Buying a Speech and Transcription Business in Canada

Evaluating a speech-to-text or voice-analytics business means looking past the demo and into what actually differentiates it from a dozen other companies wrapping the same off-the-shelf speech model. The demo will sound similar across most of the field, because the underlying speech-recognition technology is widely available. What varies enormously — and what determines whether this is a defensible business or a reseller with a nice interface — is what the company has built on top of that technology, who it sells to, and how carefully it has handled the recorded voices behind the product.

What a good one looks like

• Domain-tuned acoustic or language models with a measurable accuracy edge on the specialized vocabulary its customers actually use — medical, legal, financial or another niche it has genuinely built for.
• Enterprise contracts that renew on their own terms with call centres, healthcare providers or legal customers, rather than a string of one-off projects.
• Genuine bilingual French-English capability, built deliberately rather than bolted on.
• A documented, defensible consent chain for every recording that went into building and training the product, not a policy statement with no records behind it.

What a bad one looks like

• A thin interface layered on top of a generic third-party transcription API, with no proprietary tuning underneath it.
• Accuracy claims that have never been independently benchmarked against anything but the company’s own marketing.
• Heavy reliance on one or two customers, or on a single vendor relationship that could change price or availability without warning.
• Unclear or undocumented handling of recorded audio, including how long voiceprints are actually kept versus what the policy says.

What sellers often don’t volunteer

The true inference cost per audio-minute rarely appears in a pitch deck, and it is the number that actually determines margin once volume grows rather than the headline accuracy figure. The real gap between marketed accuracy and independently tested accuracy is another number sellers are not eager to surface unprompted. And the honest answer to how long voiceprints are actually retained, as opposed to what the written policy claims, sometimes only comes out once a buyer’s technical team starts asking about the database itself rather than the documentation describing it. None of these are necessarily dealbreakers on their own, but each one changes what the business is actually worth and how much work remains before it is fully compliant.

What you personally have to be ready for

There is no college, professional order or provincial licence gating who can own a speech or transcription business, unlike a regulated health practice or a liquor-licensed premises. What buying one does require is stepping into the compliance role PIPEDA — and, for Quebec-facing operations, Law 25 — assigns to whoever controls the personal and biometric information the business holds. That is a real, ongoing responsibility, not a one-time closing condition, and it sits with the new owner from the day the deal closes regardless of what the seller did or did not document beforehand. It also means having, or quickly building, the technical capacity to actually operate and maintain acoustic and language models day to day, not just the commercial relationships that currently sell them.

Sizing up the customer base

A book of business leaning on one or two enterprise accounts carries a materially different risk than one spread across a sector, because losing a single anchor customer in a concentrated book can remove a large share of revenue overnight. Ask how each major contract was won, whether it survives a change of ownership without the customer’s active consent, and how the company would replace that revenue if the relationship ended — the answer says as much about the business’s real durability as any figure on its income statement.

How deeply the product is wired into a customer’s systems

A speech or transcription product rarely stands alone — it typically plugs into a customer’s telephony platform, contact-centre software or practice-management system to capture audio and return results automatically, and that integration is worth evaluating on its own rather than assuming it simply comes with the deal as described. Ask for the actual integration agreements with each telephony or CRM partner the product connects to, not just a feature list, and check whether those agreements carry their own change-of-control or assignment clauses that could require a partner’s consent before the connection survives a sale. A deep, well-documented integration is genuinely valuable: it raises the cost for a customer to switch to a competitor and is part of what makes an enterprise contract renew on its own terms rather than being re-won every year. But an integration built years ago against an API version the partner no longer actively supports is closer to a liability than an asset, because it can fail without warning and the fix may not be within the buyer’s control. Ask specifically which integrations are current, which are legacy, and who is responsible for maintaining each one after closing.