When money, terms or leverage are on the table at a renewal and the user has to decide what to ask for, what to trade, and what to refuse — uplift, discounts, term, payment terms, downgrades and procurement. Also use when the user mentions 'discount to renew', 'asking for 20% off', 'pushing on price', 'procurement just got involved', 'how much can I raise their price', 'price increase at renewal', 'the uplift clause', 'they got a cheaper quote', 'best and final', 'they want to cut seats', 'should we go multi-year', 'what do I trade for that', 'my walk-away', 'they want net 60', or 'what should I concede'. Use this whenever a commercial demand has landed or a renewal price has to be set, even if they never say 'negotiation'. For the T-180 runbook and paper path, see renewal-prep. For an account already in rescue, see save-play. For the risk read, see churn-risk. For sizing an upsell, see expansion-finder.
Renewal Negotiation — position, uplift, and the concession ladder
You decide what this renewal is worth, what we will trade to keep it, and what we will not trade at
any price — and you write all three down before the first number is said aloud. renewal-prep gets
the renewal to the table on time. This skill decides what happens at the table.
The rookie arrives with a percentage and no position. The customer says "we need 20% off"; the
rookie negotiates the percentage. Twenty becomes twelve, twelve becomes eight, and every step down
is given away for nothing because there was never a written list of what we wanted in return. The
mirror image is an uplift asserted at T-30 with no value artifact behind it, raised for the first
time inside the customer's own notice window.
The elite version builds the position before the price, then gates it: every give is attached to a
get in the same row, no rung is offered to someone who cannot sign, no concession is discussed while
the only live thread runs through procurement, and the number is announced rather than requested.
Read ../cs-context/references/evidence-standard.md first.
Before Starting
Read .agents/cs-context.md (fallback .claude/cs-context.md). If absent, run cs-context
first — without §2 Commercial Model this skill guesses at the standard uplift, the discounting
norms and the payment terms it exists to defend. Never ask what that file answers: ARR,
renewal date, notice period, standard uplift, discounting norms, payment terms, segment, owner,
fiscal year, source inventory.
Take the clause stack from the executed contract, not the CRM's opinion of it. Every field is
a lever, and a wrong one is a lever pulled the wrong way in front of procurement.
Input
Field
If missing
Renewal date · notice period
subscription.renewal_date · notice_period_days
UNKNOWN — requires the executed contract; every timing rule below becomes a ceiling, not a plan
Uplift clause, cap, index · auto-renew and last change
subscription.uplift_pct + clause text · auto_renew · auto_renew_changed_at
No clause means no contractual rung, so justify on value or hold; an auto-renew change is a decision, not a posture (R2) — route to save-play before pricing anything
signs from stakeholder-map — the person on the order form, not the person on the call
Gate A fails and the ladder is blocked (Step 4). The recommended action becomes the authority test
Last non-commercial conversation · health band
Latest interaction.occurred_at whose subject was outcomes, adoption, roadmap, support or the success plan, with a counterpart outside procurement and legal · churn-risk output
Treated as older than 21 days and Gate B fails closed (Step 4); run churn-risk first, because an uplift on an At Risk account is a different decision
Compute two fields before anything else, and print both — scripts/concession_math.py gates,
which also returns Gate A and Gate B. Neither is left to the writer to remember.
Field
Formula
Why it is computed, not recalled
D — the governing date
renewal_date − notice_period_days (R1)
Where today sits against D changes the recommended position, not merely the schedule
days_since_business_thread
today − last non-commercial conversation; UNKNOWN counts as over the limit
Past 21 days with procurement active, the ladder is blocked and re-opening a business thread becomes a required action (Step 4, C10)
Ask these four once, tappably, in one batch, then run unattended.AskUserQuestion, 2–4
mutually exclusive options each, recommended first and labelled, one line under each saying what
it changes. Drop any question cs-context or the request already answers.
Header
Question
Options — recommended first, each with what it changes
Stage
Where is this negotiation?
Setting the position (Recommended) — full pack: uplift decision, gates, costed ladder, timing, comms · A demand has landed — straight to the gates, the counter and the reply draft · Structuring a multi-year or restructure — structures with the arithmetic and the terms to hold · Signed or lost, debrief it — what was conceded, what it cost, what to change
Position
What are we going for on price? (skip if cs-context §2 sets a standing uplift and this account is standard)
Standard uplift with the value artifact (Recommended) — clause plus delivered outcomes · Hold flat, traded — the hold buys term, reference rights or prepayment · Above-standard or restructure — needs the usage delta or an evidenced below-market entry price · Not decided — schedules the T-120 decision instead of assuming one
Ceiling
What can you approve without going anywhere? (skip if cs-context §2 names the bands)
Up to 5% (Recommended) — ladder stops at rung 4 · Up to 15% — manager level, rungs 5–6 open · Above 15% — VP or deal desk, rungs 7–8 with written justification · Unknown — every rung past 4 prints approval required — name the approver
Output
Who reads this?
Me, before the call (Recommended) — internal pack plus the talk track · Pack plus the customer drafts — adds the send-ready blocks in §15 · My manager or deal desk — adds the approval ask, the walk-away rationale and the precedent note
Never block, never guess. Every missing input resolves one of three ways — read it (showing
the derivation), ask it (only where two answers produce a materially different position), or
mark itUNKNOWN — requires <source> with a confidence cap. A plausible substituted value
becomes a fabricated one the moment it is quoted to procurement. If nothing comes back, run on the
recommended defaults, state them above the Bottom Line, and give each an Assumptions row
(../cs-context/references/clarification-protocol.md).
Take whatever data arrives — CSV, XLSX, JSON, warehouse results, a pasted contract clause, a
procurement thread, a transcript, or no file at all. Run ../cs-context/scripts/ingest.py first
on every supplied file. Confirm every mapping below 0.80, and arr, discount_pct,
uplift_pct, notice_period_days at any confidence — a column mapped one field left produces a
walk-away that is confidently wrong. Degrade, never refuse: under 40% coverage, name the gap
instead of setting a price, and never assume an export is current.
How This Skill Works
Mode
When
Produces
Position (default)
T-120 → T-90, before any number is exchanged
Value, price position, their alternatives, switching cost, walk-away, ask stack beyond price, the two gates, costed ladder, timing, opening comms
Live demand
A discount, downgrade, best-and-final or procurement email has landed
The gates, what they actually asked, what it costs, the counter with its required get, the approval needed, the reply draft
Structure · Debrief
Multi-year, ramp, co-term or restructure on the table; or signed or lost, within 5 working days
Structures compared with the arithmetic, uplift schedule, terms to hold, forecasting consequence; or the gate log, what was conceded, what each cost, what it bought, the precedent set, the one change before next cycle
Brief (default output)
Always, unless depth is asked for
≤20 lines: position, number, the one reason it holds, gate status, the next move with owner and date, confidence in three words, the falsifier, then Full position pack, ladder and workings on request. Brief drops the display of the reasoning, never the reasoning
Full (output on request)
Going to a deal desk or an approver, or someone will challenge the number
The complete Output Template
The ethics line — not decoration
A renewal won by trickery is next year's churn — and the churn nobody will tell you the real reason for.
Never
Instead
Rely on the customer missing their notice window
Send the courtesy notice-deadline letter at T-120, repeat at T-90, naming the date in writing
Manufacture a deadline — "this pricing expires Friday" when it does not
Anchor on D, a real date they can verify in their own contract
Withhold data as leverage, or bury an increase in an amendment with no covering note
Make leaving mechanically easy — the only thing that makes a win-back possible; and name the increase, the amount, the effective date and the reason in the email, before the paper
Auto-renewal mechanics are a legal question, not a negotiating one — New York GOL §5-903, and US
federal negative-option rulemaking that is currently unsettled. Route it to counsel, and read
the statutory position in references/procurement-tactics.md before answering a notice question.
The rules this skill enforces
From ../cs-context/references/operating-rules.md, enforced in the output. A deviation states its number, the circumstance, and what will be watched.
Rule
Enforced how
R1 · The Opt-Out Calendar
Every timing decision is an offset from D. Inside D the recommended uplift default changes (Step 3)
R2 · Decisions Beat Indicators
Auto-renew off or notice served stops the pricing conversation and routes to save-play
R11 · Value First · R12 · Co-Term
Value at T-150/T-120, price at T-90, and never an ask in a meeting carrying a miss or an apology; expansion inside 90 days of D co-terms, outside it runs separately and never as ladder currency
R18 · The Firewall
Walk-away, rung, gate status, approval band and any assessment of a named person never reach customer text
R21 · The Stop-Loss
Every position carries a walk-away and an exit date, set before the first call
R22 · Ordering · R23 · Coverage Cap
Break-even and cost figures are arithmetic, never a churn probability without a cited backtest; and confidence never exceeds coverage across the seven families
Business model, and what the seven families decide
Resolve the profile from ../cs-context/references/business-model-profiles.md before Step 1.
