AI implementation and integration business due diligence

Due diligence on an AI implementation and integration business is largely about separating what looks like revenue and IP from what actually is revenue and IP, because project-based delivery work staffed with contractors and billed partly as pass-through model-API cost creates more ambiguity on both fronts than a typical services business carries. The findings that actually kill or reprice deals in this sub-sector are specific and recurring, which means a buyer who knows what to look for can move through diligence faster and with far more confidence than one working from a generic checklist.

Trace IP ownership through every contractor who touched the code

Request the complete list of everyone — employee or contractor — who worked on the code and configuration delivered to clients, and match each name against a signed IP assignment on file; any gap in that chain means the business cannot cleanly warrant ownership of what it is selling. This is not a formality — a contractor who never assigned their work retains rights in it under Canadian law by default, and a buyer who acquires a business with this gap unresolved is acquiring exposure, not just a technicality to be cleaned up later. Where gaps exist, get them resolved, or priced, before closing rather than after.

Separate real service margin from pass-through model-API cost

Work through the revenue recognition policy line by line to determine how much of billed revenue is genuine delivery fee versus model-API or compute cost the firm is simply forwarding to a client at or near cost, because the two behave completely differently as a valuation input and a seller’s own reporting does not always draw the line clearly. Ask for engagement-level margin data rather than firm-wide averages, since a firm can show a healthy blended margin while individual engagements are actually running at a loss once true delivery cost is accounted for, and averaged figures can hide exactly the engagements most likely to cause trouble after closing.

Confirm the extent of foundation-model vendor dependence

Map every active client engagement against which foundation-model or cloud vendor it depends on, and ask specifically whether any single vendor’s pricing, terms or product roadmap changes could materially affect the economics of that work; a firm that has never had to answer this question in writing is a firm that has not thought carefully about its own concentration risk. Review the underlying vendor or reseller agreements for termination and pricing-change provisions, and confirm whether the firm’s certified partner status is personal to the current ownership or transfers with the business — the two produce very different post-closing pictures even when the client-facing work looks identical.

Check for unauthorized use of client data and warranty exposure on live systems

Ask specifically whether client data was ever used to build, test or improve the firm’s own internal accelerators or tooling, and if so, whether the underlying client contract actually permitted that use — repurposing a client’s data for the firm’s own product development without contractual permission is a distinct and serious finding, separate from ordinary data-privacy compliance. Also ask what warranty or liability exposure exists on AI systems the firm has already built and handed over to clients that are now running in production, since a defect or a compliance gap in a live system can surface as a claim well after closing, and you want to know before you buy whether that exposure sits with the seller, the firm’s insurance or, by default, with you as the new owner.

Reconcile unbilled work-in-progress and any change-order history

Pull the full unbilled work-in-progress schedule and ask for the history behind every entry older than a normal billing cycle, because an aged WIP balance is very often a fixed-price project that ran over budget and turned into a change-order dispute the parties never fully resolved. Confirm whether any client has disputed an invoice, withheld payment, or threatened to, and treat an unusually large or old WIP balance as a prompt to ask harder questions rather than as a routine timing difference — in this sub-sector it is one of the more reliable early signals of a delivery or client-relationship problem building underneath the numbers. Ask, too, whether any WIP write-offs happened shortly before the sale process began, since a seller cleaning up the balance sheet just ahead of a listing is not necessarily doing anything wrong, but it is worth understanding what prompted the timing.

What a finding here actually means

• A contractor IP gap means you may not actually own the software you are paying for, not a paperwork detail to clean up later
• A large pass-through share of revenue means the real business is smaller than the top line suggests, which should show up directly in your offer
• Undisclosed single-vendor dependence means the delivery economics could change the moment that vendor changes its terms, independent of anything the acquired business does
• An aged, unresolved WIP balance is usually evidence of a client relationship or delivery problem, not simply a billing timing issue