The AI renewal cliff, the promo year was the runway
AI add on contracts signed in 2024 and 2025 are now reaching their first renewal anniversaries, and the bills coming back are not the bills that were signed: the cliff is the gap between the promo era signing rate and the rate the vendor now quotes, the first real test of how AI prices outside a promotional window. The vendor opens a category with promo pricing, watches realized usage through the term, and moves the rate to what the data suggests the market will bear, with the first batch of renewing customers funding the test.
Prepared by Redress Compliance · August 8, 2026 · GenAI advisory. Based on 60 to 80 enterprise AI add on contracts supported 2024 to 2025.
Executive summary
The opening asks ran 20 to 45 percent, frequently above the caps buyers thought they had.
First AI renewal quotes carried an opening uplift in the 20 to 45 percent band across the vendors we tracked, and the realized uplift after a structured negotiation landed at 40 to 60 percent of the ask: a normal platform renewal moves in a low single digit band.
And the AI band is wider by an order of magnitude, which is the whole reason the cliff matters.
A 3 percent platform uplift is a budget exercise; a 30 percent AI uplift is a portfolio decision that reprices every business case built on the signing rate.
The cap clause did most of the work, and its absence cost everything. Buyers who secured a capped uplift on the AI line at signing paid roughly half of the opening ask; buyers without a cap paid the whole ask, and cliffs ran past 100 percent where no cap existed.
The strongest structure went further: buyers who treated the AI add on as a separately termed contract, with its own cap and exit, held the realized uplift below 15 percent in most cases, against the bundled alternative where the increase disappears into the platform renewal's blended number.
The vendors reprice differently, and the panel names them. Salesforce Agentforce opened at 30 percent from a zero promo base and ServiceNow Now Assist at 35, the two repricing hardest in the file.
Microsoft 365 Copilot opened narrower at 18 percent but is the largest absolute exposure for most enterprises at its scale.
The repricing mode follows the meter: seat priced add ons move through list actions, consumption models through unit cost and overage rates, and agent products through both at once, which is why the cap language has to name the meter it caps.
The calendar is the exposure, and 2027 is the bigger wave.
The Copilot, Agentforce, and Now Assist cohorts signed inside a roughly eighteen month window from mid 2024, reaching renewal twelve months later.
And the buyers most exposed are those whose promo year expires inside the next nine months, because late starts on cliff renewals carried a measurable cost.
The 2026 wave sets the tone and the 2027 wave is larger by volume, since the broader rollouts came later, and the same mechanic produced the subscription resets of the early 2020s, only faster and larger because AI has no settled pricing history to anchor against.
The vendor panel, promo to first ask
| Vendor and product | Year one promo uplift | First renewal opening ask |
|---|---|---|
| Microsoft 365 Copilot | 0 percent | 18 percent |
| Salesforce Agentforce | 0 percent | 30 percent |
| ServiceNow Now Assist | 5 percent | 35 percent |
| The realized pattern | The promo was the runway | 40 to 60 percent of the ask, with a cap |
The mechanic is the message.
A vendor opens a new AI category with promotional pricing to drive adoption inside the installed base, watches realized usage and willingness to pay through the term.
And at renewal moves the rate to the level the data suggests the market will accept, which is rarely the signing rate: the first batch of renewing customers funds the test.
Agentforce and Now Assist repriced hardest in the file, Copilot moved inside a narrower band at the largest absolute scale, and every business case built on the signing rate reprices with the renewal.
The defense clauses, written at signing
- The capped uplift on the AI line: the single most valuable clause in the file, the difference between paying half the ask and paying all of it.
- The separate term: the AI add on as its own contract with its own cap and exit, the structure that held realized uplifts below 15 percent.
- The meter named in the cap: seat products reprice through list, consumption through unit cost and overage, agents through both, and a cap that names the wrong meter caps nothing.
- The exit that survives the bundle: an AI line that can be dropped without touching the platform renewal, because the bundled increase disappears into a blended number.
- The calendar head start: renewals opened early against the promo expiry, since late starts on cliff renewals carried a measurable cost.
The AI platform contract negotiation brief
The cap language by meter type, the separate term structure, and the negotiation sequence for the first cliff renewal.
Get the white paper →The cliff by category, and the 2027 wave
The repricing mode follows the meter, which decides where the defense points: seat priced add ons like Copilot reprice through list moves the cap clause bounds directly.
