Checklist
AI business buyer checklist
An AI business buyer checklist covers where training data came from and whether the business holds rights to use it, who owns the model and fine-tuning work, how dependent the product is on a third-party foundation model provider, usage-based revenue verification, and technical team retention — questions a typical software checklist does not fully cover.
This checklist covers what to verify before buying a Canadian AI business — one built around a proprietary model, a fine-tuned application layer, or an AI-enabled product. It builds on the general software business buyer checklist but focuses on the questions specific to AI: where training data came from, who owns the model, and how dependent the product is on a third-party provider.
Confirm training data provenance and rights
Get a documented account of where training data came from and confirm the business actually holds the rights to use it commerciallyA red flag is training data scraped, licensed informally, or sourced from customer data without clear consent, since that gap can create legal exposure that only surfaces after a customer, competitor or regulator asks questions.
Where the product was trained or fine-tuned on customer data, confirm what consent was obtained and how that is documentedUsing a customer’s data to train a model without clear consent can breach that customer’s contract and Canadian privacy law at the same time, and it becomes the buyer’s liability the moment ownership changes.
Confirm who actually owns the model and the IP
Confirm ownership of any proprietary model, model weights or fine-tuning work sits with the corporation, not an individual founder or an outside collaboratorModel weights and fine-tuning artifacts are a form of IP that can be as easy to leave out of a formal assignment as source code, and a missing assignment leaves a gap in what the buyer actually owns.
Confirm every founder, employee and contractor who contributed to model development, data pipelines or prompt engineering signed a written IP assignmentThe same assignment gap that affects traditional software applies here, and it is worth checking specifically for contractors brought in for model work, since that engagement is often less formally documented than core engineering hires.
Confirm ownership and licensing terms for any third-party or open-source foundation model the product was built or fine-tuned on top ofA model built on a foundation model licence with commercial-use or redistribution restrictions can limit what the business is actually allowed to sell, a gap that only surfaces when someone reads the licence closely.
Assess dependency on third-party AI providers
Confirm which parts of the product depend on a third-party foundation model or API provider, and review the terms and pricing of that relationshipA business built on top of a third-party model provider is exposed to that provider’s pricing changes, capability changes or access restrictions in ways that are outside the business’s own control.
Model the cost of API or inference usage against revenue per customer, rather than accepting a blended margin figureUsage-based AI costs can scale differently than the revenue they generate, and a business that looks profitable on a blended basis can have a margin structure that gets worse, not better, as it grows.
Confirm what would happen to the product if the primary third-party AI provider changed its terms, pricing or availability with limited noticeA product with no realistic path to an alternative provider is more exposed than one that could migrate, even imperfectly, and that flexibility is worth pricing into the deal.
Verify revenue, customers and technical retention
Reconcile reported revenue against actual payment records, and separately verify any usage-based or consumption pricing calculationsUsage-based billing is easier to misstate than a flat subscription fee, and a buyer should confirm the reported numbers independently rather than relying on a dashboard the seller controls.
Review customer contracts for data-processing terms, liability caps tied to AI-generated output, and any change-of-control clausesA customer contract with an uncapped liability exposure tied to how the AI performs is a risk that transfers directly to the new owner along with the relationship.
Confirm key technical staff with irreplaceable knowledge of the model, data pipeline or prompt architecture intend to remain through and after the transitionAn AI product can depend on a very small number of people who understand how the model was actually built and tuned, and losing that knowledge right after closing can be difficult to recover from.
Ask directly about known bias, accuracy or safety issues in the model and how they have been tested for and addressedA seller unable to describe what testing has been done for accuracy or bias, rather than a documented process and results, is a pattern worth taking seriously before relying on the product’s reliability.
Sources
Every item on this checklist traces to a primary source. Links were last confirmed on the dates shown.
- 01Canada Revenue AgencyGovernmentSelling a business
- 02Office of the Privacy Commissioner of CanadaGovernmentThe Personal Information Protection and Electronic Documents Act (PIPEDA)
- 03Treadstone LawLegal commentaryIntellectual Property Due Diligence When Buying a Business in Ontario
- 04Treadstone LawLegal commentaryCybersecurity and Data Privacy Due Diligence When Buying a Business in Ontario
- 05Treadstone AssociatesAdvisoryArtificial Intelligence Services
Deavo is an advertising and listings platform, not a brokerage, law firm or valuation firm. This page is general information, not legal, tax, accounting or valuation advice, and rules differ by province. Confirm anything you rely on with a qualified professional before you act on it.