GenAI commitments sized on vendor forecasts over-run actual consumption by 22 to 38 percent, and only a swap clause converts that stranded spend back into value

Every enterprise GenAI deal signed in 2026 spans at least three meters: a seat at $21 to $99 per user per month, an agent credit at $0.01 apiece, and a consumption pool billed at API rates on top. The forecast that sizes those meters is the vendor's, the telemetry is yours, and the gap is 22 to 38 percent. A substitution and reallocation clause is the only mechanism that lets you move that money to the meter your users actually chose.

Prepared by Redress Compliance · September 6, 2026 · GenAI and agent pricing advisory. Microsoft, Salesforce, OpenAI and Anthropic enterprise negotiations, 2024 to 2026.

A swap right is not a discount and it is not portability.

It is the ability to move dollars you have already committed across three axes inside one paper: SKU to SKU (Copilot Business at $21 to a full Copilot add-on at $30, or an Agentforce user license at $5 to Flat Fee Access at $125).

Meter to meter (seat dollars into Copilot Credits at $0.01 or Flex Credits at $500 per 100,000), and entity to entity (the division that over-forecast funding the division that under-forecast).

Each axis spans a different unit of account, which is exactly why vendors keep them separate. Microsoft's July 1, 2026 repricing pushed E3 to $39 and E5 to $60 while holding the Copilot add-on at $30, and shipped E7 at $99 as a single bundled container of E5, Copilot, Agent 365 and Entra Suite.

E7 is not a discount vehicle. It is an anti-swap container: once the seat, the agent identity and the suite live in one SKU, there is nothing left to reallocate. Salesforce is more honest about the mechanism because it sells the currency directly.

Flex Credits are already fungible across Actions, Prompts, Translations and Voice Actions, and the same three-action ticket costs $0.30 on Flex Credits versus $2.00 on the Conversations SKU, a 6.7x delta on identical work. That arbitrage is what a reallocation clause captures.

The standard exclusions are predictable and appear in nearly every first draft we review: no reduction in seat count, no movement into a lower-margin SKU, no rollover past the Order End Date, and no cross-entity movement without a new order form.

The clause has three parts and each does one job. Part one replaces the unit commitment with a committed value construct: "Customer commits to spend not less than $X across the Term.

Commitment is measured in currency, not in units, quantities, seats, credits or conversations." That single substitution moves the entire forecasting risk.

If the vendor's sizing is wrong by 22 to 38 percent, which is the range we see across engagements, the money is still spent, just spent correctly. Part two grants the reallocation right: exercisable quarterly on 30 days written notice, no repricing, no new order form, no approval.

Vendors will push for annual exercise and 90 days notice; that pushes correction outside the quarter in which the telemetry appeared, which is the point of the resistance.

Part three locks the conversion table at signature, so the ratio between a seat dollar and a credit dollar cannot be repriced mid-term. Without it, the vendor grants the right and then moves the exchange rate, and you have negotiated a right worth nothing.

Pair this with a price hold on the underlying rate card, or the conversion table floats on top of a moving base.

Two sentences carry the legal weight. The first: "All substitutions are valued at Customer's then-current contracted rates, not list price." Vendors will accept a swap right and then convert at list, which on a 30 percent Copilot discount destroys roughly a third of the value on every move.

The second: "Reallocation does not constitute a reduction, cancellation or termination of the Commitment, and shall not trigger any true-down restriction, minimum quantity floor or discount recapture." Without that sentence.

The swap right collides with the seat-count floor elsewhere in the agreement and legal will read the floor as controlling.

Add the anti-expiry language directly at the Order End Date rule: "Credits.

Capacity and prepaid units purchased under this Order do not expire at the Order End Date and carry forward into any successor Order or renewal Term." Salesforce will resist this harder than any pricing point, because expiry is where the unconsumed forecast becomes revenue.

Extend the entire construct across both the credit and seat meters so a single affiliate definition covers cross-entity movement.

Ask a Microsoft or Salesforce rep for ten points of discount and you will get a process: an approval matrix, a desk review, a concession traded against term length or a workload you were going to buy anyway.

Ask the same rep to let you move committed dollars between a seat SKU and a credit pool mid-term, and the conversation changes character entirely. The resistance is disproportionate to the money involved, and that tells you the objection is not about margin.

A discount reduces revenue by a known amount on a known SKU in a known quarter. A swap right does something worse from the vendor's side: it makes the composition of the revenue unpredictable while leaving the total intact.

Understand what the account team is actually protecting. Quota is not built on total contract value alone. It is built on SKU mix, because product-line targets, accelerators and channel comp all attach to specific product codes.

A rep carrying a Copilot seat number and a separate Agent 365 number does not get credit for a customer who moves $400,000 out of the first bucket into the second, even though the customer's total spend has not changed by a dollar.

