The Contract Clause That Saved My AI Consulting Business
In year two of AI consulting, I got burned. Client disputed the deliverable. "This isn't what we agreed." I had a contract — but it didn't have specific acceptance criteria. My vague "AI system" description meant we were arguing about what "done" meant.
$14,000 project. Six-week dispute. Eventually settled at $8,000 to make it go away.
Here's what I added to every contract after that.
The 5 Clauses Every AI Consulting Contract Needs
1. Acceptance Criteria (The Lesson I Learned)
Never say "deliver an AI system." Define success in measurable terms:
System acceptance requires all of the following:
(a) Classification accuracy ≥ 92% on the provided test dataset
(b) API response time ≤ 400ms at p95 under 100 concurrent requests
(c) Successful processing of 500 test invoices without human intervention
(d) Client sign-off on UAT results within 10 business days of delivery
If they can't sign off on UAT within 10 days, the project is deemed accepted.
2. Scope Creep Clause
Any request for functionality, features, or deliverables not explicitly listed
in Section 2 (Scope of Work) constitutes a Change Order. Change Orders require
written approval and are billed at $[RATE]/hour with a minimum 4-hour increment.
This one phrase stops 80% of scope creep conversations before they start.
3. Data Warranty
Client warrants that all data provided for model training or system integration
is lawfully collected, properly licensed for AI training, and compliant with
applicable privacy regulations including GDPR and CCPA. Client indemnifies
Consultant against any claims arising from data sourcing or compliance failures.
AI projects constantly involve client data. This shifts liability where it belongs.
4. AI Output Disclaimer
Consultant does not warrant that AI-generated outputs will be error-free,
unbiased, or suitable for all use cases. Client is responsible for human review
of AI outputs before use in consequential decisions including but not limited to
hiring, lending, medical, or legal matters.
Regulators are watching AI. This clause protects you from downstream liability.
5. IP Ownership (Read This Carefully)
Client owns all custom models trained exclusively on Client's proprietary data.
Consultant retains ownership of general-purpose tooling, frameworks, and
methodologies developed during the engagement. Consultant may reuse methodologies
but not Client-specific trained models in future engagements.
This lets you build reusable frameworks (huge for efficiency) while still giving clients what they need.
What Most Consultants Miss: The SLA
If you're doing retainer or post-delivery support, define it exactly:
- Response time commitment (business hours vs 24/7)
- What constitutes "support" vs "new work"
- Escalation path
- What happens when you don't meet SLA
The full AI Consulting Contract & SLA Pack with all five clauses, a change order template, and a dispute resolution addendum is at wedgemethod.gumroad.com.
Top comments (0)