Consumption: no seat price to discount — the negotiation is the commitment tier and the overage rate.
Product-led: below the enterprise threshold there is no negotiation, and inventing one creates a
market for complaining. Monthly evergreen: no notice window, so no D. Regulated/public sector:
price may be framework-fixed; the negotiable surface is scope.
All seven families are checked every time, including the clean ones. Product usage & adoption decides whether
the value evidence is real and whether a downgrade ask is genuine · commercial & contract holds the clause
stack · relationship & engagement supplies signs and days_since_business_thread, which together decide
whether the ladder opens at all · support & reliability decides whether an increase is defensible this cycle
· sentiment & VoC prices reference rights · billing & payment prices the payment-terms rungs · firmographic
& external gives their budget cycle.
Run sequence: position → price → time the ask → gate the ladder → cost the ladder → discount discipline → structure → procurement → downgrade and the competitive line → comms, then debrief.
Step 1 — Build the position before you name a number
Six elements. A position missing one of them is a price with a story attached.
#
Element
What it must contain
If missing
1
Value delivered
Three outcomes in their own metrics: baseline · current · delta · a customer-side validator, provenance-tagged
The only rung left is the contract clause, the weakest one. Build it via value-case
2–3
Price position and their alternatives
What this account pays per unit against our own realised book; and the named alternative, its status, what it replaces and what it does not
UNKNOWN — requires realised ACV per unit by segment; never quote a public benchmark as our price position, and if nobody has asked about the alternative, ask — an unnamed "the market" is not one
4
Switching cost
Computed from their numbers (below)
UNKNOWN — requires integration inventory and trained-user count. Never a market average
5
Walk-away
The number and the terms below which we do not sign, approved before the first conversation (R21)
A walk-away invented mid-call is not a walk-away, it is a feeling
6
Ask stack beyond price
Term, multi-year, reference rights, case study, advisory seat, expansion path, payment timing, exec sponsor access, logo rights
Every concession becomes a pure loss, because there is nothing to trade it for
Switching cost, from their own numbers — the strongest figure in the room, and almost nobody brings
it: integration rebuild + data migration + retraining + parallel run + process rewrite, each priced
from their integration inventory, record counts, trained-user count and loaded rates (arithmetic in
references/deal-structures.md). Every input comes from their data or is marked UNKNOWN — requires X; never quote a published switching-cost average, which procurement dismantles in one question.
Exit criteria: six elements populated or marked, walk-away signed off by a named approver on a date.
Step 2 — Choose the uplift and pick the rung that justifies it
Decide internally at T-120. Pick exactly one rung — 0 hold flat (traded, never free) · 1
contractual index · 2 standard uplift · 3 above-standard · 4 restructure — and be able to produce its
evidence in the meeting, not after it. Rung 2 needs the clause and a documented value-realisation
summary; an uplift asserted without the value artifact is a price increase in search of a reason.
Rungs 3 and 4 need deal-desk approval. Full rung table: references/uplift-justification.md.
The justification stack, in this order — the sequence is the argument. (1) Outcomes in their
metrics. (2) Product shipped since the last renewal and actually adopted. (3) Usage growth and
entitlement true-up. (4) The contractual clause. (5) List-price movement. Leading with 4 or 5 tells
procurement this is a commercial fight, and they are better at those than you are.
Standard negotiated annual uplifts of 3–5%, often pegged to CPI or a fixed percentage "whichever is
higher", are reported by SaaS contracting and procurement guides (2026) — [V] practitioner guidance,
not a measured benchmark. Vendor price indices reporting double-digit averages run on unaudited
methodology [V]; never quote one to a customer. An uplift of u is value-neutral at u / (1 + u)
of added churn probability — scripts/concession_math.py uplift, arithmetic for ordering and not a
forecast (R22).
Step 3 — Time the ask
Situation
The rule
If you are already past it
Standard renewal
Decide by T-120, deliver with the proposal at T-90, negotiate T-60 → T-45
Compress, do not skip — the value artifact still lands first
Value and price · after a miss or apology
Value at T-150/T-120, price at T-90, never in one conversation, and never an ask in a meeting carrying a miss (R11)
Split across two days and say aloud which conversation this is
Inside the opt-out window
Never raise an increase for the first time inside D.
Hold flat this cycle and put the increase in writing for the next one with its effective date named. State the deviation under R1
Expansion alongside the renewal
Inside 90 days of D, co-term it (R12)
Outside 90 days run it separately — and never use an expansion as ladder currency
Step 4 — Gate the ladder before you cost it
Two refusal conditions, run before any rung is priced. Both fail closed, and both print in the pack
with their evidence whether they pass or fail.
signs names a person and that person is in the conversation, or has confirmed in writing the terms they will sign
Every Offer? cell in §3 reads BLOCKED — Gate A. The recommended action is the authority test, not the concession
B · Business thread (C10)
days_since_business_thread ≤ 21, or procurement is not active
Every Offer? cell reads BLOCKED — Gate B. The recommended action is re-opening a business thread, with an owner and a date, before any concession is discussed
A pack that offers a rung under a failed gate is invalid output. The costed ladder is still
built — an approver needs it — but no rung is offered, and §4's counter carries the gate's action.
The authority test, said before any rung is priced. A concession given to someone without
signing authority is spent twice: once with the person who cannot sign, and again with the person
who can, who now opens from your reduced number.
"If I could get that approved, is this something you could sign this quarter?"
Where signs is UNKNOWN the question is who, not whether: "Before I take a number to our deal
desk — who signs the order form on your side, and are they behind the figure you have asked me
for?" An answer of "I would still need to take it to Dana" is a blocked gate, not a soft yes.
Re-opening the business thread is a named meeting, not a check-in. A named person, a subject
drawn from their own data, a date, and no price in it. The commercial reply goes out once that
meeting is booked, not once it has happened, and procurement is copied — a second thread, not a
secret one. Both gates in full with the three plays: references/concession-ladder.md.
Step 5 — Cost the ladder, and never climb a rung without its get
Cheapest to most expensive, costed in dollars for this account before the call —
scripts/concession_math.py ladder.
#
The give
What we get
Approver
1
Payment terms — Net 30 → Net 45/60
Signature by a named date
CSM / RM
2
Waive or reduce the uplift
Multi-year term, or a two-year price lock in our favour
RM
3
Enablement, training credits, premium-support trial, advisory seat
Exec sponsor meeting plus a written success plan
Manager
4
0–5% discount
Named signature date plus reference rights (two calls a quarter, one case study a year)
Rep / RM
5
5–10% discount
Multi-year commitment, or increased scope or seats at list
Manager
6
10–15% discount
Multi-year plus prepayment, annual or full-term upfront
Manager / Director
7
15–25% discount
Multi-year plus prepay plus expanded scope plus logo and case-study rights
VP
8
>25%, or any unprecedented term
Strategic justification, a board-visible logo, or a documented displacement threat
C-level / deal desk exception
A row with an empty What-we-get cell is invalid output (C12). Not incomplete — invalid: delete
the row or name the get. There is no — and no n/a, and four things read as gets without being
one — goodwill or the relationship (unpaperable) · "they renew" (the renewal cannot be the
payment for the renewal) · "they stop asking" (you bought quiet and set next year's floor) · an
expansion inside 90 days of D (it co-terms by R12).
Five rules make the ladder work: one rung at a time, never two in a meeting · never open at the
walk-away · ask what the constraint is before pricing it, since it is often timing, budget
period or scope, none of which cost margin · non-price levers first · log every rung offered,
since an offered rung is next year's floor whether or not it was taken.
Never without VP+ approval — these four set next year's leverage: the notice window, the
auto-renew clause, the uplift clause, audit and true-up rights. Never at any level: MFN
pricing · unlimited liability · TfC with no fee or notice · source-code escrow without a deal-desk
exception · unmetered usage · a permanent base reprice disguised as a one-time credit. Per-rung
scripts and the What-we-get catalogue: references/concession-ladder.md.
Step 6 — Discount discipline
Compute the lifetime cost before offering a point — scripts/concession_math.py discount. A discount
decides every year they stay, plus the escalator you no longer apply to the reduced base; round to two
significant figures (§4F). The precedent is the real price: it is the floor they negotiate from next
cycle, and it travels through procurement networks and benchmarking services that exist to collect it [P].
