Withdrawal only works when the vendor has more to lose from your absence than you do from the delay. This page sets out when to stop responding, how long to hold, what the vendor does on days 3, 10 and 21, and the numbers a well timed silence should deliver.
Withdrawal only works when the vendor has more to lose from your absence than you do from the delay. This page sets out when to stop responding, how long to hold, what the vendor does on days 3, 10 and 21, and the numbers a well timed silence should deliver.
Going quiet is not sulking and it is not a bluff. It is a measurement. When you stop responding, one of two things happens: the rep's forecast line goes red and they escalate internally to save it, or nothing happens at all and you learn, cheaply, that you were never load-bearing in their quarter. That answer is worth having before you concede anything. Right now the question silence is answering is specific: vendors are asking for AI-driven renewal increases of 20 to 37 percent against a historical annual uplift band of 3 to 9 percent. That gap is not cost recovery, it is a test of whether you will pay for the word "AI." Tropic's renewal data shows negotiation strips roughly 55 percent off the initial ask in relative terms, and the vendors demanding the most concede the most. Silence is the cheapest way to find out which half of that spread you are in.
Withdrawal only earns its keep when three preconditions hold at once. First, no contractual deadline lands inside 60 days, because a hold clock that runs past a hard date converts leverage into a fire drill. Second, you have a live alternative under evaluation or a credible in-house path, and the rep knows it exists. Third, nothing in production breaks while you are dark. Miss any one and you are not testing the vendor, you are testing yourself.
If nothing happens while you are dark, you have learned something worth more than the discount you were chasing.
Three states make silence outright malpractice. An expiring pilot converting to production is the worst of them: the free beta was engineered to build operational dependency precisely so that the uplift at conversion is unrefusable, and every day of silence deepens the user base that will fight you. A SKU retirement notice already served is a countdown, not a conversation, and the correct response is an immediate market evaluation, not absence. And any autorenew clause with a notice window under 90 days turns a 45-day hold into an automatic renewal at the vendor's number. Check the notice mechanics before you draft the withdrawal note, not after. Our broader view on sequencing sits in the GenAI negotiation timing and leverage playbook.
Exposure is vendor-specific, and the same 21 days of silence produces wildly different results depending on who is on the other side. OpenAI and Anthropic sell direct with no published list price for enterprise seats. The ChatGPT Enterprise band lands roughly $45 to $75 per user per month, commonly cited near $60, which means the number on your quote is entirely the rep's discretion. There is no price list to hide behind and no channel partner to blame. When you go dark, the rep owns the forecast miss personally. Claude Enterprise behaves similarly: across 47 tracked negotiations the average discount was 28 percent with a 15 to 45 percent range, and 40 percent-plus went to buyers committing $1M or more. Discretion that wide is discretion you can move, and we set out the seat-versus-token structure in the Claude Enterprise pricing analysis.
Microsoft is the inverse. The Copilot add-on is anchored at $30, held there even as the underlying E3 and E5 licences rose on July 1, 2026, and the discount ladder is published: roughly 15, 20, 30 and 40 percent by seat band, expiring June 30, 2026. Copilot passed 20 million paid seats against a base of 450 million plus commercial seats. Your 2,000 seats are a rounding error in that forecast. Silence moves Microsoft only when it is attached to a named EA anniversary and a broader renewal envelope, which is the argument in our Microsoft EA negotiation tactics for 2026. Salesforce sits between the two: Agentforce PreCommit gives the rep a contracted revenue floor with a true-up on any shortfall and no rollover of unused Flex Credits, so your silence costs them engagement, not revenue.
