Buyers who table a hard consumption ceiling and a 5 to 7% uplift cap before signature beat the vendor's 20 to 37% AI uplift demand by 8 to 27 percent on total three-year cost

The 2026 GenAI quote runs three meters at once: the seat, the credit, and the agent identity. Vendors price the first one visibly and leave the second and third open-ended, which is where the 20 to 37% AI uplift and the five-to-six-figure monthly surprises actually live. The clauses below are ranked by what buyers realistically win, with the replacement wording to paste into the redline, because the discount on the order form is undone by the schedule you did not write.

Prepared by Redress Compliance · August 22, 2026 · GenAI and enterprise software advisory. Renewal and first-purchase engagements, 2024 to 2026.

The 2026 GenAI quote is not a price list, it is three price lists stapled together, and the vendor will only defend one of them in the room. Meter one is the per-user seat: visible, benchmarked, and the one your CFO already has a number for.

Meter two is per-action consumption: Copilot Credits at $200 per 25,000-credit pack or $0.01 pay-as-you-go, Agentforce Flex Credits at $500 per 100,000 credits ($0.10 per standard action, $0.15 per voice action).

Meter three is the agent identity itself, which is what Agent 365 and the Agentforce editions are actually selling. Sequence matters here because the discount you win on meter one is the discount the vendor uses to justify holding the line on meters two and three.

I have watched buyers celebrate 22% off a seat SKU and then absorb a consumption line that ran five figures a month within two quarters, because the credit burn multipliers were never in the contract.

A classic answer costs 1 credit, a generative answer 2, premium AI tools run up to 100 credits per 10 responses, and generative voice costs 35 credits per minute.

That is a 5x to 30x spread inside a single unit the vendor calls "a credit," and the vendor controls which end of the spread your users land on through product defaults you do not set.

The frontier labs have made the seat conversation easier and the consumption conversation harder. Claude Enterprise lists at $20 per seat per month, and the seat is an access gate, not an allowance: usage bills separately at standard API rates with nothing included.

Gemini Enterprise Business starts at $21. ChatGPT Enterprise stays quote-only, with 2026 procurement reporting converging on $45 to $75, roughly $60 average, a 150-seat minimum and annual prepay, which sets a realistic entry near $108,000.

When three vendors sit within a dollar of each other on the visible meter, "our list is our list" stops working as an argument and the negotiation migrates to where the vendor still has cover: the unit definition and the volume driver.

Treat each meter as a separate schedule with its own cap, its own baseline, and its own termination consequence. The AI contract red lines that hold up in year two are the ones written per meter, not per agreement.

Rank this first because it is the clause buyers win most often and the one that compounds hardest.

The vendor is walking in with a 20 to 37% AI uplift demand against a market backdrop of roughly 12.2% annualized software inflation, and the gap between those numbers is not cost recovery, it is margin capture during a window when buyers feel behind on AI.

Your counter is a fixed 5 to 7% ceiling, or CPI plus 2, whichever is lower, applied to the aggregate of every meter in the agreement.

Anchor it against what the vendor's own suite did: Microsoft moved E3 from $36 to $39 and E5 from $57 to $60 effective July 1, 2026, roughly 5 to 8%, while explicitly excluding standalone Copilot SKUs from the increase.

When the vendor's published, unnegotiated list movement on core productivity sits near 8%, a 30% ask on the AI line has no defensible cost basis and you should say so in those words.

The cap has to be written against three specific evasions or it will not survive a repricing year.

First, the cap must attach to the effective per-unit price actually paid in the prior term, not to list, or the vendor simply shrinks the discount and holds list flat, which is a price increase with better paperwork.

Second, the cap must follow the SKU through renaming, successor products, and repackaging: "the Cap shall apply to each Licensed Item and to any successor, renamed, repackaged, or functionally equivalent item offered by Vendor in substitution." Third.

It must survive tier consolidation, because the E7 Frontier Suite at $99 launching May 1, 2026 is exactly the mechanism that voids a per-SKU cap by making the old SKU disappear.

Add: "Where Vendor consolidates two or more Licensed Items into a bundled tier, the capped price shall be the sum of the prior-term effective prices of the consolidated items, increased by no more than the Cap."

Predict the counter, because it is the same every time.

The vendor will concede the cap on the seat, congratulate you on a strong deal, and quietly exclude consumption, credits, overage rates, and anything labeled "usage-based services." That exclusion is the whole ballgame given credit burn runs 5x to 30x on premium tools.

