The 30-Minute Read
A serious AI vendor proposal runs 8–20 pages. You can evaluate one in 30 minutes with a structured read: executive summary, scope, pricing, SLA, data, termination. Most of the “impressive” pages are filler. The deal-breaking pages are usually 4.
Executive Summary Signals
A good executive summary names the specific business problem, the specific outcome, the timeline, and the price. A bad one talks about “AI transformation” in the abstract. If the executive summary doesn't name your specific use case, the rest of the proposal is generic.
Scope Clarity
Look for: clear deliverables, dependencies, who provides what, milestones. Look out for: vague phrases like “ongoing optimization” without definition. Vagueness in scope is where scope creep lives.
Pricing Structure
The structure matters more than the number. Three things to check:
- Variable components. What changes the bill? Usage, seats, API calls?
- Pass-throughs. Are AI provider costs on top of the listed price?
- Annual escalators. What's the price in year 2?
- • If the price is much lower than competitors, ask what's missing.
- • If pricing is “custom only,” ask for a typical range.
- • If there's no usage cap, ask what an upper bound looks like.
SLA and Reliability
Look for: uptime guarantee, response time on incidents, credits for breach. Look out for: SLAs with no enforcement mechanism (just “we'll try”).
Data Provisions
The non-negotiables: data ownership stays with you, no training on your data, deletion on contract end, location of data residency, breach notification terms.
Termination and Migration
What happens at end of contract? Can you export your data? What format? What's the off-boarding timeline? If the answers are vague, you have vendor lock-in by design.
Top 10 Red Flags
- No specific success metric.
- Auto-renewal with short notice window.
- No SLA enforcement.
- Data training rights granted.
- Vague “professional services” line items with no scope.
- No off-boarding terms.
- Hidden pass-through costs.
- Single-vendor dependency on a third-party model with no fallback.
- References that can't be contacted independently.
- Pricing changes mid-negotiation.
Every AI vendor proposal is a 30-minute test of whether the vendor wants a partner or a hostage. Read for that.
FAQ
Should we always negotiate? Yes, especially on data terms and renewal mechanics.
What about pilot terms? Insist on pilots with clear success criteria and a no-obligation off-ramp.
What if it's a startup vendor? Lower the risk: shorter contract, escrow source code (if critical), and survival clauses.