Canada EMR and Clinical AI Add-On Market Report, Size, Share and Forecast 2026-2033

Canada's EMR Base Is Full; the Growth Has Moved to the Layer Above It

Canada does not have an EMR adoption problem. It has a replacement problem that nobody wants to pay for, and a much more interesting problem sitting on top of it.

By 2024, 95 per cent of Canadian physicians were using electronic records to enter and retrieve clinical notes — up from 87 per cent in 2021, 82 per cent in 2017 and 39 per cent in 2010. Four in five no longer use paper at all. Sixty-five per cent run exactly one system. That is what a finished market looks like: there is no greenfield left to sell into.

The only organic growth left in the core layer is the growth in the number of clinicians, and that is barely there. CIHI counted 99,555 physicians in 2024, or 241 per 100,000 population — down from 243 the year before. Family physicians per 100,000 fell for a second consecutive year, from 124 in 2022 to 120 in 2023 to 119 in 2024, and for the first time since the mid-1990s the supply of family physicians grew more slowly than the population. The seat count is not going to carry this market.

So this report measures something narrower and more useful than "EMR adoption". It measures the annual software spend of roughly 67,800 community physician seats in Canada, split into two layers: a core platform layer that is saturated, repriced annually and growing slowly, and an AI add-on layer that was a rounding error before 2025 and is now the only part of the market carrying a genuine growth question.

The AI layer is 11.2 per cent of the market in 2025 and 28.5 per cent of it in 2033. That is what "spending is shifting to AI add-ons" means in Canadian dollars: not that the market doubles, but that between a quarter and a third of a slow-growing market changes composition inside eight years.

The 2025 AI layer was bought by the state, not by doctors

This is the fact that governs everything downstream, and it is easy to miss because the adoption numbers look organic. They are not. Infoway's AI Scribe Program launched in June 2025 and supplied fully funded one-year licences for pre-qualified tools to eligible primary care clinicians. Clinicians had to sign a vendor agreement no later than 31 December 2025. The program portal is explicit about the limits: funding covers "specific primary care providers and baseline solution functionality" and applies for twelve months, and "any use of more advanced functionality, or use beyond the time period set out above, will be at your cost."

By February 2026 the program had enrolled 12,019 primary care clinicians — 24.5 per cent of the country's roughly 49,150 family physicians — of whom 7,400 were active users, defined as ten or more uses a month. Monthly AI-assisted encounters rose fivefold, from 244,000 to 1.22 million.

None of that revenue came from a clinician's own wallet. Which means the Canadian AI add-on market enters its first real test in 2026–27, when those twelve-month licences lapse and 12,019 clinicians have to decide whether to keep paying themselves.

And the auditor published a hallucination rate in the same quarter

In June 2026 Infoway released the six-month evaluation, run independently by the Centre for Digital Health Evaluation at Women's College Hospital. The experience results are strong: 1.11 hours a week saved in work hours and nearly an hour more after hours, same-day note completion up from 55.9 to nearly 75 per cent, delayed documentation down from 9.8 to under 3 per cent, more than 90 per cent of clinicians agreeing the tools added value.

In May 2026, Ontario's Auditor General reported on a provincial procurement of AI scribes. Of the 20 systems tested, nine hallucinated — fabricating information or proposing referrals and blood tests the doctor never ordered. Twelve transcribed a different drug than the one prescribed. Seventeen missed key details about patients' mental health. Some vendors did not submit third-party audit reports, certifications or threat risk assessments, and four systems were approved anyway.

Those two documents are not in conflict. They measure different things: whether clinicians like the tool, and whether the tool is correct. The market's next phase is not an adoption curve. It is an assurance curve, and the incremental spend accrues to whoever can sell verification rather than convenience.

The add-on is becoming a feature

The single most consequential competitive fact in this market is on a vendor's own home page. Accuro — the second-largest EMR in Ontario, with 7,784 community seats — now ships "Accuro AI Scribe, powered by Heidi," integrated inside the EMR.

That is the whole value-chain argument compressed into one product line. When the scribe is a menu item inside the platform the clinician already uses, the platform owner sets the price, captures the attach, and the standalone scribe vendor becomes an engine behind someone else's brand. The 19.1 per cent growth in the AI layer is real; the question this report answers is who books it.

What American physicians did without a programme

Canada's 12,019 funded seats read as an adoption story. Set them against the only comparable national survey and they read as a procurement story. The American Medical Association's 2026 Physician Survey on Augmented Intelligence, fielded in January and February 2026 across 1,692 physicians, found 72 per cent incorporating at least one AI use case into practice — up from 48 per cent in 2024 and 38 per cent in 2023 — with no national licensing programme, no pre-qualified vendor list and no public purse paying for the first year. The share reporting no AI use at all fell from 34 per cent to 19 per cent in two years, and the average number of use cases per physician doubled from 1.1 to 2.3.