A one-time concession that does not reprice the relationship needs all six: list price stated and
unchanged on the order form · a named credit line with an end date · snap-back language in the same
document · a get named in the same paper (C12) · the words "one-time" in the covering email as well
as the contract · logged as value_delta_reason = discount_concession. Fewer than six and it is a new
price with an optimistic note attached. When something must be given, in order of preference: payment
timing → scope or entitlement → a term-limited credit → a percentage off the base; only the last is
permanent by default. Detail: references/concession-ladder.md.
Step 7 — Structure the deal
Six structures, their arithmetic and their traps: references/deal-structures.md. Four rules hold
whichever you pick. Write each year's fee in dollars on the order form, not as a formula.A
multi-year with an annual opt-out is not a multi-year: every anniversary is its own renewal event with
its own D (R1), forecast and staffed as annual. A ramp step is tied to a milestone you can
observe, or it is a discount with a delay. Payment terms are money: 30 days of extension costs
about ARR × 30/365 × cost of capital, and prepayment is the same in reverse.
Step 8 — Handle procurement
Procurement is not your enemy and not the decision-maker. They are measured on savings against a
baseline, cycle time and risk terms accepted — so give them something to save against, early, and
never let list price be the only anchor in the room. Never let their thread be the only one open
(C10, Gate B).
Their move
The counter
Entering at T-90 or earlier (normal and healthy)
Give everything at once: pricing, entitlement report, security package, insurance certificates, entity and tax data, redlines accepted elsewhere [V]
Entering inside T-30
Name the paper path and its measured lead times, and propose a 30–60 day extension at current terms. That removes the artificial deadline and is a real option, not a bluff
"We have quotes 30% lower"
Ask for a scope-normalised comparison, then price the switching cost from their numbers. Never match a number whose scope you have not seen
"Send your best and final" at T-14, or a deadline in no contract
A best-and-final is conditional on a named signature date and a named get, or it is simply a lower price; anchor timing on D, which they can verify in their own paper
Splitting the CSM from the AE and price-testing each
The RM owns price, the CSM owns value, and neither prices anything in a hallway. Say it plainly
Ten tactics with their tells, the re-opening plays and the paper lead times:
references/procurement-tactics.md. The renewal manager owns procurement, not the CSM[P] —
where one person holds both, run the two conversations on different days.
Step 9 — Test a downgrade, and hold the competitive line
Test
Genuine right-sizing
Negotiation move
Utilisation
Provisioned seats far above active users, unused for two quarters
Utilisation healthy; the ask appeared with the renewal
Who is asking
The admin or budget owner, with a headcount number
Procurement, with a round percentage
When raised
Before the renewal conversation, or at a QBR
First raised at T-30, attached to a deadline
How specific
Named teams, named seats, a date
"About 20%"
If refused
They renew anyway, smaller
Explicit or implied non-renewal
Three or more on the left: take the right-size, and take it well. Median B2B SaaS gross revenue retention
was 84% in the 2026 Benchmarkit B2B SaaS & AI-native metrics report (FY2025 data), down from 88% [V] —
most of that erosion is contraction, not logo loss. Right-size to real usage plus a named growth buffer ·
trade the reduction (C12), since a smaller contract at list is frequently worth more than a larger one
at 15% off · keep entitlement above the point where the use case breaks · put the recovery path in the
paper with a pre-agreed unit price for adds during the term · book it as contraction with
value_delta_reason = seat_reduction, never as a flat renewal, which is how GRR lies to a board. Three or
more on the right: a price negotiation wearing a downgrade's clothes — answer it with Step 5's ladder.
Under competitive threat, do not compete on price. Compete on switching cost, integration depth,
data gravity, and what a migration costs them in the quarter they would have to run it. Ask directly
to be in the evaluation; never disparage, never promise a roadmap date to win a bake-off (R19).
Price-match rule: only where you have seen the scope, and only as a term-limited credit with
written snap-back. If they are genuinely leaving, stop spending (R21) and hand to save-play.
Step 10 — Write the comms, then debrief
Five drafts leave the building, written out in assets/negotiation-comms.md under
../cs-context/references/customer-voice.md: the business-thread re-opener (first, whenever
Gate B failed), the uplift notification (T-90), the discount response, the downgrade
response, and the best-and-final / deadline response.
The price paragraph is constructed, not written. Four slots, and the number is slot 3:
Slot
Contains
Rule
1
What changed, in their numbers, with dates
Two facts, each verifiable in their own systems
2
What we shipped that they adopted
Named feature, named team, when it went live
3
The number
Declarative present tense. Last sentence of the paragraph. Full stop (C3)
4
Effective date, notice date, and the meeting offer
New paragraph. Does not restate the number
C13 · Announce, do not ask. The price paragraph carries no question mark and no request
construction — would you be open to, is there room, we were hoping, would you consider and
their relatives. A price framed as a request invites a counter before the conversation starts.
Correct: "The annual fee for the 2027 term is $514,000."
C3 · Say the number, then stop. Justification precedes the number; nothing follows it in the
same paragraph — no of course, that said, however, we're flexible, happy to discuss,
nothing is set in stone. Softening invites negotiation against yourself; the silence does the
work. Questions are legitimate — the authority test is one — and live in their own paragraph, away
from the number. Both banned lists in full, with worked rewrites:
references/uplift-justification.md §6b.
Every draft is scanned before it is fenced.python3 scripts/pre_send_scan.py draft.txt must exit
0 — it checks C13, C3, the R18 firewall vocabulary and unfilled placeholders — then the
eight-step leak scan in customer-voice.md. A hit is rewritten, not softened. The firewall
(R18) keeps all of this out of customer text in any wording: the walk-away · the rung or the word
"concession" · the gate status · the approval band · health or risk band · revenue at risk · forecast
category · save play · the precedent note · any assessment of a named person on their side.
Debrief within five working days, signed or lost (assets/debrief-template.md): the gate log ·
what was conceded · what each cost · what each bought · the precedent set · what we refused and what
happened · which lead times we mis-estimated · the one process change before next cycle.
Output Template
Brief — the default
**<Account> · ATR $<X> · <position> · decide by <D> (<N> days)**
<Two sentences: the position and the one reason it holds, with provenance on the numbers.>
**Gates:** A <PASS/FAIL — signer, presence> · B <PASS/FAIL — <n> days since business thread>.
**Ask:** <uplift or hold, and the rung.> **Trade:** <first rung → what we get.> **Walk-away:**
<number and terms — internal only.> **Do:** <Owner> <action> by <date>.
Confidence: <level> (<n>/7 families). **What would change this:** <2–3 observable events.>
*Full position pack, ladder and workings on request.*
Where either gate fails, the Do line is that gate's action — the authority test or the
business-thread meeting — and no rung appears in the Trade slot.
Full — on request
The complete pack is assets/position-pack.md, emitted verbatim. Sections in this order: Bottom Line
(carrying signs · decides · influences and days_since_business_thread as header rows) → 1. The
Position → 2. Uplift and Justification → 3. The Gates, then the Ladder → 4. Live Demand →
Counter, or Structures Compared → 5. Timing and Actions → What would change this position →
Assumptions → Coverage Ledger. Two orderings in it are structural: §3 generates the Gates
table before the ladder table and prints Ladder status: OPEN / BLOCKED — under a failed gate every
Offer? cell reads BLOCKED, §4's counter is the gate's action, and §5's first row is the authority
test or the business-thread meeting; and every ladder and counter row carries a non-empty
What-we-get cell (C12), a row without one being deleted rather than shipped.
§15 closes the pack when a draft is due — the only part that leaves the building:
## 15. Customer-Facing Draft
<Only the draft due now. All five written out in `assets/negotiation-comms.md`.>
════════════════════════════════════════════════════════════
CUSTOMER-FACING — copy the block below and send as written.
Everything above this line is internal. Do not forward it.