The fix is a single sentence in the definitions: "Licensed Items includes all per-user, per-agent, per-action.

Per-credit and consumption-based charges under this Agreement and any Order Form or rate card referenced herein." Then add the "lower of" mechanic: if the vendor's published rate for a metered unit drops during the term, your rate drops with it.

Anthropic cancelled the September 1 move to $3/$15 and made Sonnet 5's $2/$10 permanent on August 11, 2026, and buyers who had reforecast on the higher figure are now over-budgeting by 50%.

A one-line "lower of contracted or then-current published rate" clause costs the vendor nothing to sign in month one and is worth real money in month twenty.

Buyers running the same pattern against consumption vendors should read across to the clause set behind the order form discount, where the same cap-versus-baseline fight plays out on credits.

The consumption ceiling is the clause you will never win after the first overage invoice arrives, because at that point the vendor holds a delivered service, a signed order form, and an accounts payable clock. Table it before execution or accept that your credit spend is uncapped for the term.

The mechanic that makes this urgent is the burn-rate spread: a classic answer costs 1 credit, a generative answer 2, premium AI tools run up to 100 credits per 10 responses, and generative voice consumes 35 credits per minute.

That is a 35x to 100x range inside a single meter, which means a workload mix shift that nobody approved can multiply your monthly bill without a single new seat being sold.

At $0.01 per credit pay-as-you-go, or $200 per 25,000-credit capacity pack, a modest voice deployment running four hours a day across twenty agents burns roughly 1.68 million credits a month. Microsoft's own field guidance concedes that high-volume agents reach five to six figures monthly.

Your job in the redline is to make that number a decision, not an invoice.

Three pieces of paper do the work. First, attach the credit consumption table as a numbered contract annex, not a link to a documentation page the vendor can edit unilaterally.

Second, set a monthly and annual dollar ceiling above which the vendor must notify and cannot bill without written approval from a named signatory. Third, and this is the piece buyers routinely surrender, elect the throttle yourself.

A vendor-imposed throttle degrades service quietly and preserves the vendor's revenue narrative; a buyer-elected throttle suspends non-critical agent classes and forces a conversation.

The wording to table: "Supplier shall notify Customer in writing upon consumption reaching 70% and again at 90% of the applicable Monthly Consumption Ceiling.

Supplier shall suspend further metered consumption for the remainder of the period and shall not invoice for any consumption above the Ceiling absent Customer's prior written authorization signed by an Authorized Approver." Expect the vendor to counter with automatic overage billing at list plus a "courtesy notification" at 90%.

That counter is worth nothing: notification without a suspension right is a receipt, not a control.

The tell that you are negotiating with someone who has authority is whether they will name the suspension mechanism. Vendors who cannot suspend at the platform level will offer a credit true-up instead, which is fine if the true-up is capped and priced at your committed rate rather than list.

Anchor the ceiling at 115 to 125% of your modeled year-one consumption, not at the vendor's forecast, and require the ceiling to reset annually by written agreement rather than by vendor notice.

Buyers who run this alongside the wider clause set in our AI contract red lines work typically land the ceiling in exchange for a modest lengthening of the term, which is a trade worth making.

Assist, action, task, conversation, credit: none of these terms has an external standard behind it. There is no ISO definition of a Copilot Credit and no auditable benchmark for what constitutes one Agentforce action.

The vendor writes the definition, publishes it on a page it controls, and reserves the right to update it. That means a change to what counts as a billable action is a price increase executed without touching your price.

If a vendor reclassifies grounded retrieval from one credit to two, your unit rate is unchanged and your bill doubles.

This is the single most under-redlined clause in 2026 GenAI agreements, and it is more valuable than another two points of discount because it protects the whole base rather than a slice.

Freeze the rate card. The wording: "The Consumption Rate Card attached at Annex [X] is fixed for the Initial Term.

The lower rate shall apply automatically to Customer from the effective date of the reduction, without amendment." That last sentence is the lower-of provision and it costs the vendor nothing to concede in a rising market, which is exactly why you should take it while they still believe rates only go up.

Anthropic proved the point in 2026. Sonnet 5's $2 in, $10 out introductory rate was made permanent on August 11, 2026 and the planned September 1 increase to $3/$15 was cancelled. Teams that had already reforecast on the higher number were carrying roughly 50% surplus in their Sonnet budget line.