Worse, the finance organization behind that rep has already recognized ratable seat revenue against a forecast that assumed those seats stay seats. Fungible dollars break the forecast in both directions and create a restatement risk on a product line that the vendor's investors are watching closely.

That is why a 10 percent discount, which costs real money, is easier to win than a full swap right, which costs none.

Read Microsoft's E7 launch in that light. The first new enterprise tier in a decade lands at $99 per user per month and folds E5, M365 Copilot, Agent 365 and Entra Suite into a single container. The buyer-side reading is not that Microsoft found a pricing sweet spot.

It is that bundling is the structural answer to buyers who want decomposable meters. Once seat, agent governance and identity live inside one SKU, there is nothing left to swap between, because the components no longer have independent prices you can point at.

The July 2026 base increases (E3 to $39, E5 to $60) with standalone Copilot SKUs excluded from the rise reinforce the same intent: keep the AI meters attached to a rising base rather than floating where they can be reallocated.

Buyers who want optionality should hold the segmented position (Copilot at 20 to 40 percent of the population, defined by role) precisely because it keeps the meters separable, and separability is the precondition for any swap right at all.

Salesforce chose the opposite trade, and it is instructive. Flex Credits are genuinely fungible across Actions, Prompts, Translations and Voice Actions, and Salesforce markets that breadth as freedom from renegotiating pricing or forecasting volume.

Then it draws a hard line: credits expire at the Order End Date. The vendor conceded fungibility inside its own currency and recovered every point of that concession by time-boxing it.

Breadth of movement is cheap to the vendor because the dollars stay inside the Salesforce currency and the same fiscal year. Duration is expensive, because carryover breaks the annual recognition schedule.

Every vendor you negotiate with will make some version of this trade, and knowing which axis they defend tells you where the concession actually lives.

Which produces the buyer's winning frame: total committed value held flat. Do not open by asking for flexibility, because flexibility sounds like a request for less money. Open by fixing the number the vendor's forecast depends on, then argue only about where inside that number the dollars land.

In that frame the vendor's aggregate does not move, the rep's total attainment does not move, and the only thing changing is your ability to correct a forecast that was theirs to begin with.

It is the one construction where the vendor has no revenue argument left, only an account-planning inconvenience, and account-planning inconvenience is not a thing you should pay 22 to 38 percent of a committed base to spare them.

The same discipline applies across the broader GenAI and agent contract redline set: fix the total, contest the composition.

Five objections arrive in a predictable order, and each has a graded fallback that keeps the conversation moving rather than ending it.

"Our systems cannot do it" is answered with the vendor's own precedent: Salesforce built Flex Credits as a fungible currency across four action types, and Microsoft prices Copilot Credits at $0.01 with 25,000-credit packs, so cross-meter accounting demonstrably exists.

"We can do it at renewal" is answered by pointing out that renewal-only flexibility is worth nothing against a 22 to 38 percent oversizing gap discovered in month five: counter with a mid-term quarterly window.

"Only upward swaps" (seats into credits, never back) is answered with a symmetric 25 percent bidirectional band, because one-way movement is a pre-approved upsell, not a swap right.

"No cross-entity movement" is answered with a named affiliate schedule attached to the order form rather than an open pooling right, which is a concession the vendor can control and audit.

"Ratios set at list" is the one that quietly deletes the whole clause: if a $30 seat converts to credits at list rather than at your contracted net, a 40 percent discount evaporates on conversion. Freeze conversion ratios at contracted net, in writing, in the same paragraph.

The credit versus seat economics only work in your favor when the exchange rate is yours.

A strong outcome reads as one sentence: at least 25 to 30 percent of annual committed value reallocable across named meters each contract year, exercisable quarterly, at conversion ratios frozen to contracted net, with express confirmation that no exercise constitutes a repricing event, a true-up.

Or a change to tier eligibility.

Do not walk in asking for "flexibility." Walk in with a percentage derived from your own telemetry, because a round number invites a round counter and a computed number forces the vendor to argue with your data instead of your ambition.

The method is simple: take the last two full quarters of actual consumption per meter (seats activated versus seats billed, credits burned versus credits purchased, API spend versus pooled allocation), annualize it, and compare it to the commitment the rep's forecast produced.

Across the engagements we run, that gap lands at 22 to 38 percent. That range is your evidence base, not your ask.

Your ask is the top of your own measured gap, rounded up, with a floor of 25 percent, because anything below 25 percent gets consumed by normal seasonal variance and delivers no real optionality.

Work the Claude Enterprise example. An 800-person org gets rep-supplied math of roughly $1.1M per year: a ~$20 per user per month seat that buys access only, with all usage billing at API rates on top.

That structure is precisely the one where the seat leg holds and the usage leg runs 30 percent light, because rep forecasts assume every licensed user becomes a heavy user. A 30 percent swap band on that deal keeps $330,000 of committed spend in play rather than expiring against a meter nobody used.