| Vendor | What your silence actually costs them | Strong outcome after a 21-day hold |
|---|---|---|
| OpenAI (ChatGPT Enterprise) | A discretionary forecast line with no list price defence; rep-level exposure | Land at or below $48 per user per month, seats matched to active users, price held across term |
| Anthropic (Claude Enterprise) | Same direct-sale exposure; 28% average discount is the floor, not the ceiling | 33 to 40% off opening at a $500K plus commit, 3-year term adds 5 to 10% |
| Microsoft (Copilot, Agent 365) | Statistically nil outside a named EA cycle; promo ladder already public | Secure the 40% band at 1,000 plus seats before June 30, 2026, plus credit price protection |
| Salesforce (Agentforce) | Engagement only; PreCommit true-up protects the revenue floor regardless | Kill the shortfall true-up or win credit rollover, then negotiate the $0.10 per action rate |
Do this first: pull your autorenew notice window and your next hard date, then rank your vendors against the exposure column above. Go quiet only where the forecast pain is theirs.
Announce it once, in writing, then say nothing. The message is four lines and contains no counter, no date, and no complaint: the requirement is under budget review, other initiatives have moved ahead of it in the internal sequence, we will come back to you when there is something to discuss, and please hold further outreach until then. That is the whole email. The moment you attach a number, you have made a counteroffer and the vendor will negotiate against it while you think you are silent. The moment you offer a date, you have given the account executive a forecast line to defend and the silence stops costing them anything. Withdrawal works because it removes certainty from someone whose compensation depends on certainty, and the version that keeps a date in it removes nothing.
Then enforce it, which is harder than sending it. No replies to the AE, including one line acknowledgements. No solution engineer calls, no architecture whiteboards, no security questionnaire returns, no procurement portal activity, no data on seat counts, active user rates, or token consumption. Every one of those artifacts feeds the vendor's internal business case and tells the deal desk the opportunity is alive. Send a written instruction to every internal stakeholder who has the vendor's mobile number, naming the engineering leads and the pilot owner specifically, and stating that any vendor contact routes to procurement unanswered. In my experience the leak is almost never procurement. It is a well meaning platform lead who takes a friendly call and mentions the rollout date.
A pilot that keeps burning tokens through your silence is doing the vendor's negotiating for them.
The expensive failure mode is consumption. If a pilot team keeps running through the hold, the vendor watches the meter climb and reads your silence as posturing over a dependency you have already accepted. On the API side, a hold at $2.50 input and $10 output per million tokens on GPT-4o produces a rising usage curve that becomes the vendor's opening slide when you return. Freeze the pilot at a documented ceiling, or at minimum stop reporting on it, and time the whole exercise the way the pre-spike negotiation window describes rather than after adoption has priced you.
The vendor response curve is predictable enough to plan against. Days one to seven produce nothing of value: two polite check-ins, a forwarded case study, maybe a webinar invitation. Anyone who breaks in week one has paid for silence and collected nothing. Days eight to fourteen produce the first structural move, usually escalation to the AE's manager and a revised proposal that is repackaging rather than discount, more seats at the same effective rate, credits bundled in, a term extension dressed as value. Days fifteen to twenty one produce the real number when a quarter close sits inside thirty days, which is where the quarter end discount question becomes the operative variable rather than a talking point. Past forty five days the mechanics turn against you: the opportunity drops out of the commit forecast, the account moves to a lower touch or partner led motion, and the relationship rebuild costs six to eight weeks before you get back to the number you already had.
| Hold duration | What the vendor does | What you should expect on price |
|---|---|---|
| Days 1 to 7 | Check-ins from the AE, no internal escalation | Nothing. Opening ask unchanged |
| Days 8 to 14 | Manager escalation, deal desk review begins | Repackaging, 5 to 10 points of apparent value, little real discount |
| Days 15 to 21 | Deal desk approval sought if quarter close is within 30 days | The genuine move. 15 to 25 points off opening is realistic on seat based GenAI |
| Days 22 to 45 | Executive sponsor outreach, sometimes a bypass to your CIO | Best and final territory, plus term and price hold concessions |
| Beyond 45 days | Forecast removal, account downgraded to low touch | Position resets. Rebuild costs 6 to 8 weeks |
Fourteen to twenty one days is the standard hold. Extend to a quarter boundary only when your renewal date leaves genuine slack, because a silence that runs into a service expiry converts leverage into an emergency purchase at list. Two anchors from the research set the target: negotiation cuts the initial vendor ask by roughly 55% in relative terms against AI uplift demands of 20% to 37%, and enterprise GenAI discount bands run 15% to 45% with 28% typical at commitments of $300,000 to $750,000. A 21 day hold that returns a repackage rather than a rate change tells you the exposure is not where you assumed, and the timing and leverage playbook is where you go to re-read the account before you re-enter. First action: confirm the renewal or expiry date, subtract the hold, and check the answer is still comfortably positive before you send the withdrawal note.
