Selling a conversational AI platform in Canada

A conversational AI platform sells on trust as much as on revenue, because the buyer isn’t just acquiring a customer list — it’s inheriting every commitment the platform has made to those customers about what happens to their data and their end-users’ conversations. Preparing well for a sale means resolving the questions a buyer’s technical and legal teams will ask before they ask them, rather than scrambling to answer once due diligence has already started and the seller looks unprepared.

What to fix before listing

• Put a written data-processing agreement in place with every customer covering what happens to conversation data, including whether and how it can be used to train or fine-tune models, retroactively if that hasn’t been documented before now
• Get signed IP assignment agreements from any contractor who worked on model fine-tuning or prompt engineering, closing a gap that’s common on platforms that started as a small technical team
• Document the platform’s human-handoff and fallback design, since a buyer’s counsel will treat the absence of one as a live liability question, not a product nice-to-have to fix eventually
• Review the foundation-model vendor’s terms of service for any resale, embedding or reseller-program restrictions that could limit what a new owner is allowed to do with the platform after a change of control
• Pull together the actual source data behind any containment or resolution-rate figures used in marketing, so they can be defended with evidence rather than merely repeated in a data room

What the regulator wants, and how long that runs

Quebec’s Law 25 does something no other provincial privacy regime currently does explicitly for automated systems: where a chatbot’s output materially affects a person, the law requires disclosure of that fact and grants the affected person a right to have a human review the decision on request. A seller whose customer base includes any Quebec-facing deployments needs a documented process for honouring that right before a buyer’s counsel goes looking for one, because retrofitting it mid-diligence is one of the more common causes of delay in this category. Federal direction on transparency obligations for automated customer-facing systems more broadly is still evolving, which means a buyer’s counsel will typically test against the clearest existing standard — Quebec’s — even for deployments outside the province.

Confidentiality during marketing

What makes a conversational AI platform defensible — its fine-tuning approach, its prompt libraries, the specific integrations that make switching costly for a customer — is also what’s easiest to describe generically in early marketing and disclose in damaging detail too soon to a competitor posing as a buyer. Early materials should describe outcomes and architecture at a level that lets a genuine buyer evaluate the opportunity without handing a lookalike competitor a blueprint, with technical specifics reserved for after a signed non-disclosure agreement, and prompt libraries or fine-tuning data reserved further still, for exclusivity or post-letter-of-intent stages of the deal.

What the buyer will ask for

Expect requests for the actual data-processing agreements in force with customers, not a summary of them; the foundation-model vendor’s current terms of service and any change-of-control or resale restrictions in them; source data behind any containment or resolution-rate claims, broken out by customer rather than blended into one figure; and documentation of the fallback and human-handoff design. A buyer’s technical diligence will typically also probe how much of the platform’s logic sits above the underlying foundation model versus how much is essentially default behaviour from the API itself — a question that goes directly to whether the buyer is acquiring real technology or a well-marketed integration wrapped around someone else’s product.

What commonly delays a conversational AI platform sale

The most frequent hold-up is a foundation-model dependency question a seller hadn’t fully mapped out — discovering, once a buyer’s counsel reads the vendor agreement closely, that a change-of-control clause requires the vendor’s consent before the business can transfer, or that the vendor’s terms restrict the exact resale model the business actually runs on. A close second is a data-consent gap: conversation logs used for training without a clear contractual basis to do so, found during diligence rather than resolved before listing began. Sellers who confirm both — vendor consent and training-data consent — before going to market close faster and avoid the re-pricing that follows when a buyer discovers either one first.