════════════════════════════════════════════════════════════
**<Draft name> — due <date>, to <recipient>** · `pre_send_scan.py`: PASS
```text
<Send-ready text. Plain, blank line between paragraphs, • bullets, no markdown headings or pipe
tables. Opens on something only this account's data could produce. The price paragraph runs
justification → number, the number is its last sentence, and it carries no question mark. Names the
opt-out date in plain words wherever the draft concerns price. A fence containing [Name] is not send-ready.>
```
Quality Bar
D = renewal_date − notice_period_days computed from the executed contract and printed with days remaining, with no timing decision anchored on the renewal date (R1); all six position elements populated or marked UNKNOWN — requires X; the walk-away carrying a named approver and a date (R21); switching cost computed from the customer's own numbers, never a market average
Exactly one uplift rung chosen, its evidence named, the justification stack ordered value-first (R11), and no increase raised for the first time inside D unless the deviation states R1, the circumstance and what will be watched
C14 — Gate A printed with its evidence. Where signs is UNKNOWN or the signer is not in the conversation, no rung is offered and the recommended action is the authority test in its exact words
C10 — days_since_business_thread computed and printed. Over 21 with procurement active, Gate B is FAIL, the ladder reads BLOCKED, and a dated business-thread meeting is the first action
C12 — every ladder and counter row carries a non-empty What we get cell. A row without one is deleted, not shipped. "Goodwill", "they renew" and "they stop asking" are not gets
C13 and C3 — the price paragraph of every draft is declarative with zero question marks and zero request constructions, justification precedes the number, the number is the last sentence of its paragraph, no softener follows it, and pre_send_scan.py exits 0 on every draft
Every rung costed in dollars for this account, both never-concede lists printed even when nothing was requested, any discount modelled over its lifetime and rounded to two significant figures (§4F), and any one-time concession carrying all six paper requirements including written snap-back
Downgrade requests scored on all five tests before a response is drafted, and booked with the correct value_delta_reason
Every number carries a provenance tag with a date or window, every inference states its rule, every action has action · owner · date · expected effect · success measure, and confidence is stated at or below the Coverage Ledger cap across all seven families (R23) with no churn probability and no certainty language — arithmetic and bands only (R22)
Marked internal (R18); customer drafts sit inside a ```text fence below the divider, email-formatted, zero unfilled placeholders, leak scan run on each, and no walk-away, rung, gate status, approval band or precedent note anywhere in them
Every missing input resolved read it / ask it / mark it; four questions asked once, tappably, batched, with nothing asked that cs-context §2 answers; every default stated above the Bottom Line with an Assumptions row naming a concrete consequence
Any supplied file went through ingest.py; mappings below 0.80 confirmed plus arr, discount_pct, uplift_pct, notice_period_days at any confidence; as-of date printed; business-model profile resolved first, so no seat-price negotiation is proposed on a consumption contract and no negotiation is invented for a self-serve plan
Anti-Patterns
Anti-pattern
Correction
A concession row with an empty get, or a get of "goodwill"
Every give is attached to a get in the same row, or the row is deleted (C12). The renewal cannot be the payment for the renewal
Offering a rung to someone who cannot sign
Run Gate A. Where signs is UNKNOWN or absent, the move is the authority test, not the concession — a discount given to a messenger is spent twice (C14)
Letting procurement be the only live thread
Compute days_since_business_thread. Past 21 days, book a non-commercial meeting with the economic buyer before any concession is discussed (C10)
Asking for the uplift, or softening straight after the price — "would you be open to a small increase?", "…but of course there's flexibility"
Announce it: rationale, then the number, declarative, no question mark in the price paragraph (C13); justification precedes the number, the number ends the paragraph, and nothing follows it (C3) — the silence does the work
A customer draft carrying a rung, a band, a walk-away or [Name]
Fill every slot or drop the sentence; pre_send_scan.py exits 0 before the fence is emitted
Negotiating the percentage they opened with
Negotiate the structure. Ask what the constraint is, then price that
Raising the price for the first time inside the notice window
Decide at T-120, deliver at T-90. Inside D, hold flat and put next cycle's increase in writing with its effective date (R1)
Asserting an uplift with no value artifact, or leading the justification with the clause
Rung 2 needs the clause and the outcomes; value first, clause fourth — leading with the clause starts a commercial fight you will lose
A "one-time" discount with no snap-back, or a multi-year with annual opt-outs treated as a multi-year
Six paper requirements including written snap-back, and a lifetime cost that includes the escalator forgone on the reduced base; an opt-out multi-year is three renewal events with three opt-out deadlines, priced, forecast and staffed as annual (R1)
Fighting every downgrade, or letting their deadline set the calendar
Score the five tests — genuine right-sizing is taken, traded and papered with a recovery path. Anchor timing on D, and where their deadline is artificial propose a 30–60 day extension at current terms
The CSM negotiating price directly, or matching a competitor quote whose scope you have not seen
RM owns price and CSM owns value, run on different days where one person holds both; scope-normalise a rival quote first, then price the switching cost from their numbers
Related Skills
Skill
Relationship
cs-context · renewal-prep
cs-contextruns first, supplying §2 Commercial Model — standard uplift, discounting norms, payment terms, notice period; renewal-prepruns before and around this, owning the T-180→T-0 ladder, gate audit, MEDDPICC-R, paper critical path and the notice-deadline letter, while this skill owns only price and terms
stakeholder-map · churn-risk
Run before.stakeholder-map supplies signs, without which Gate A cannot resolve; the health band decides whether an uplift is on the table at all
save-play
Runs instead once notice is served, auto-renew flips or the account is Critical. Its ladder is a rescue instrument; this one is commercial
value-case · qbr-builder · expansion-finder
The first two supply position element 1 and the business thread Gate B requires (R11); expansion-finder sizes the expansion, and this skill decides whether it co-terms (R12) and forbids it as ladder currency
renewal-forecast · churn-postmortem
renewal-forecastconsumes this — the position sets the delta reason codes and the category, and a downgrade booked as a flat renewal breaks GRR; churn-postmortemruns after a loss, taking the gate log and concession log
Going Deeper
Read
When
references/concession-ladder.md
Steps 4–6 — both gates in full, the What-we-get catalogue, every rung with its cost driver and script, discount discipline, the never-concede register
references/uplift-justification.md
Steps 2 and 10 — rungs and their evidence, cap-and-collar language, and the constructed announcement paragraph with worked rewrites
Step 8 and the moment a procurement address appears — the tactics with their tells, the business-thread re-opening plays, the statutory notice position and the paper lead times; and Step 7 — multi-year, ramp, opt-out, price protection, co-term and TfC, with arithmetic and forecasting consequence
Step 10 — the five scanned send-ready drafts; the debrief within five working days of signature or loss; and the Full output template, emitted verbatim
Before any rung is priced — gates first, then ladder costs, discount lifetime cost, uplift break-ev
…(truncated)
1---2name: renewal-negotiation3description: When money, terms or leverage are on the table at a renewal and the user has to decide what to ask for, what to trade, and what to refuse — uplift, discounts, term, payment terms, downgrades and procurement. Also use when the user mentions 'discount to renew', 'asking for 20% off', 'pushing on price', 'procurement just got involved', 'how much can I raise their price', 'price increase at renewal', 'the uplift clause', 'they got a cheaper quote', 'best and final', 'they want to cut seats', 'should we go multi-year', 'what do I trade for that', 'my walk-away', 'they want net 60', or 'what should I concede'. Use this whenever a commercial demand has landed or a renewal price has to be set, even if they never say 'negotiation'. For the T-180 runbook and paper path, see renewal-prep. For an account already in rescue, see save-play. For the risk read, see churn-risk. For sizing an upsell, see expansion-finder.4license: MIT5---67# Renewal Negotiation — position, uplift, and the concession ladder89You decide what this renewal is worth, what we will trade to keep it, and what we will not trade at10any price — and you write all three down before the first number is said aloud. `renewal-prep` gets11the renewal to the table on time. This skill decides what happens at the table.1213The rookie arrives with a percentage and no position. The customer says "we need 20% off"; the14rookie negotiates the percentage. Twenty becomes twelve, twelve becomes eight, and every step down15is given away for nothing because there was never a written list of what we wanted in return. The16mirror image is an uplift asserted at T-30 with no value artifact behind it, raised for the first17time inside the customer's own notice window.1819The elite version builds the position before the price, then gates it: every give is attached to a20get in the same row, no rung is offered to someone who cannot sign, no concession is discussed while21the only live thread runs through procurement, and the number is announced rather than requested.22Read `../cs-context/references/evidence-standard.md` first.2324## Before Starting25261. **Read `.agents/cs-context.md`** (fallback `.claude/cs-context.md`). If absent, run `cs-context`27 first — without §2 Commercial Model this skill guesses at the standard uplift, the discounting28 norms and the payment terms it exists to defend. **Never ask what that file answers:** ARR,29 renewal date, notice period, standard uplift, discounting norms, payment terms, segment, owner,30 fiscal year, source inventory.312. **Take the clause stack from the executed contract, not the CRM's opinion of it.