Without a lower-of clause, none of that flowed back automatically; it sat with the vendor until someone noticed and asked. Multiply that across a portfolio and the lower-of provision is worth more than the uplift cap in any year where a frontier lab discounts to buy share.

The same logic applies to Salesforce, where the 20-action crossover between Flex Credits and the $2 conversation model means a definitional change to what counts as an action moves you across the crossover without anyone electing anything.

Pair this clause with the model election rights covered separately in this pack and in our Claude enterprise contract clauses analysis.

Expect resistance framed as operational necessity: the vendor will say product evolution makes a frozen rate card unworkable. The concession that closes it is a carve-out for genuinely new capabilities, priced by agreement at introduction, coupled with the freeze on everything already listed.

Do not accept "commercially reasonable notice" in place of ninety days, and do not accept notice without consent. Notice alone lets the vendor reprice and dares you to terminate mid-term, which you will not do.

The single most expensive assumption in an Agentforce deal is that the pricing model you pick at signature is the one that fits your workload twelve months later. It will not be.

Salesforce's own rate card sets standard actions at 20 Flex Credits, or $0.10 each at $500 per 100,000 credits, and offers a parallel $2 per conversation model. The crossover is arithmetic: above roughly 20 actions per conversation, the conversation model wins; below it, Flex Credits win.

Nobody knows their true action-per-conversation ratio before production traffic, and the ratio moves when you add tools, voice (30 credits per action, $0.15) or grounding. Salesforce knows this, which is why the two models cannot run in the same org at once.

That restriction is what turns a pricing choice into a lock-in event, and it is what your redline has to defuse.

Table the election as a contractual right exercisable annually, not a purchasing decision made once.

The wording to paste: "Customer may, effective on each anniversary of the Effective Date and upon thirty (30) days' written notice, elect between the Flex Credits consumption model and the Conversations model for any org, at the unit rates set out in Exhibit A, without repricing, requalification.

Uplift, or loss of any committed-spend credit." Add a mid-term trigger at 90 days after go-live, because the first quarter of real traffic is when your ratio becomes visible and the vendor's rep is still motivated.

Expect the response you always get: the account team will agree in principle, then route it to deal desk, who will offer the election "subject to then-current list pricing." That defeats the clause. Hold the Exhibit A rate lock or the right is decorative.

The second half of this redline is fungibility. Unused seat commitment should convert to credits and credits back to seats at a stated ratio, minimum 1:1 on dollar value, with 90 days' notice.

Salesforce packaging works against you here: the per-user agent SKUs at $125 to $150 per user per month, and Agentforce 1 at $550 and up, are bought through standard contracting and sit outside the pay-as-you-go and pre-commit structures entirely.

So the swap right has to be written to cross that boundary explicitly, naming the per-user SKUs as eligible destinations for reallocated commitment. A strong outcome looks like a single dollar-denominated commitment with named draw-down meters, not three separate pools.

Buyers who get there routinely recover 10 to 15% of a mispredicted commitment; buyers who do not simply forfeit it. The same logic applies across vendors, as our AI contract red lines work shows.

The risk that ends up costing the most over three years is not the price of the SKU you bought. It is the SKU disappearing. Microsoft's M365 E7 Frontier Suite lands May 1, 2026 at $99 per user per month, bundling E5, M365 Copilot and Agent 365 into the first new enterprise tier in about a decade.

Priced against the components, E7 is not obviously a rip-off. That is the point. It is designed so that the sales motion at your next renewal is "you are already paying for most of it," and the AI capability you were treating as optional becomes structural.

ServiceNow ran the same play more bluntly by gating agentic capability behind Pro Plus, taking fulfiller pricing from roughly $160 to $200 up to $240 to $320, a 50 to 60% uplift for buyers who needed one feature on the far side of the gate.

Three pieces of language do the work.

First, legacy SKU continuity: "Customer may continue to license the SKUs listed in Exhibit B for the Term and one full successor renewal term, at the rates stated, notwithstanding any vendor product retirement, consolidation, or repackaging." Second.

Price protection carried across the successor: "If a licensed SKU is retired or superseded, Customer shall receive the successor SKU at the same effective per-unit rate, adjusted only by the uplift cap in Section X, for the remainder of the Term and one renewal." Third.

An anti-bundling line: "Vendor shall not condition access to capability licensed hereunder on Customer's purchase of additional capability not requested by Customer." The vendor will counter that product roadmaps cannot be frozen.