Even at the 25 percent floor it is $275,000. Frame it that way in the room: you are not asking for a discount, you are asking that money you have already promised to pay remains spendable.

That is a materially easier concession for a rep to book than list price movement, and our analysis of credits versus seats shows why the meter you commit against usually is not the meter your users pick.

Then attach the aggregation math, because it is what makes the ask defensible rather than opportunistic. Every additional workload you pull onto an enterprise agreement deepens the tier discount by 3 to 5 percent but adds 100 percent of its own commitment to the base.

The vendor sells that trade as inevitable. It is only rational if the added commitment is fungible. Without swap rights, you are buying a 4 percent discount with a full new commitment you cannot redirect. With a 30 percent band, aggregation stops being a trap and starts being a portfolio decision.

Three patterns repeat. First, most Copilot deployments run three simultaneous meters (per-user seat, per-action credit consumption at $0.01 per credit, and Agent 365 where agents are governed as identities), and buyers budget only the first.

The surprise bill arrives from meters two and three while meter one sits underconsumed, which is the exact shape a substitution clause is built to fix.

Second, the 2026 repricing environment turns stranded commitment into a compounding liability rather than a one-year write-off: SaaS inflation running at 13.2 percent, Salesforce at +6 percent on list, Microsoft at +5 to +16 percent across suites (E3 $36 to $39.

E5 $57 to $60) and up to +43 percent on frontline (F3 $8 to $10), Slack at +20 percent.

Unused commitment does not simply expire, it gets rebased upward at renewal. Third, roughly 60 percent of vendors mask increases during negotiation, folding uplift into bundling or SKU changes rather than stating it.

That makes a swap right an inflation hedge as much as a utilization tool: money that can move to a cheaper meter is money the vendor cannot reprice by moving you to a more expensive one. Pair it with the uplift cap redline and the two clauses reinforce each other.

Related advisory: the full topic guide.

What is a swap right in a GenAI contract?

A swap right, also called a substitution or reallocation clause, lets you move committed spend between SKUs and meters inside the same agreement without renegotiating price.

In practice it means dollars committed to Copilot seats at $30 per user per month can be redirected to agent credits at $0.01 each, or to a consumption pool, when adoption lands differently than forecast. It is distinct from portability, which is about moving spend to another vendor.

How much reallocation should I ask for?

Ask for at least 25 to 30 percent of annual committed value, exercisable quarterly with 30 days notice. That band is anchored on evidence that vendor-forecast commitments over-run trailing twelve-month consumption by 22 to 38 percent, so anything below 25 percent leaves stranded spend on the table.

Vendors typically open at zero, counter at 10 percent upward-only, and settle at 20 to 25 percent bidirectional if you hold.

Will Microsoft agree to swap rights between Copilot seats and Copilot Credits?

Microsoft resists cross-meter movement because the seat and the credit sit in different revenue lines, and its structural counter is E7 at $99 per user per month, which bundles E5, Copilot, Agent 365 and Entra Suite so nothing needs to be swapped.

The realistic outcome is a limited band, usually 15 to 25 percent of committed value, moving from seats into prepaid credit packs at $200 per 25,000 credits. Refusing E7 until the band is granted is the practical lever.

Does Salesforce already offer this with Flex Credits?

Partially. Flex Credits at $500 per 100,000 are fungible across Actions, Prompts, Translations and Voice Actions, and Salesforce markets that as expansion without renegotiating pricing.

The hard limit is time: credits must be consumed before the Order End Date, with no rollover, and the Agentforce user license at $5 per user per month still requires credits on top. The negotiation target is carry-forward, not breadth.

What is the difference between a swap right and a true-down right?

A true-down reduces total commitment and therefore reduces vendor revenue, which is why it is almost never granted mid-term. A swap right holds total committed value flat and only changes where that value lands, which removes the vendor's revenue objection entirely.

Framing your ask as value-neutral reallocation rather than reduction is the single change that moves this clause from rejected to negotiable.

How do I stop the vendor repricing the conversion ratio mid-term?

Lock the conversion table at signature and state expressly that ratios are calculated at contracted net rates, not then-current list.

Without that sentence the vendor can grant a swap right and then devalue it by moving list price, which is a live risk in a year where SaaS prices rose 13.2 percent and Microsoft suites went up 5 to 16 percent.

Pair the swap clause with a price hold clause so both the ratio and the underlying rate are fixed.

Can swap rights move spend between business units or legal entities?

Only if you name them. Standard GenAI paper restricts consumption to the contracting entity, which strands credits bought by one division while another over-consumes.

The fix is a named affiliate schedule attached to the order form plus a sentence permitting reallocation across any listed affiliate at no additional fee, refreshed annually to catch acquisitions.