When the vendor breaks a 21 day silence, the first offer is almost always packaging dressed as generosity. Microsoft's E7 Frontier Suite is the clean example: $99 per user per month against $117 for E5, Copilot and Entra Suite bought separately. That is a 15% mechanical saving that exists whether or not you ever left the room, and it costs the seller nothing because it was published pricing before your silence began. Grade it as zero. The same applies to term extensions offered as concessions, credit packs "included" at their standard $500 per 100,000 rate, and promotional ladders (the 15/20/30/40% seat band promos) that any buyer of your size qualifies for by walking through the front door. A concession is only real if it is off the number the vendor named to you, at your volume, and it would not have been available to the buyer next door. Against the market, the benchmarks are specific: Claude enterprise negotiations average 28% off with a 15 to 45% range, and 40%+ requires $1M+ annual commitment, with a three year term adding another 5 to 10 points, which is why the seat versus token split matters when you price Claude Enterprise properly. ChatGPT Enterprise volume discounts land in the 10 to 25% band, with the top of that range attaching to 5,000+ seats on multi-year paper. And on renewal uplifts, where the AI tax ask is running 20 to 37% against a historical 3 to 9%, negotiation cuts the initial ask by roughly 55% in relative terms. Use that as your grading rubric: a vendor that opened at a 30% uplift and comes back at 13 to 14% has responded to pressure. A vendor that comes back at 26% has responded to nothing.
The failure signal is unambiguous. Anything under 10% off the opening number after a full 21 day hold means the account team read your silence as theatre and briefed their manager accordingly. In my experience across the table, that response is not a negotiating position, it is a diagnostic result: they have concluded you have no alternative, no internal sponsor for delay, or a renewal date they can see and you cannot move. Do not counter it. Escalate the silence one level (skip the rep, address the regional director) and attach a dated alternative, because the second break is where the real number lives. Grade every return on four axes before you respond: unit price against the benchmark band, commitment level required to hold it, price protection across the term, and what happens at renewal 24 months out. The 23% below opening outcome (5,000 seats at $60 plus a $2M API commit, restructured to 3,800 active seats and $1.4M of API, landing at $3.9M with a two year price hold) came from attacking all four, not from discount on one.
A repackaged bundle is not a concession; it is the price list with a ribbon on it.
Silence is free before you sign and expensive after. The meters running underneath a live agreement do not pause because you stopped answering email, and several are engineered so that inactivity is billable. Agentforce PreCommit is the sharpest: it is a volume commit with negotiated rates, a true-up charge if actual consumption lands below the committed amount at term end, and no rollover of unused Flex Credits into the next term. Go quiet for six weeks during a live PreCommit and you are not applying pressure, you are funding a shortfall invoice on credits sold at $500 per 100,000 (20 credits per action, so roughly $0.10 per action). Microsoft's consumption stack punishes drift in both directions: Copilot Credits at $0.01 pay-as-you-go versus $200 per 25,000 credit prepaid pack means underestimating costs you nothing structurally but overcommitting strands cash, and Windows 365 for Agents bills $0.40 per hour rounded up to the next full hour plus $5 per month for every always-available Cloud PC. Idle agents on always-available instances accrue while you are demonstrating resolve.