Fine. You are not freezing the roadmap. You are fixing the price at which their roadmap reaches you.

Expect resistance to be strongest on the second clause, because that is the one that actually costs them money. Trade for it: accept a longer term or a modest volume floor in exchange for successor-rate protection.

It is the cheapest concession you will make all year, and it prices a repricing event the vendor has already scheduled. Microsoft-specific sequencing is covered in our Microsoft GenAI contracts guidance.

Strip the branding off the 2026 quote and look at what actually changed. Microsoft raised E3 from $36 to $39 and E5 from $57 to $60 effective July 1, 2026, and pointedly excluded the standalone Copilot SKUs from the increase.

Read that sequencing carefully: the AI product was held flat while the base underneath it moved. That is not a product launch, it is a base-price reset with an AI narrative wrapped around it.

The same pattern shows up across the vendor set in the 20 to 37% AI uplift line item that now appears on renewal quotes.

A decade of discount anchoring had pushed effective per-seat pricing well below list on most enterprise agreements, and vendors have no clean way to walk that back on the existing SKU without triggering a defensive procurement response.

An adjacent capability with no historical discount record solves that problem. The uplift is not priced against your prior effective rate, it is priced against nothing at all, which is exactly why it is being asked for.

The convergence at the seat is the tell. Claude Enterprise lists at $20, Gemini Enterprise Business starts at $21, ChatGPT Enterprise clusters near $60 in the quote-only band buyers report at $45 to $75.

Three vendors with radically different cost structures landing within a few dollars of each other is not competitive discovery, it is coordination by observation. And it is convenient.

Once the seat number is public and matched across the field, the vendor no longer has to defend it, because you can verify it in ten minutes and there is nothing to argue about. The negotiation moves to the number that is settled and away from the number that is not.

Meanwhile the consumption meter stays opaque: $200 for a 25,000 credit pack or $0.01 per credit at pay-as-you-go tells you the unit price and nothing about the units, and premium tools, grounding and voice multiply burn by 5 to 30x.

A generative voice minute at 35 credits is 35 times a classic answer. No published rate card resolves that variance. That is deliberate architecture, not immaturity.

The consequence for the buyer who negotiates only the visible line is arithmetic, not opinion. A 1,000 seat E3 rollout runs $828,000 a year all-in, of which $360,000 is Copilot.

Win a hard-fought 15% off the Copilot line and you have saved $54,000 against a bill where $468,000 sits in the platform underneath and the credit meter sits outside both numbers entirely.

High-volume agent estates reaching five and six figures monthly are not a tail risk, they are the observed outcome for teams that budgeted the seat and never noticed the second and third meters.

The buyer who books that 15% as a win has negotiated 6.5% of the visible spend and 0% of the unbounded portion.

This is why the uplift must be treated as a base-price event: the cap has to apply to the aggregate of seat, credit and agent identity spend, not to the SKU carrying the AI label, and the year-two and year-three uplift ceiling has to be written against total contract value rather than per line item.

A 5 to 7% aggregate cap is worth more than a 20% discount on one meter.

The vendor response to an aggregate cap is predictable and worth rehearsing. They will argue the meters are technically distinct products under separate schedules, that consumption is buyer-controlled and therefore uncappable, and that credits are a pass-through cost.

The first two are contract positions dressed as facts, and the third is refuted by their own margin structure. Expect an offer of a soft "true-up review" instead of a cap, which is a cap with the enforcement removed.

Hold the line that the cap is aggregate or there is no signature, and be specific: a fixed dollar ceiling on total year-two and year-three spend, with any consumption above the ceiling billed at the prior year unit rate rather than list.

What actually moves the band is a credible alternative analysis, not an executed switch.

Seat convergence at $20, $21 and roughly $60 means the substitution case writes itself, and Claude's structure (a $20 access seat with usage billed separately at standard API rates, no included allowance) gives you a clean apples-to-apples cost model to put in front of the incumbent.

The point is to be ready to switch, not to switch. Across 2025 engagements that posture produced 8 to 27% against the first proposal, and the variance tracked how early the buyer started far more than estate size.

The decay is sharp: by month three most of the leverage is gone, because the vendor has read your renewal calendar, your pilot has generated internal advocates, and your alternative has stopped being credible.

