Free Guide // White Paper

AI Procurement Checklist: What to Negotiate Before Signing

Enterprise AI contracts are unlike any software deal you've negotiated before. Token-based pricing, data training rights, model deprecation risk, and embedded switching costs can trap organisations for years. This 40-point checklist covers every commercial, legal, and technical term that must be addressed before your organisation commits to any AI platform.

What's Inside This Checklist

Token & Usage Pricing Negotiation

Input vs output token pricing, volume tiers, batch discounts, and rate caps — including enterprise commit structures for OpenAI, Azure OpenAI, AWS Bedrock, and Google Gemini.

Data Rights & Training Restrictions

How to prevent vendors from using your proprietary data to train foundation models. The specific contract language required to protect IP and maintain data residency compliance.

Model Availability & Deprecation SLAs

What happens when a vendor retires the model your applications depend on? Minimum notice periods, transition rights, and pricing freeze provisions during model migrations.

Performance & Accuracy Guarantees

How to contractualise AI performance — accuracy benchmarks, latency SLAs, uptime guarantees, and remedies when model outputs fall below agreed performance thresholds.

AI Governance & Liability Provisions

Indemnification for AI-generated errors, bias liability, regulatory compliance obligations, and audit rights. Critical clauses as AI regulations tighten globally in 2026.

Exit & Portability Rights

Data export format requirements, model fine-tuning portability, integration dependency mapping, and minimum notice periods for contract exit without commercial penalty.

Who This Checklist Is For

CIO

CIOs & AI Leaders

Protect your organisation from AI vendor lock-in, model deprecation risk, and data rights exposure before signing any enterprise AI contract.

CFO

CFOs & Finance Teams

Understand token-based cost models, commit discount structures, and the true total cost of enterprise AI platforms beyond headline API pricing.

Procurement & Legal

The exact contract provisions, data processing amendments, and liability clauses required to protect your organisation in any enterprise AI agreement.

Download Your Free Checklist

Business email required. We respect your privacy.

✓

Your Checklist Is On Its Way

Your guide is unlocked and ready to read below. Scroll down to access the full content.

Guide Unlocked

AI & GenAI Procurement Checklist for Enterprise Buyers

1. AI Pricing Models: What You Are Actually Committing To

Four models dominate enterprise AI pricing: (1) Token-based consumption — OpenAI, Anthropic, Google charge per token. Input and output tokens price differently. A complex prompt can cost $0.01-0.10+. Scale creates significant unpredictability. (2) Seat-based subscriptions — Microsoft Copilot ($30/user/month), GitHub Copilot ($19/user/month), Salesforce Einstein ($50-75/user/month). Predictable, but shelfware risk is high — initial Copilot deployments average 30-40% adoption in year one. (3) API call pricing — flat per-call fee regardless of content volume. Predictable for well-defined use cases. (4) Committed spend discounts — AWS Bedrock, Azure OpenAI, GCP Vertex AI offer 20-40% discounts for annual committed consumption. These are negotiable and underutilised.

2. The 15-Point AI Contract Checklist

Data rights (1-5): (1) Training data prohibition — confirm vendor will NOT use your inputs/outputs to train models without explicit consent. All major vendors offer this for enterprise tier — get it in writing. (2) Data residency — specify processing and storage location; EU customers require EU data residency for GDPR. (3) Data retention limits — require deletion of inputs and outputs within 30-90 days unless you opt in to extended retention. (4) IP ownership — confirm you own the outputs generated using your data and prompts. (5) Subprocessor disclosure — require disclosure of and consent rights over all subprocessors handling your data.

Commercial terms (6-10): (6) Spend caps and automated alerts at defined thresholds — without this, a runaway process generates uncapped costs. (7) Price lock — minimum 12-month price freeze on consumption rates. (8) Most favoured customer clause — as AI prices fall (they are), ensure your contract benefits automatically. (9) Step-down right — if actual usage is materially below committed level, right to reduce commitment without full penalty. (10) Exit provisions — define how you extract fine-tuned models, embeddings, and data on termination.

SLAs (11-15): (11) Uptime SLA — minimum 99.9% for business-critical applications. (12) Response latency SLA — define p95 and p99 latency maximums. (13) Model version control — mandatory notification before updates changing output characteristics. (14) Accuracy metrics — for classification/processing use cases, minimum accuracy as a contractual commitment. (15) Incident response — defined SLAs for AI hallucination incidents and data breach notifications.

Take the Next Step

Apply These Tactics to Your Deal

A 30-minute conversation with a senior IT Negotiations advisor gives you a tailored view of your specific situation, pricing position, and savings opportunity — at no cost.

Speak to an AI Procurement Specialist →Read More Free Guides

Related Resources

AI & GenAI Platform Negotiation

Expert advisory on OpenAI, Anthropic, Google Gemini, AWS Bedrock, and Azure OpenAI enterprise contracts. Token pricing, data rights, and governance frameworks.

Learn More →

IT Contract Negotiation

Comprehensive IT contract negotiation across all major vendors. From SaaS and cloud to on-premise and hybrid — our advisors protect your commercial interests.