at_risk, there is no expansion ask this quarter — fix delivery first. Naming that skip explicitly is the point; an agent that asks anyway is the failure mode.

Referral ask. The highest-return thing in this file. Most B2B purchases begin with a referral, referred deals close in a fraction of the time of cold-sourced ones, and firms rate referrals as their most effective channel by a wide margin over outbound — yet most never systematically ask. The repo automates the channel that works worst and, without this, ignores the one that works best.

Preconditions, all required:

• the quarter scored on_track;
• referenceable is not false in !_account.md;
• a specific, easy ask — a named person or a described situation, not "know anyone?"

Ask for one introduction, not a list. Make it forwardable: two or three sentences the client can pass on without rewriting, plus a plain description of who the firm helps. The client's effort is the constraint, not their goodwill.

If a precondition fails, record referral_ask: skipped and the reason. Asking a client whose quarter went badly is how a recoverable account becomes a lost one.

Workflow

1. Check the gate. Missing !_account.md or onboarding.md → run client-onboarding.
2. Assemble last quarter's commitments and mark each honestly.
3. Score the success definition. Note per-stakeholder criteria that diverge — the champion and the user often disagree, and renewal follows whoever holds budget.
4. Pull real results only. Anything unsourced becomes an open question for the client.
5. Decide the next quarter: at most three commitments, each with a date and an owner.
6. Decide the two asks against their preconditions, and record skips with reasons.
7. Write the file, then update !_account.md: health, health_reason, reviewed_at, last_contact, and referenceable if the client agreed.
8. If this client's proof record is still an empty stub and the quarter produced confirmable numbers, say so and offer sales-outcome-log to fill it. A stub nobody returns to is the documented failure mode of the proof library, and the QBR is the natural moment it gets fixed — the numbers are already on the table.

Output

workspace/clients/{slug}/qbr/{YYYY}-Q{N}.md

---
client: northwind-logistics
period: 2026-Q4
success_definition_status: on_track   # on_track | at_risk | not_measurable_yet
commitments_delivered: 3
commitments_missed: 1
health: green                          # green | amber | red
expansion_ask: release-automation      # service slug, or: skipped
expansion_ask_skipped_reason:
referral_ask: made                     # made | skipped
referral_ask_skipped_reason:
proof_refs: [northwind-release-automation]
next_review: 2027-01-15
generated: 2026-12-18
---

Body: What we promised, What was delivered (with the honest marks), Success definition status, Next quarter, Where we need you (client-side dependencies), Expansion if any, Referral if any.

This document is client-facing. Render it with tool-weasyprint if it is being sent:

uv run --with markdown --with weasyprint python3 \
  .agents/skills/tool-weasyprint/scripts/md-to-pdf.py \
  workspace/clients/{slug}/qbr/{YYYY}-Q{N}.md

Internal risk assessment stays in account-plan.md and notes.md, not here. Keep the blunt language where the client will not read it.

Testing Requirements

1. Run with one missed commitment; confirm it is marked missed, not softened, and that firm-caused attribution is stated.
2. Score a quarter at_risk; confirm no expansion ask and no referral ask are made, and both skips carry reasons.
3. Run with referenceable: false; confirm no referral ask regardless of the score.
4. Confirm every number in the file traces to a proof record or a client-side source.
5. Confirm an empty proof stub for this client triggers the offer to fill it.
6. Render the PDF; confirm no internal risk language appears in it.

Related Skills