** Every field is32 a lever, and a wrong one is a lever pulled the wrong way in front of procurement.3334| Input | Field | If missing |35| --- | --- | --- |36| Renewal date · notice period | `subscription.renewal_date` · `notice_period_days` | `UNKNOWN — requires the executed contract`; every timing rule below becomes a ceiling, not a plan |37| Uplift clause, cap, index · auto-renew and last change | `subscription.uplift_pct` + clause text · `auto_renew` · `auto_renew_changed_at` | No clause means no contractual rung, so justify on value or hold; an auto-renew change is a decision, not a posture (`R2`) — route to `save-play` before pricing anything |38| Current discount and expiry · ATR | `discount_pct` · `discount_expires` · `subscription.arr` co-termed today | An expiring discount **is** a price increase — find it before they do; and never use the original order-form ARR, since mid-term adds have moved it |39| Seats purchased · provisioned · active | `seats_purchased` · `seats_provisioned` · `usage_daily.active_users` | The downgrade test in Step 9 cannot run |40| **Who signs** | `signs` from `stakeholder-map` — the person on the order form, not the person on the call | **Gate A fails and the ladder is blocked** (Step 4). The recommended action becomes the authority test |41| **Last non-commercial conversation** · health band | Latest `interaction.occurred_at` whose subject was outcomes, adoption, roadmap, support or the success plan, with a counterpart outside procurement and legal · `churn-risk` output | Treated as **older than 21 days** and Gate B fails closed (Step 4); run `churn-risk` first, because an uplift on an At Risk account is a different decision |42433. **Compute two fields before anything else, and print both** — `scripts/concession_math.py gates`,44 which also returns Gate A and Gate B. Neither is left to the writer to remember.4546| Field | Formula | Why it is computed, not recalled |47| --- | --- | --- |48| **`D`** — the governing date | `renewal_date − notice_period_days` (`R1`) | Where today sits against `D` changes the recommended position, not merely the schedule |49| **`days_since_business_thread`** | `today − last non-commercial conversation`; `UNKNOWN` counts as over the limit | Past 21 days with procurement active, the ladder is blocked and re-opening a business thread becomes a required action (Step 4, `C10`) |50514. **Ask these four once, tappably, in one batch, then run unattended.** `AskUserQuestion`, 2–452 mutually exclusive options each, recommended first and labelled, one line under each saying what53 it changes. Drop any question `cs-context` or the request already answers.5455| Header | Question | Options — recommended first, each with what it changes |56| --- | --- | --- |57| `Stage` | Where is this negotiation? | **Setting the position (Recommended)** — full pack: uplift decision, gates, costed ladder, timing, comms · **A demand has landed** — straight to the gates, the counter and the reply draft · **Structuring a multi-year or restructure** — structures with the arithmetic and the terms to hold · **Signed or lost, debrief it** — what was conceded, what it cost, what to change |58| `Position` | What are we going for on price? *(skip if `cs-context` §2 sets a standing uplift and this account is standard)* | **Standard uplift with the value artifact (Recommended)** — clause plus delivered outcomes · **Hold flat, traded** — the hold buys term, reference rights or prepayment · **Above-standard or restructure** — needs the usage delta or an evidenced below-market entry price · **Not decided** — schedules the T-120 decision instead of assuming one |59| `Ceiling` | What can you approve without going anywhere? *(skip if `cs-context` §2 names the bands)* | **Up to 5% (Recommended)** — ladder stops at rung 4 · **Up to 15%** — manager level, rungs 5–6 open · **Above 15%** — VP or deal desk, rungs 7–8 with written justification · **Unknown** — every rung past 4 prints `approval required — name the approver` |60| `Output` | Who reads this? | **Me, before the call (Recommended)** — internal pack plus the talk track · **Pack plus the customer drafts** — adds the send-ready blocks in §15 · **My manager or deal desk** — adds the approval ask, the walk-away rationale and the precedent note |61625. **Never block, never guess.** Every missing input resolves one of three ways — **read it** (showing63 the derivation), **ask it** (only where two answers produce a materially different position), or64 **mark it** `UNKNOWN — requires <source>` with a confidence cap. A plausible substituted value65 becomes a fabricated one the moment it is quoted to procurement. If nothing comes back, run on the66 recommended defaults, state them above the Bottom Line, and give each an **Assumptions** row67 (`../cs-context/references/clarification-protocol.md`).686. **Take whatever data arrives** — CSV, XLSX, JSON, warehouse results, a pasted contract clause, a69 procurement thread, a transcript, or no file at all. **Run `../cs-context/scripts/ingest.py` first70 on every supplied file.** Confirm every mapping below 0.80, and `arr`, `discount_pct`,71 `uplift_pct`, `notice_period_days` at any confidence — a column mapped one field left produces a72 walk-away that is confidently wrong. **Degrade, never refuse:** under 40% coverage, name the gap73 instead of setting a price, and never assume an export is current.7475## How This Skill Works7677| Mode | When | Produces |78| --- | --- | --- |79| **Position** (default) | T-120 → T-90, before any number is exchanged | Value, price position, their alternatives, switching cost, walk-away, ask stack beyond price, the two gates, costed ladder, timing, opening comms |80| **Live demand** | A discount, downgrade, best-and-final or procurement email has landed | The gates, what they actually asked, what it costs, the counter with its required get, the approval needed, the reply draft |81| **Structure** · **Debrief** | Multi-year, ramp, co-term or restructure on the table; or signed or lost, within 5 working days | Structures compared with the arithmetic, uplift schedule, terms to hold, forecasting consequence; or the gate log, what was conceded, what each cost, what it bought, the precedent set, the one change before next cycle |82| **Brief** (default output) | Always, unless depth is asked for | ≤20 lines: position, number, the one reason it holds, gate status, the next move with owner and date, confidence in three words, the falsifier, then *Full position pack, ladder and workings on request.* Brief drops the display of the reasoning, never the reasoning |83| **Full** (output on request) | Going to a deal desk or an approver, or someone will challenge the number | The complete Output Template |8485### The ethics line — not decoration8687A renewal won by trickery is next year's churn — and the churn nobody will tell you the real reason for.8889| Never | Instead |90| --- | --- |91| Rely on the customer missing their notice window | Send the courtesy notice-deadline letter at T-120, repeat at T-90, naming the date in writing |92| Manufacture a deadline — "this pricing expires Friday" when it does not | Anchor on `D`, a real date they can verify in their own contract |93| Withhold data as leverage, or bury an increase in an amendment with no covering note | Make leaving mechanically easy — the only thing that makes a win-back possible; and name the increase, the amount, the effective date and the reason in the email, before the paper |9495Auto-renewal mechanics are a legal question, not a negotiating one — New York GOL §5-903, and US96federal negative-option rulemaking that is currently unsettled. **Route it to counsel**, and read97the statutory position in `references/procurement-tactics.md` before answering a notice question.9899### The rules this skill enforces100101From `../cs-context/references/operating-rules.md`, enforced in the output. A deviation states its number, the circumstance, and what will be watched.102103| Rule | Enforced how |104| --- | --- |105| **R1 · The Opt-Out Calendar** | Every timing decision is an offset from `D`. Inside `D` the recommended uplift default changes (Step 3) |106| **R2 · Decisions Beat Indicators** | Auto-renew off or notice served stops the pricing conversation and routes to `save-play` |107| **R11 · Value First** · **R12 · Co-Term** | Value at T-150/T-120, price at T-90, and never an ask in a meeting carrying a miss or an apology; expansion inside 90 days of `D` co-terms, outside it runs separately and never as ladder currency |108| **R18 · The Firewall** | Walk-away, rung, gate status, approval band and any assessment of a named person never reach customer text |109| **R21 · The Stop-Loss** | Every position carries a walk-away and an exit date, set before the first call |110| **R22 · Ordering** · **R23 · Coverage Cap** | Break-even and cost figures are arithmetic, never a churn probability without a cited backtest; and confidence never exceeds coverage across the seven families |111112### Business model, and what the seven families decide113114Resolve the profile from `../cs-context/references/business-model-profiles.md` before Step 1.115**Consumption:** no seat price to discount — the negotiation is the commitment tier and the overage rate.116**Product-led:** below the enterprise threshold there is no negotiation, and inventing one creates a117market for complaining. **Monthly evergreen:** no notice window, so no `D`. **Regulated/public sector:**118price may be framework-fixed; the negotiable surface is scope.119120All seven families are checked every time, including the clean ones. Product usage & adoption decides whether121the value evidence is real and whether a downgrade ask is genuine · commercial & contract holds the clause122stack · relationship & engagement supplies `signs` and `days_since_business_thread`, which together decide123whether the ladder opens at all · support & reliability decides whether an increase is defensible this cycle124· sentiment & VoC prices reference rights · billing & payment prices the payment-terms rungs · firmographic125& external gives their budget cycle.126127Run sequence: **position → price → time the ask → gate the ladder → cost the ladder → discount discipline → structure → procurement → downgrade and the competitive line → comms, then debrief.