The other silence that costs money is the one you cannot credibly perform. The free pilot is not a courtesy, it is a dependency play: no-cost beta functionality for a preliminary term, aimed squarely at building operational reliance among end users so the uplift at renewal is unarguable. Once 800 people run their day through a tool, your withdrawal is theatre and the account team knows it, because their telemetry shows daily active usage while you claim indifference. Test this before you go quiet. If you cannot switch the tool off for ten business days without a business unit escalating to your CIO, you do not have a silence tactic, you have a renewal. Fix the dependency first (cap the pilot population, keep a second model in production, hold the data outside the vendor), then go quiet. Price the platform before usage grows, not after. First action this week: map your live meters and your pilot dependency in one page, then decide whether silence is leverage or an invoice.
Do not re-enter with a phone call. A call lets the rep control the sequence, reset the anchor, and walk you back through the value story you already rejected. Re-enter in writing, one document, with a specific price, a specific structure, and a specific signature date, and make clear that the date expires. Market experience across GenAI renewals says the deal that closes is the one where the buyer names the number first after the silence, because the vendor has spent three weeks building an internal case for a concession and needs a target to attach it to. Give them the target. The benchmark to hold against is the worked outcome: a $5.6M opening (5,000 seats at $60 plus a $2M API commit) resolved at $3.9M, 23% below opening. That reduction did not come from a discount argument. It came from splitting seats from consumption and pricing each on its own logic: 3,800 seats matched to verified active users at a 12% volume discount, and the API commit set at $1.4M after model routing moved the traffic mix off premium models. An explicit price hold across the two-year term stopped the vendor from recovering the concession at month 13. That structure, not the headline percentage, is what your re-entry letter should describe.
First move this week: confirm the notice date, then build the seat count from actual active users before you draft anything. See the timing and leverage playbook for the opening sequence, and price the platform before usage grows.
Fourteen to twenty-one days is the working range. Under 10 days the vendor reads it as scheduling friction and nothing moves. Beyond 45 days you fall out of the active pipeline and reconstructing the deal typically costs 6 to 8 weeks, which is worse than the discount you were chasing.
It reduces the price when the vendor's forecast depends on your signature and you have no hard deadline. Against OpenAI or Anthropic, where seat pricing is fully discretionary and unpublished, a 21 day hold near quarter end has moved deals into the 25 to 45% discount band. Against Microsoft Copilot, where the $30 add-on is anchored and adoption is under 5% of the installed base, silence alone rarely moves the unit price and you need volume banding or term commitment instead.
When a pilot is converting to production, when a SKU retirement notice has been served, or when your autorenewal notice window is under 90 days. In those cases the clock is working for the vendor and every day of silence transfers leverage. Deal with the deadline first, then consider withdrawal.
Send one short message: the evaluation is paused pending an internal budget review, you will come back when there is something to discuss, and no date is offered. Do not attach a counter, do not explain the reasoning, and do not answer follow ups. A silence with a rationale attached is just a slow negotiation.
Not without shutting off consumption first. Agentforce PreCommit carries a shortfall true-up and unused Flex Credits do not roll over, and agent runtime meters such as Windows 365 for Agents bill $0.40 per hour plus $5 per always-available Cloud PC regardless of whether you are talking to the vendor. Freeze usage before you freeze the conversation.
Measure against the opening ask, not list. A strong outcome is 20 to 30% below the vendor's opening number with seats and consumption priced separately and a written price hold across the term. A 15% bundle saving that comes from mechanical SKU packaging is not a negotiated concession and should not close the deal.
The buyer side playbook for When to open a GenAI negotiation, and when to go quiet, free behind a work email.
Gated with a work email on the download page. No sales follow up you did not ask for.
Get the White Paper →
500+ enterprise clients. 11 vendor practices. Industry recognized. One conversation can change what you pay for the next three years.
One buyer side briefing a week. Renewal signals, audit moves, and the levers that work. No vendor spin.
