Token economics, data residency, indemnity, rate card lock, and the renewal clauses that decide the long term cost of every enterprise AI vendor contract. Buyer side. Independent.
Enterprise AI procurement is not the same as SaaS procurement. The cost moves with usage, not seats. The risk lives in the training data, not the application logic. The renewal price is rarely locked at signature. The twenty questions below run against any enterprise AI vendor contract and surface the clauses that decide the outcome.
The checklist comes out of 70 enterprise AI vendor contracts that Redress Compliance has advised over the last twelve months, across foundation model vendors, application vendors with embedded AI, and cloud platform AI services.
Most AI vendors quote a token unit price at signature and reserve the right to change the price book at any anniversary. Lock the unit price for the term, or accept a small year over year cap.
A rate card lock in an email or a side letter does not survive the contract. Require the lock to appear in the master agreement, not the order form.
Volume commit on prepaid tokens looks like a discount. The discount disappears if you do not consume the commit. Match the commit to a 90 percent confidence consumption forecast, not the vendor projection.
Tokens consumed over the commit run at list rate by default. Negotiate the overage rate at the same discount as the commit, or accept a defined uplift cap.
Hybrid models stack a seat fee on top of consumption. The seat fee anchors a floor below which the cost cannot fall. Watch the seat fee on hybrid contracts as carefully as the consumption rate.
| Commercial question | Default vendor position | Buyer side counter |
|---|---|---|
| Token unit price | Vendor controls price book | Lock unit price for term |
| Rate card lock | Email or side letter | Master agreement clause |
| Volume commit | Vendor forecast | P90 customer forecast |
| True up math | List rate on overage | Same discount as commit |
| Seat versus consumption | Hybrid floor | Either pure consumption or seat with cap |
The default position varies by vendor. Foundation model vendors increasingly default to opt out of training. Application vendors often default to opt in. Confirm the position in writing.
Data residency matters for regulated industries and EU customers. Confirm the region for both the model inference and the prompt log storage.
Vendors swap underlying models without notice in many contracts. Reserve the right to pin a model version for a defined period if the swap materially changes behavior.
Audit logs decide whether you can investigate a regulator inquiry or a model behavior incident. Confirm log retention, export format, and access controls.
A model trained on your prompts is a model that has absorbed your business. The training opt out clause and the data export clause are the two most consequential clauses in any AI contract. They outlive the contract itself.
Generative model outputs can replicate protected material. Three major vendors now offer a defined copyright indemnity. Confirm the scope, the cap, and the exclusions.
Defect indemnity covers loss caused by a model behaving incorrectly. The default is no indemnity in most contracts. Negotiate a defined cap.
Regulator indemnity covers cost of regulator inquiry tied to model output. This is rarely offered but increasingly relevant for regulated industries.
Breach response covers timeline, notification, and remediation. The default vendor position is 72 hour notification. Negotiate down to 24 hour notification for material breaches.
Latency drives user experience on agent and assistant workloads. The vendor default is no latency commitment. Negotiate a percentile based latency target for production workloads.
Throughput limits cap the requests per minute. The vendor default is a generic tier ceiling. Negotiate a dedicated capacity tier for production workloads if the ceiling is binding.
The model swap right lets you move workload from one model to another without contract penalty. Reserve the right in the master agreement.
Uptime credit on AI services is rarely material as a percentage of contract value. Negotiate a credit structure that scales with the severity of the outage, not just the duration.
Data export covers prompt history, fine tuning data, audit logs, and any embeddings stored by the vendor. Define the export format, the timeline, and the cost.
Model continuity covers the ability to continue using a specific model version after non renewal, on a different platform or self hosted. Most vendors do not offer continuity. Negotiate a transition window.
The renewal price lock is the highest value clause in any AI contract. Without it, the vendor sets the renewal terms unilaterally. Negotiate a renewal cap or a defined renewal price grid.
Question 20, the renewal price lock. Without it, every other clause unlocks at renewal. Vendors know this and price accordingly. The buyer side posture is to refuse signature without a defined renewal cap or grid.
Use a P90 consumption forecast based on year one usage. Most enterprise AI deployments see 2.5 to 4 times the year one consumption in year two as adoption matures. Model the cost at the P90 figure, not the vendor projection.
Yes for content generation workloads. The indemnity is now standard at three major vendors and the cap and exclusion structure varies materially. The legal review effort is justified for any creative or marketing workload.
Two to four weeks with a buyer side advisor. The review covers the master agreement, the order form, the data processing addendum, and any product specific terms. The output is a redlined contract plus a scorecard.
Yes, with modifications. Embedded AI features in Microsoft, Salesforce, ServiceNow, and Workday all carry similar clauses inside the parent contract. The same checklist applies, scoped to the AI addendum.
We review the master agreement, run the scorecard, redline the contract, and sit at the negotiation table with your procurement and legal teams. We are not an AI vendor partner and do not take vendor commissions.
The cost of an enterprise AI contract in year two is not the cost in year one. The price stays the same. The token consumption triples. The contract decides who absorbs it.
A buyer side playbook for AI platform contract negotiation. Token economics, rate card lock, and indemnity language.
Independent. Buyer side. Written for CIOs, CFOs, and procurement leaders carrying enterprise software contracts. No vendor influence. No sales kickback.
Open the white paper in your browser. Corporate email only.
Open the Paper →
Independent. Buyer side. We have advised on 70 enterprise AI vendor contracts in the last twelve months.
Buyer side GenAI insight. No vendor influence. Read in five minutes.
Once a month. Audit patterns, renewal benchmarks, vendor commercial signals across Oracle, Microsoft, SAP, Salesforce, IBM, Broadcom, AWS, Google Cloud, ServiceNow, Workday, Cisco, and the GenAI vendors. No follow up sales pressure.
Free providers (Gmail, Yahoo, Outlook) cannot subscribe. Work email only. Unsubscribe in one click.