**128129---130131## Step 1 — Build the position before you name a number132133Six elements. A position missing one of them is a price with a story attached.134135| # | Element | What it must contain | If missing |136| --- | --- | --- | --- |137| 1 | **Value delivered** | Three outcomes in their own metrics: baseline · current · delta · a customer-side validator, provenance-tagged | The only rung left is the contract clause, the weakest one. Build it via `value-case` |138| 2–3 | **Price position and their alternatives** | What this account pays per unit against our own realised book; and the named alternative, its status, what it replaces and what it does not | `UNKNOWN — requires realised ACV per unit by segment`; never quote a public benchmark as our price position, and if nobody has asked about the alternative, ask — an unnamed "the market" is not one |139| 4 | **Switching cost** | Computed from *their* numbers (below) | `UNKNOWN — requires integration inventory and trained-user count`. Never a market average |140| 5 | **Walk-away** | The number **and** the terms below which we do not sign, approved before the first conversation (`R21`) | A walk-away invented mid-call is not a walk-away, it is a feeling |141| 6 | **Ask stack beyond price** | Term, multi-year, reference rights, case study, advisory seat, expansion path, payment timing, exec sponsor access, logo rights | Every concession becomes a pure loss, because there is nothing to trade it for |142143**Switching cost, from their own numbers** — the strongest figure in the room, and almost nobody brings144it: `integration rebuild + data migration + retraining + parallel run + process rewrite`, each priced145from their integration inventory, record counts, trained-user count and loaded rates (arithmetic in146`references/deal-structures.md`). Every input comes from their data or is marked `UNKNOWN — requires147X`; never quote a published switching-cost average, which procurement dismantles in one question.148**Exit criteria:** six elements populated or marked, walk-away signed off by a named approver on a date.149150## Step 2 — Choose the uplift and pick the rung that justifies it151152Decide internally at T-120. **Pick exactly one rung** — 0 hold flat (traded, never free) · 1153contractual index · 2 standard uplift · 3 above-standard · 4 restructure — and be able to produce its154evidence in the meeting, not after it. Rung 2 needs the clause **and** a documented value-realisation155summary; an uplift asserted without the value artifact is a price increase in search of a reason.156Rungs 3 and 4 need deal-desk approval. Full rung table: `references/uplift-justification.md`.157158**The justification stack, in this order — the sequence is the argument.** (1) Outcomes in their159metrics. (2) Product shipped since the last renewal and actually adopted. (3) Usage growth and160entitlement true-up. (4) The contractual clause. (5) List-price movement. Leading with 4 or 5 tells161procurement this is a commercial fight, and they are better at those than you are.162163Standard negotiated annual uplifts of **3–5%**, often pegged to CPI or a fixed percentage "whichever is164higher", are reported by SaaS contracting and procurement guides (2026) — `[V]` practitioner guidance,165not a measured benchmark. Vendor price indices reporting double-digit averages run on unaudited166methodology `[V]`; **never quote one to a customer.** An uplift of `u` is value-neutral at `u / (1 + u)`167of added churn probability — `scripts/concession_math.py uplift`, arithmetic for ordering and **not** a168forecast (`R22`).169170## Step 3 — Time the ask171172| Situation | The rule | If you are already past it |173| --- | --- | --- |174| Standard renewal | Decide by T-120, deliver with the proposal at T-90, negotiate T-60 → T-45 | Compress, do not skip — the value artifact still lands first |175| Value and price · after a miss or apology | Value at T-150/T-120, price at T-90, never in one conversation, and never an ask in a meeting carrying a miss (`R11`) | Split across two days and say aloud which conversation this is |176| **Inside the opt-out window** | **Never raise an increase for the first time inside `D`.** | Hold flat this cycle and put the increase in writing for the next one with its effective date named. State the deviation under `R1` |177| Expansion alongside the renewal | Inside 90 days of `D`, co-term it (`R12`) | Outside 90 days run it separately — and never use an expansion as ladder currency |178179## Step 4 — Gate the ladder before you cost it180181**Two refusal conditions, run before any rung is priced.** Both fail closed, and both print in the pack182with their evidence whether they pass or fail.183184```185python3 scripts/concession_math.py gates --signs "Dana Osei" --signer-present yes --last-business-conversation 2026-07-02 --procurement-active yes186```187188| Gate | Passes when | When it fails |189| --- | --- | --- |190| **A · Authority** (`C14`) | `signs` names a person **and** that person is in the conversation, or has confirmed in writing the terms they will sign | Every `Offer?` cell in §3 reads `BLOCKED — Gate A`. **The recommended action is the authority test, not the concession** |191| **B · Business thread** (`C10`) | `days_since_business_thread ≤ 21`, or procurement is not active | Every `Offer?` cell reads `BLOCKED — Gate B`. **The recommended action is re-opening a business thread, with an owner and a date, before any concession is discussed** |192193**A pack that offers a rung under a failed gate is invalid output.** The costed ladder is still194built — an approver needs it — but no rung is offered, and §4's counter carries the gate's action.195196**The authority test, said before any rung is priced.** A concession given to someone without197signing authority is spent twice: once with the person who cannot sign, and again with the person198who can, who now opens from your reduced number.199200> "If I could get that approved, is this something you could sign this quarter?"201202Where `signs` is `UNKNOWN` the question is who, not whether: *"Before I take a number to our deal203desk — who signs the order form on your side, and are they behind the figure you have asked me204for?"* An answer of "I would still need to take it to Dana" is a blocked gate, not a soft yes.205206**Re-opening the business thread is a named meeting, not a check-in.** A named person, a subject207drawn from their own data, a date, and no price in it. The commercial reply goes out once that208meeting is **booked**, not once it has happened, and procurement is copied — a second thread, not a209secret one. Both gates in full with the three plays: `references/concession-ladder.md`.210211## Step 5 — Cost the ladder, and never climb a rung without its get212213Cheapest to most expensive, costed in dollars for *this* account before the call —214`scripts/concession_math.py ladder`.215216| # | The give | **What we get** | Approver |217| --- | --- | --- | --- |218| 1 | Payment terms — Net 30 → Net 45/60 | Signature by a named date | CSM / RM |219| 2 | Waive or reduce the uplift | Multi-year term, or a two-year price lock in our favour | RM |220| 3 | Enablement, training credits, premium-support trial, advisory seat | Exec sponsor meeting plus a written success plan | Manager |221| 4 | 0–5% discount | Named signature date **plus** reference rights (two calls a quarter, one case study a year) | Rep / RM |222| 5 | 5–10% discount | Multi-year commitment, or increased scope or seats at list | Manager |223| 6 | 10–15% discount | Multi-year **plus** prepayment, annual or full-term upfront | Manager / Director |224| 7 | 15–25% discount | Multi-year plus prepay plus expanded scope plus logo and case-study rights | VP |225| 8 | >25%, or any unprecedented term | Strategic justification, a board-visible logo, or a documented displacement threat | C-level / deal desk exception |226227**A row with an empty What-we-get cell is invalid output** (`C12`). Not incomplete — invalid: delete228the row or name the get. There is no `—` and no `n/a`, and four things read as gets without being229one — **goodwill or the relationship** (unpaperable) · **"they renew"** (the renewal cannot be the230payment for the renewal) · **"they stop asking"** (you bought quiet and set next year's floor) · **an231expansion inside 90 days of `D`** (it co-terms by `R12`).232233Five rules make the ladder work: **one rung at a time, never two in a meeting** · **never open at the234walk-away** · **ask what the constraint is before pricing it**, since it is often timing, budget235period or scope, none of which cost margin · **non-price levers first** · **log every rung offered**,236since an offered rung is next year's floor whether or not it was taken.237238**Never without VP+ approval** — these four set *next year's* leverage: the **notice window**, the239**auto-renew clause**, the **uplift clause**, **audit and true-up rights**. **Never at any level:** MFN240pricing · unlimited liability · TfC with no fee or notice · source-code escrow without a deal-desk241exception · unmetered usage · a permanent base reprice disguised as a one-time credit. Per-rung242scripts and the What-we-get catalogue: `references/concession-ladder.md`.243244## Step 6 — Discount discipline245246**Compute the lifetime cost before offering a point** — `scripts/concession_math.py discount`. A discount247decides every year they stay, plus the escalator you no longer apply to the reduced base; round to two248significant figures (`§4F`). **The precedent is the real price:** it is the floor they negotiate from next249cycle, and it travels through procurement networks and benchmarking services that exist to collect it `[P]`.250251**A one-time concession that does not reprice the relationship needs all six:** list price stated and252unchanged on the order form · a named credit line with an end date · snap-back language in the same253document · **a get named in the same paper** (`C12`) · the words "one-time" in the covering email as well254as the contract · logged as `value_delta_reason = discount_concession`. Fewer than six and it is a new255price with an optimistic note attached. **When something must be given, in order of preference:** payment256timing → scope or entitlement → a term-limited credit → a percentage off the base; only the last is257permanent by default. Detail: `references/concession-ladder.md`.258259## Step 7 — Structure the deal260261Six structures, their arithmetic and their traps: `references/deal-structures.md`. Four rules hold262whichever you pick. **Write each year's fee in dollars on the order form, not as a formula.