Consumption priced models reprice through unit cost and overage rates, where the pool size, the rate, and rollover terms carry the exposure; and agent products reprice both at once, the widest surface.
The 2027 wave will be larger by volume because the broader enterprise rollouts came later, arriving with the 2026 tone already set, which makes this year's renewals the precedent purchases.
The vendor by vendor mechanics run through the practice, the two layer Now Assist bill in the Now Assist strategy, the eight Copilot meters in the Copilot licensing guide, the consumption side in the token cost report, and the off books demand the renewals never see in the shadow AI report.
- Percentile standing for your exact deal size and industry, from real closed transactions
- Scenario simulation before the call: test alternative terms and see the financial impact of each
- A negotiation playbook, talking points, and a two page executive brief on day one
What we saw across AI renewals, 2024 to 2025
Across roughly 60 to 80 enterprise AI add on contracts our team supported between 2024 and 2025, the first renewal was almost never priced at the signing rate:
After a structured negotiation, with the cap clause doing most of the work.
The realized uplift where the AI add on carried its own term, cap, and exit.
The report reads bands from an anonymized file, and the strategic read is calendar shaped: AI add ons now represent two to three times their share of net new software spend on a typical renewal, the buyers most exposed hold promo years expiring inside nine months.
And the difference between the capped and uncapped outcomes is written entirely at signing, before any renewal exists.
The cliff is not a grievance, it is the predictable end of a promotional window, and the buyer who treats the signing rate as a runway price, negotiating the cap and the separate term while the vendor still wants adoption, lands the renewal the report's prepared cohort landed.
Your first five moves
- Map every AI line's promo expiry date now, because exposure concentrates inside the next nine months.
- Cap the AI uplift at signing, naming the meter, the clause that halved the opening ask.
- Term the AI add on separately with its own exit, the structure that held uplifts under 15 percent.
- Refuse the bundle roll in, where the increase disappears into the platform renewal's blended number.
- Open cliff renewals early with a benchmark, since late starts carried a measurable cost. The cost optimization practice runs the renewal with you.
Frequently asked questions
What is the AI renewal cliff?
The gap between the rate a buyer signed in 2024 or 2025, usually inside a promotional window, and the rate the same vendor quotes at first renewal: opening asks ran in a 20 to 45 percent band across the vendors we tracked, with cliffs past 100 percent where no cap existed.
It is the first real test of how AI prices outside a promo window, and the first renewing customers fund the test.
Which AI vendors are repricing hardest?
Salesforce Agentforce and ServiceNow Now Assist in our engagement file, opening at 30 and 35 percent respectively from promo bases of zero and five percent, while Microsoft 365 Copilot opened narrower at 18 percent but represents the largest absolute exposure for most enterprises given its scale.
The realized uplift after structured negotiation landed at 40 to 60 percent of the ask across the panel.
How do you defend against AI renewal increases?
At signing, not at renewal: the capped uplift on the AI line was the single most valuable clause in the file, with capped buyers paying roughly half the opening ask and uncapped buyers paying all of it.
And the separately termed AI contract with its own cap and exit held realized uplifts below 15 percent.
The cap must name the meter it caps, because seat, consumption, and agent products reprice through different mechanisms.
Should AI add ons be bundled into the platform renewal?
No: rolling the AI line into the platform renewal makes the increase disappear into a blended number where it cannot be challenged line by line, while the separately termed AI contract preserves the cap, the exit, and the ability to drop the line without touching the platform.
The bundle trap is the vendor's preferred structure for exactly that reason.
When does the AI renewal cliff hit?
Through 2026 and 2027: the Copilot, Agentforce, and Now Assist cohorts signed inside a roughly eighteen month window from mid 2024, reaching first renewal twelve months later, and the buyers most exposed hold promo years expiring inside the next nine months.
The 2027 wave is larger by volume because broader rollouts came later, arriving with the 2026 precedents already set.
How is the AI cliff different from normal renewal uplifts?
By an order of magnitude: a capped platform renewal moves in low single digits, while first AI renewals opened at 20 to 45 percent, because the category has no settled pricing history to anchor against and vendors are testing what the market will bear.
A 3 percent platform uplift is a budget exercise; a 30 percent AI uplift reprices every business case built on the signing rate, which makes it a portfolio decision.
