Computer-vision business due diligence

Due diligence on a computer-vision business looks different from a standard small-business review because the assets that matter most — training data, trained models, and the right to keep using both commercially — are intangible, easy to misrepresent by omission rather than outright lie, and expensive to fix after the fact. A buyer under a letter of intent needs a checklist built around three questions: who actually owns what the company is selling, who built it, and what happens to the people whose images or biometric data the system has processed along the way.

The documents to request

• Every licence or agreement covering the training data currently in use — not a summary, the actual agreement — confirmed to cover commercial resale rather than only internal research or evaluation
• Signed intellectual-property assignment agreements from every employee and contractor who worked on the core detection or classification models, with any gaps flagged before closing rather than discovered after
• Accuracy and performance logs broken out by deployment site over time, not a single aggregate figure, so drift and degradation on real installations are visible rather than hidden in an average
• The hardware and upstream vision-model vendor agreements currently in force, including any change-of-control, assignment or termination clauses that could affect the business after closing
• Documentation of what consent or notice process, if any, applies to biometric data the system collects, and any prior correspondence with a privacy regulator on that use

Registry and public-record searches

A patent and trademark search through the Canadian Intellectual Property Office confirms whether any detection method the company claims as proprietary is actually filed, and in whose name — it’s common enough to find a patent application still sitting in a founder’s personal name rather than the corporation being sold. An execution and judgment search, run in the relevant provincial registry, confirms there’s no outstanding judgment against the corporation or its key hardware assets that could attach after closing and complicate the transfer. Neither search is optional simply because the company is technology-focused rather than asset-heavy; unresolved IP ownership is, if anything, a bigger risk in a computer-vision deal than in most other small-business purchases, precisely because the technology is most of what’s being bought.

Findings that commonly kill computer-vision deals

Three findings recur often enough to be worth naming directly. Training images sourced from customer sites, public scraping, or a partner feed under a licence that never covered commercial resale is the single most common deal-breaker, because it means the buyer would be inheriting a product built on data it can’t legally keep using going forward. A missing IP assignment from a contractor who built a core detection model is close behind, since it leaves open the possibility that the contractor — not the company — actually owns the model outright. And biometric data collected without documented consent, or without notice to the relevant regulator where required, creates liability exposure that transfers directly to the buyer, not the seller, the moment the deal closes.

What a finding actually means when it appears

Not every finding is a deal-breaker, and treating them all as equally serious wastes negotiating leverage on the buyer’s side. A licensing gap on a small, replaceable slice of the training data is usually a fixable pre-closing condition, not a reason to walk away entirely. A missing IP assignment from a former contractor who’s easy to locate and willing to sign is a delay, not necessarily a collapse of the deal. What separates a fixable finding from a genuinely serious one is usually whether it can be resolved before closing with a signature or a relicensing agreement, or whether it requires rebuilding a core asset from scratch — and getting a straight answer on which kind of finding you’re looking at is exactly what technical and legal diligence, run together rather than in isolation, are for.

Single-vendor dependency as a diligence finding, not just a risk factor

Heavy dependence on one upstream vision-model provider or one hardware vendor shows up in diligence as more than a strategic concern to note in a memo — it’s a contractual fact to pull and read directly. Confirm what happens if that vendor changes its terms of service, raises pricing, or discontinues the product line the business runs on, and whether the business has any contractual protection against that risk at all. A buyer who finds no such protection isn’t necessarily walking away from the deal, but is pricing in the real cost of eventually re-platforming onto a different vendor — a cost the purchase price should reflect rather than leave as an unpriced surprise for later.