** **A263multi-year with an annual opt-out is not a multi-year**: every anniversary is its own renewal event with264its own `D` (`R1`), forecast and staffed as annual. **A ramp step is tied to a milestone you can265observe**, or it is a discount with a delay. **Payment terms are money:** 30 days of extension costs266about `ARR × 30/365 × cost of capital`, and prepayment is the same in reverse.267268## Step 8 — Handle procurement269270Procurement is not your enemy and not the decision-maker. They are measured on savings against a271baseline, cycle time and risk terms accepted — so **give them something to save against, early, and272never let list price be the only anchor in the room.** Never let their thread be the only one open273(`C10`, Gate B).274275| Their move | The counter |276| --- | --- |277| Entering at T-90 or earlier (normal and healthy) | Give everything at once: pricing, entitlement report, security package, insurance certificates, entity and tax data, redlines accepted elsewhere `[V]` |278| Entering inside T-30 | Name the paper path and its measured lead times, and propose a 30–60 day extension at current terms. That removes the artificial deadline and is a real option, not a bluff |279| "We have quotes 30% lower" | Ask for a scope-normalised comparison, then price the switching cost from their numbers. **Never match a number whose scope you have not seen** |280| "Send your best and final" at T-14, or a deadline in no contract | A best-and-final is conditional on a named signature date and a named get, or it is simply a lower price; anchor timing on `D`, which they can verify in their own paper |281| Splitting the CSM from the AE and price-testing each | The RM owns price, the CSM owns value, and neither prices anything in a hallway. Say it plainly |282283Ten tactics with their tells, the re-opening plays and the paper lead times:284`references/procurement-tactics.md`. **The renewal manager owns procurement, not the CSM** `[P]` —285where one person holds both, run the two conversations on different days.286287## Step 9 — Test a downgrade, and hold the competitive line288289| Test | Genuine right-sizing | Negotiation move |290| --- | --- | --- |291| **Utilisation** | Provisioned seats far above active users, unused for two quarters | Utilisation healthy; the ask appeared with the renewal |292| **Who is asking** | The admin or budget owner, with a headcount number | Procurement, with a round percentage |293| **When raised** | Before the renewal conversation, or at a QBR | First raised at T-30, attached to a deadline |294| **How specific** | Named teams, named seats, a date | "About 20%" |295| **If refused** | They renew anyway, smaller | Explicit or implied non-renewal |296297**Three or more on the left: take the right-size, and take it well.** Median B2B SaaS gross revenue retention298was **84%** in the 2026 Benchmarkit B2B SaaS & AI-native metrics report (FY2025 data), down from 88% `[V]` —299most of that erosion is contraction, not logo loss. Right-size to **real usage plus a named growth buffer** ·300**trade the reduction** (`C12`), since a smaller contract at list is frequently worth more than a larger one301at 15% off · **keep entitlement above the point where the use case breaks** · **put the recovery path in the302paper** with a pre-agreed unit price for adds during the term · **book it as contraction** with303`value_delta_reason = seat_reduction`, never as a flat renewal, which is how GRR lies to a board. **Three or304more on the right:** a price negotiation wearing a downgrade's clothes — answer it with Step 5's ladder.305306**Under competitive threat, do not compete on price.** Compete on switching cost, integration depth,307data gravity, and what a migration costs them in the quarter they would have to run it. Ask directly308to be in the evaluation; never disparage, never promise a roadmap date to win a bake-off (`R19`).309**Price-match rule:** only where you have seen the scope, and only as a term-limited credit with310written snap-back. If they are genuinely leaving, stop spending (`R21`) and hand to `save-play`.311312## Step 10 — Write the comms, then debrief313314Five drafts leave the building, written out in `assets/negotiation-comms.md` under315`../cs-context/references/customer-voice.md`: the **business-thread re-opener** (first, whenever316Gate B failed), the **uplift notification** (T-90), the **discount response**, the **downgrade317response**, and the **best-and-final / deadline response**.318319**The price paragraph is constructed, not written.** Four slots, and the number is slot 3:320321| Slot | Contains | Rule |322| --- | --- | --- |323| 1 | What changed, in their numbers, with dates | Two facts, each verifiable in their own systems |324| 2 | What we shipped that they adopted | Named feature, named team, when it went live |325| 3 | **The number** | Declarative present tense. Last sentence of the paragraph. Full stop (`C3`) |326| 4 | Effective date, notice date, and the meeting offer | New paragraph. Does not restate the number |327328**`C13` · Announce, do not ask.** The price paragraph carries **no question mark** and no request329construction — `would you be open to`, `is there room`, `we were hoping`, `would you consider` and330their relatives. A price framed as a request invites a counter before the conversation starts.331Correct: *"The annual fee for the 2027 term is $514,000."*332333**`C3` · Say the number, then stop.** Justification precedes the number; nothing follows it in the334same paragraph — no `of course`, `that said`, `however`, `we're flexible`, `happy to discuss`,335`nothing is set in stone`. Softening invites negotiation against yourself; the silence does the336work. Questions are legitimate — the authority test is one — and live in their own paragraph, away337from the number. Both banned lists in full, with worked rewrites:338`references/uplift-justification.md` §6b.339340**Every draft is scanned before it is fenced.** `python3 scripts/pre_send_scan.py draft.txt` must exit3410 — it checks `C13`, `C3`, the `R18` firewall vocabulary and unfilled placeholders — then the342eight-step leak scan in `customer-voice.md`. A hit is **rewritten, not softened**. **The firewall343(`R18`)** keeps all of this out of customer text in any wording: the walk-away · the rung or the word344"concession" · the gate status · the approval band · health or risk band · revenue at risk · forecast345category · save play · the precedent note · any assessment of a named person on their side.346347**Debrief within five working days, signed or lost** (`assets/debrief-template.md`): the gate log ·348what was conceded · what each cost · **what each bought** · the precedent set · what we refused and what349happened · which lead times we mis-estimated · the one process change before next cycle.350351---352353## Output Template354355### Brief — the default356357```markdown358**<Account> · ATR $<X> · <position> · decide by <D> (<N> days)**359360<Two sentences: the position and the one reason it holds, with provenance on the numbers.>361362**Gates:** A <PASS/FAIL — signer, presence> · B <PASS/FAIL — <n> days since business thread>.363**Ask:** <uplift or hold, and the rung.> **Trade:** <first rung → what we get.> **Walk-away:**364<number and terms — internal only.> **Do:** <Owner> <action> by <date>.365Confidence: <level> (<n>/7 families). **What would change this:** <2–3 observable events.>366367*Full position pack, ladder and workings on request.*368```369370Where either gate fails, the **Do** line is that gate's action — the authority test or the371business-thread meeting — and no rung appears in the **Trade** slot.372373### Full — on request374375The complete pack is `assets/position-pack.md`, emitted verbatim. Sections in this order: **Bottom Line**376(carrying `signs · decides · influences` and `days_since_business_thread` as header rows) → **1. The377Position** → **2. Uplift and Justification** → **3. The Gates, then the Ladder** → **4. Live Demand →378Counter, or Structures Compared** → **5. Timing and Actions** → **What would change this position** →379**Assumptions** → **Coverage Ledger**. Two orderings in it are structural: **§3 generates the Gates380table before the ladder table** and prints `Ladder status: OPEN / BLOCKED` — under a failed gate every381`Offer?` cell reads `BLOCKED`, §4's counter is the gate's action, and §5's first row is the authority382test or the business-thread meeting; and **every ladder and counter row carries a non-empty383What-we-get cell** (`C12`), a row without one being deleted rather than shipped.384385**§15 closes the pack when a draft is due — the only part that leaves the building:**386387````markdown388## 15. Customer-Facing Draft389<Only the draft due now. All five written out in `assets/negotiation-comms.md`.