Start six to nine months out, halt the auto-renewal in writing, and table the aggregate cap before any pilot expansion is approved. The same discipline applies wherever consumption meters sit alongside seats, which is why the AI contract red lines apply regardless of which logo is on the quote.

In a consumption deal the termination clause is not an exit clause, it is a commitment-forfeiture clause, and the vendor drafted it that way on purpose. Prepay against a credit pool, terminate for convenience, and the standard language leaves your unconsumed balance with the vendor.

Table termination for convenience at any anniversary on 90 days notice with pro-rata refund of prepaid and unconsumed commitment.

You will not usually get it, and you should ask anyway, because it sets the anchor for what you will get: reduction rights of 15 to 20% of committed value at each anniversary.

Exercisable in writing, with no requirement to justify the reduction and no clawback of the volume discount tier earned in the prior year.

On a $1.5 million three-year commitment, a 20% annual reduction right is roughly $300,000 of exposure you can retire without litigation. That is the realistic win and it is worth more than the theoretical refund you will not be granted.

Credit expiry is the second forfeiture mechanism and it is quieter. Unused credits must roll to the next contract year or refund at termination, full stop.

Vendors will offer a partial rollover (typically 25%) as an opening concession; push for 100% rollover within the term and pro-rata refund on non-renewal.

Pair that with a ramp structure that makes the pilot reversible: year one commitment at no more than 40% of the projected steady-state, with the year two and year three commitments set by written election 60 days before each anniversary rather than fixed at signature.

A vendor confident in the consumption forecast will accept this. One that resists is telling you the forecast is theirs, not yours. The same principle governs credit-based platforms generally, as the Snowflake clause set shows in a mature consumption market.

Then write the degradation trigger, because model deprecation is the risk nobody prices.

Proposed wording: "Customer may terminate this Order Form, or reduce the committed amount pro rata, upon 30 days written notice if (a) Vendor deprecates, retires or restricts access to any model designated in Exhibit A.

Or (b) measured output quality against the benchmark set in Exhibit B declines by more than 10% across two consecutive monthly measurement periods, with any prepaid and unconsumed amounts refunded within 45 days." Name the models in the exhibit and name the benchmark.

Without both, the clause is decorative.

The clause list is worth nothing if you table it in month eleven of a twelve-month term. Every redline above has a date attached to it, and the dates are not yours to pick, they are set by the vendor's fiscal calendar and by your own auto-renewal mechanics.

The working rule from 2025 to 2026 engagements is blunt: halt the auto-renewal and open the negotiation six to nine months before expiry, and secure Outcome Measurement Agreements and hard consumption ceilings before any document is executed, not as a post-signature amendment.

The reason is the leverage decay curve. By month three of a negotiation cycle, most of your leverage is gone, because the vendor has read your renewal date, your deployment velocity, and your internal announcements, and has priced accordingly.

Fiscal alignment is the second lever and it is quantifiable: Snowflake's January 31 year end makes November to January consistently worth 5 to 8% better per-credit pricing than the same deal closed in Q1 or Q2, and every vendor has an equivalent window. Microsoft's is June, Salesforce's is January.

You do not have to move your renewal to hit the window, but you do have to decide whether a two-month extension at current rates buys you an 8% better consumption rate for three years. It usually does.

Our Snowflake clause work shows the same pattern: the discount lands on the order form, the money moves in the schedule you negotiate before the quarter closes.

Two commitment thresholds change the conversation rather than just the price.

Above $100,000 annual commitment, frontier-lab vendors will waive the per-seat fee entirely where 100% of spend routes to credits or API consumption, which converts a fixed-plus-variable structure into a single variable meter you can actually forecast.

Above $200,000 annual, the published volume token discount runs 10 to 20%. If your forecast lands at $85,000, the correct move is to model whether committing $100,000 with a carry-forward provision beats paying $85,000 plus seat fees.

It frequently does, and the vendor will not volunteer the comparison. Sign-off discipline matters here: the CFO signs the commitment, the CIO signs the consumption ceiling, and legal signs nothing until both schedules are attached as exhibits rather than referenced as URLs the vendor can edit.

The patterns repeat with uncomfortable consistency. First, buyers cap the seat and leave consumption open.

The $30 Copilot seat gets negotiated hard, the Copilot Credit meter at $0.01 per credit gets waved through, and then premium tools burning up to 100 credits per 10 responses and generative voice at 35 credits per minute produce five and six figure monthly bills nobody modeled.