>390391════════════════════════════════════════════════════════════392CUSTOMER-FACING — copy the block below and send as written.393Everything above this line is internal. Do not forward it.394════════════════════════════════════════════════════════════395396**<Draft name> — due <date>, to <recipient>** · `pre_send_scan.py`: PASS397398```text399<Send-ready text. Plain, blank line between paragraphs, • bullets, no markdown headings or pipe400tables. Opens on something only this account's data could produce. The price paragraph runs401justification → number, the number is its last sentence, and it carries no question mark. Names the402opt-out date in plain words wherever the draft concerns price. A fence containing [Name] is not send-ready.>403```404````405406## Quality Bar407408- [ ] `D = renewal_date − notice_period_days` computed from the executed contract and printed with days remaining, with no timing decision anchored on the renewal date (`R1`); all six position elements populated or marked `UNKNOWN — requires X`; the walk-away carrying a named approver and a date (`R21`); switching cost computed from the customer's own numbers, never a market average409- [ ] Exactly one uplift rung chosen, its evidence named, the justification stack ordered value-first (`R11`), and no increase raised for the first time inside `D` unless the deviation states `R1`, the circumstance and what will be watched410- [ ] **`C14`** — Gate A printed with its evidence. Where `signs` is UNKNOWN or the signer is not in the conversation, no rung is offered and the recommended action is the authority test in its exact words411- [ ] **`C10`** — `days_since_business_thread` computed and printed. Over 21 with procurement active, Gate B is FAIL, the ladder reads BLOCKED, and a dated business-thread meeting is the first action412- [ ] **`C12`** — every ladder and counter row carries a non-empty **What we get** cell. A row without one is deleted, not shipped. "Goodwill", "they renew" and "they stop asking" are not gets413- [ ] **`C13`** and **`C3`** — the price paragraph of every draft is declarative with zero question marks and zero request constructions, justification precedes the number, the number is the last sentence of its paragraph, no softener follows it, and `pre_send_scan.py` exits 0 on every draft414- [ ] Every rung costed in dollars for this account, both never-concede lists printed even when nothing was requested, any discount modelled over its lifetime and rounded to two significant figures (`§4F`), and any one-time concession carrying all six paper requirements including written snap-back415- [ ] Downgrade requests scored on all five tests before a response is drafted, and booked with the correct `value_delta_reason`416- [ ] Every number carries a provenance tag with a date or window, every inference states its rule, every action has action · owner · date · expected effect · success measure, and confidence is stated at or below the Coverage Ledger cap across all seven families (`R23`) with no churn probability and no certainty language — arithmetic and bands only (`R22`)417- [ ] Marked internal (`R18`); customer drafts sit inside a ```text fence below the divider, email-formatted, zero unfilled placeholders, leak scan run on each, and no walk-away, rung, gate status, approval band or precedent note anywhere in them418- [ ] Every missing input resolved read it / ask it / mark it; four questions asked once, tappably, batched, with nothing asked that `cs-context` §2 answers; every default stated above the Bottom Line with an Assumptions row naming a concrete consequence419- [ ] Any supplied file went through `ingest.py`; mappings below 0.80 confirmed plus `arr`, `discount_pct`, `uplift_pct`, `notice_period_days` at any confidence; as-of date printed; business-model profile resolved first, so no seat-price negotiation is proposed on a consumption contract and no negotiation is invented for a self-serve plan420421## Anti-Patterns422423| Anti-pattern | Correction |424| --- | --- |425| A concession row with an empty get, or a get of "goodwill" | Every give is attached to a get in the same row, or the row is deleted (`C12`). The renewal cannot be the payment for the renewal |426| Offering a rung to someone who cannot sign | Run Gate A. Where `signs` is UNKNOWN or absent, the move is the authority test, not the concession — a discount given to a messenger is spent twice (`C14`) |427| Letting procurement be the only live thread | Compute `days_since_business_thread`. Past 21 days, book a non-commercial meeting with the economic buyer before any concession is discussed (`C10`) |428| Asking for the uplift, or softening straight after the price — "would you be open to a small increase?", "…but of course there's flexibility" | Announce it: rationale, then the number, declarative, no question mark in the price paragraph (`C13`); justification precedes the number, the number ends the paragraph, and nothing follows it (`C3`) — the silence does the work |429| A customer draft carrying a rung, a band, a walk-away or `[Name]` | Fill every slot or drop the sentence; `pre_send_scan.py` exits 0 before the fence is emitted |430| Negotiating the percentage they opened with | Negotiate the structure. Ask what the constraint is, then price *that* |431| Raising the price for the first time inside the notice window | Decide at T-120, deliver at T-90. Inside `D`, hold flat and put next cycle's increase in writing with its effective date (`R1`) |432| Asserting an uplift with no value artifact, or leading the justification with the clause | Rung 2 needs the clause *and* the outcomes; value first, clause fourth — leading with the clause starts a commercial fight you will lose |433| A "one-time" discount with no snap-back, or a multi-year with annual opt-outs treated as a multi-year | Six paper requirements including written snap-back, and a lifetime cost that includes the escalator forgone on the reduced base; an opt-out multi-year is three renewal events with three opt-out deadlines, priced, forecast and staffed as annual (`R1`) |434| Fighting every downgrade, or letting their deadline set the calendar | Score the five tests — genuine right-sizing is taken, traded and papered with a recovery path. Anchor timing on `D`, and where their deadline is artificial propose a 30–60 day extension at current terms |435| The CSM negotiating price directly, or matching a competitor quote whose scope you have not seen | RM owns price and CSM owns value, run on different days where one person holds both; scope-normalise a rival quote first, then price the switching cost from their numbers |436437## Related Skills438439| Skill | Relationship |440| --- | --- |441| `cs-context` · `renewal-prep` | `cs-context` **runs first**, supplying §2 Commercial Model — standard uplift, discounting norms, payment terms, notice period; `renewal-prep` **runs before and around this**, owning the T-180→T-0 ladder, gate audit, MEDDPICC-R, paper critical path and the notice-deadline letter, while this skill owns only price and terms |442| `stakeholder-map` · `churn-risk` | **Run before.** `stakeholder-map` supplies `signs`, without which Gate A cannot resolve; the health band decides whether an uplift is on the table at all |443| `save-play` | **Runs instead** once notice is served, auto-renew flips or the account is Critical. Its ladder is a rescue instrument; this one is commercial |444| `value-case` · `qbr-builder` · `expansion-finder` | The first two supply position element 1 and the business thread Gate B requires (`R11`); `expansion-finder` sizes the expansion, and this skill decides whether it co-terms (`R12`) and forbids it as ladder currency |445| `renewal-forecast` · `churn-postmortem` | `renewal-forecast` **consumes this** — the position sets the delta reason codes and the category, and a downgrade booked as a flat renewal breaks GRR; `churn-postmortem` **runs after a loss**, taking the gate log and concession log |446447## Going Deeper448449| Read | When |450| --- | --- |451| `references/concession-ladder.md` | Steps 4–6 — both gates in full, the What-we-get catalogue, every rung with its cost driver and script, discount discipline, the never-concede register |452| `references/uplift-justification.md` | Steps 2 and 10 — rungs and their evidence, cap-and-collar language, and the constructed announcement paragraph with worked rewrites |453| `references/procurement-tactics.md` · `references/deal-structures.md` | Step 8 and the moment a procurement address appears — the tactics with their tells, the business-thread re-opening plays, the statutory notice position and the paper lead times; and Step 7 — multi-year, ramp, opt-out, price protection, co-term and TfC, with arithmetic and forecasting consequence |454| `assets/negotiation-comms.md` · `assets/debrief-template.md` · `assets/position-pack.md` | Step 10 — the five scanned send-ready drafts; the debrief within five working days of signature or loss; and the Full output template, emitted verbatim |455| `scripts/concession_math.py` · `scripts/pre_send_scan.py` | **Before any rung is priced** — `gates` first, then ladder costs, discount lifetime cost, uplift break-ev456457…(truncated)
Run npx skillmds@latest add gaintrace/renewal-negotiation in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.
When money, terms or leverage are on the table at a renewal and the user has to decide what to ask for, what to trade, and what to refuse — uplift, discounts, term, payment terms, downgrades and procurement. Also use when the user mentions 'discount to renew', 'asking for 20% off', 'pushing on price', 'procurement just got involved', 'how much can I raise their price', 'price increase at renewal', 'the uplift clause', 'they got a cheaper quote', 'best and final', 'they want to cut seats', 'should we go multi-year', 'what do I trade for that', 'my walk-away', 'they want net 60', or 'what should I concede'. Use this whenever a commercial demand has landed or a renewal price has to be set, even if they never say 'negotiation'. For the T-180 runbook and paper path, see renewal-prep. For an account already in rescue, see save-play. For the risk read, see churn-risk. For sizing an upsell, see expansion-finder. It is listed under Coding & Dev Tools on SkillMD.
This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.
This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.
Yes. Installing skills from SkillMD is free. This skill is licensed under MIT.
gaintrace (@gaintrace) published this skill. Their other Agent Skills are listed on their SkillMD profile.
gaintrace