Second, the AI uplift demand has converged on 20 to 37% against general software inflation running near 12.2%, which tells you the uplift is not cost recovery, it is repricing.

Third, the buyers who land at the 27% end of the savings band are the ones who opened six to nine months out; the buyers at 8% opened at three months and negotiated against a clock the vendor set.

Be honest with your own board about which numbers are which.

The Microsoft SKU moves (E3 $36 to $39, E5 $57 to $60, E7 at $99 from May 2026), the Salesforce Flex Credits rate card (20 credits per standard action, $500 per 100,000 credits), the Claude Enterprise $20 seat with its 20-seat minimum and no included usage allowance.

And the Gemini Enterprise Business $21 entry are all vendor-published and citable in the room.

The 8 to 27% savings band, the leverage decay curve, and the fiscal window premium are engagement observations from our practice, defensible but not published rate cards.

And ChatGPT Enterprise at $45 to $75 per seat, averaging near $60 with a reported 150-seat minimum, is a band estimate assembled from 2026 procurement reports, because OpenAI still routes every buyer to a sales call.

Quote it as a band, never as a price, and use it the way it is meant to be used: as the credible alternative that moves the incumbent's number without requiring you to switch.

The wider point is set out in our AI contract red lines work, and it holds here: the seat is the visible meter, and the visible meter is the one the vendor is happy to discuss.

What uplift cap can I realistically win on a GenAI renewal in 2026?

A fixed 5 to 7% annual ceiling, or CPI plus a small margin, is the achievable outcome when tabled before signature.

Vendors are opening at 20 to 37% AI uplift against a general software price inflation rate of 12.2% annualized, so the gap is wide and the cap is the single highest-value clause in the pack.

The critical drafting point is that the cap must apply to the aggregate of all meters, not just the per-seat SKU, because vendors will accept a seat cap and leave consumption uncapped.

Why does the Copilot seat price understate the real cost?

For a 1,000-seat Microsoft 365 E3 rollout the all-in annual figure is $828,000, of which only $360,000 is the Copilot line. The rest comes from the base suite, which repriced on July 1, 2026 (E3 from $36 to $39, E5 from $57 to $60), and from Copilot Credit consumption on agents.

Credits run $200 per 25,000-credit capacity pack or $0.01 pay-as-you-go, and premium tools, grounding and voice multiply burn by 5 to 30x.

Should I choose Agentforce Flex Credits or the per-conversation model?

The crossover is 20 actions per conversation. At $0.10 per standard action, 20 actions equals $2.00, which matches the $2 per conversation rate, so anything above 20 actions favours conversations and anything below favours Flex Credits.

The two models cannot run in the same Salesforce org at the same time, so the redline is not about picking correctly today but about securing an annual right to re-elect once you have real usage data.

When should I start the redline cycle?

Six to nine months before expiry, with auto-renewal formally halted at the outset. Engagement data shows most leverage has evaporated by month three of a compressed cycle, and hard consumption ceilings plus outcome measurement terms have to be secured before any document is executed.

Aligning signature to the vendor's fiscal quarter end is separately worth 5 to 8% on consumption unit rates.

Does the Claude Enterprise seat price include usage?

No, and this is the most commonly misread line in enterprise AI quotes. Claude Enterprise lists at $20 per seat per month billed annually with a 20-seat annual minimum, but usage is billed separately at standard API rates with no included allowance.

Above roughly $100k of annual commitment the seat fee is typically waived entirely with 100% of spend routing to API credits, and volume token discounts of 10 to 20% appear above $200k.

What is the tier consolidation risk and how do I redline it?

The risk is that the SKU you negotiated stops existing.

Microsoft's M365 E7 Frontier Suite at $99 per user per month launched May 1, 2026 bundling E5, Copilot and Agent 365, and ServiceNow gates Now Assist behind Pro Plus with a mandated 50 to 60% base uplift that moves a fulfiller from $160 to $200 up to $240 to $320.

The clause to table is a right to remain on the legacy SKU through the term plus one renewal, with price protection carried across at the same effective per-unit rate to any successor SKU.

How much can I expect to save against the first vendor proposal?

Across 2025 engagements the band was 8 to 27% against the first proposal. Variance is driven by estate size, prior contract posture, and above all how early the buyer started.

A credible alternative analysis usually moves the discount band without running a full RFP; the leverage comes from being demonstrably ready to switch, not from actually switching.