Two things follow. First, the Canadian figure measures how fast a funder can put a tool in front of a clinician, not how fast clinicians take one up. 24.5 per cent of family physicians inside eight months is a procurement achievement; 72 per cent inside three years without a programme is closer to an adoption ceiling, and it is the direction Canada moves once the tool stops being free.

Second, and more useful to anyone selling into this market, the AMA's barrier ranking is the same argument this report makes from the auditor's findings, reached from the opposite direction. Asked what would facilitate adoption, 88 per cent named validation of safety and efficacy by a trusted entity and 86 per cent named data privacy assurance; asked to rank regulatory actions, physicians put clear liability frameworks first. Documentation is simultaneously the use case they are most enthusiastic about and among the least used — 57 to 58 per cent enthusiasm against 28 per cent current use. The demand is not for a better note. It is for a defensible one.

What this report is not

It is not a global EHR forecast with a Canada row attached. Most "Canada EHR" sizing in circulation measures enterprise and hospital deployment with implementation services and hardware inside the number. This report measures physician-office software — the thing a family practice actually pays for — in Canadian dollars, because that is the currency the spend occurs in. It also does not present two independent routes to the same figure. There is one construction, a stated band of 4.6 to 7.5 per cent, and an explicit list of what would falsify it, in the market risk chapter.

Table of Contents

1. Canada’s EMR Base Is Full; the Growth Has Moved to the Layer Above It
2. Four Vendors Hold 91 Per Cent of Ontario Seats, and None of Them Owns the Add-On
3. The Add-On Is Priced by the Clinician, Sold by the Province, and Delivered Behind a Brand
4. The 2026 Renewal Is the Whole Investment Case
5. The Scribe Is Solved Enough to Sell and Not Solved Enough to Trust
6. The Auditor Found What the Satisfaction Survey Could Not
7. Nobody Certifies the Scribe; Procurement Is Doing the Regulating

Competitive Landscape

Four Vendors Hold 91 Per Cent of Ontario Seats, and None of Them Owns the Add-On

Canada has no national EMR market. It has thirteen provincial ones, and public, vendor-reported seat counts now exist for two of them. OntarioMD publishes what vendors self-report monthly, and the August 2026 numbers are the best available picture of who holds the clinician relationship in Canadian primary care.

Three vendor groups account for 24,541 of those 27,101 seats — 90.6 per cent. TELUS Health holds 39.0 per cent across three offerings, QHR’s Accuro 28.7 per cent, and the WELLSTAR group 22.9 per cent. The long tail is 1,230 seats shared among four vendors, the largest of which has 145.

The structure matters more than the ranking, for three reasons.

First, the deployment mix is finished. 24,793 of 27,101 seats — 91.5 per cent — are hosted rather than local. The cloud migration that drove the last decade of Canadian EMR revenue is essentially complete. There is no remaining installed base to convert, which removes the industry’s most reliable repricing lever.

Second, the consolidation is already done. The OSCAR-heritage base was assembled by acquisition, not growth: WELL Health bought Juno EMR and ClinicAid from CloudMD in 2022 for approximately CAD 5.75 million, describing itself in the release as the largest provider of OSCAR 1 products and services. There is no fourth column left to acquire.

Third — and this is the reason the add-on is a different contest — the only genuine share movement in the table is a cloud-native entrant. Avaros holds 1,330 Ontario seats, 4.9 per cent of the market, and is the one vendor in the table with zero local deployments. It got there without buying an installed base. In a market where the top three are flat to declining in relative share, that is the only organic growth line in the census.

The second census, and the eviction dated inside it

Ontario is no longer the only public seat count. When New Brunswick moved from a single provincial EMR to an open market in 2021 it kept the certification pipeline, and its provincial EMR programme now reports vendor-level subscriber counts through the same OntarioMD release as Ontario does.

At 888 seats against Ontario’s 27,101, New Brunswick is too small to move any national figure in this report, and it does not. It is useful for two other reasons.

The first is that incumbency survived an open market. From 2012 to 2021 the province ran a single EMR — Intrahealth’s Profile, delivered by Velante, a subsidiary of the New Brunswick Medical Society. Five years after the market was opened to certified competitors, that same product still holds 684 of 888 seats, 77.0 per cent. Opening a market is not the same as contesting one.

The second is that a certification requirement, not a price, is what removes an installed base — and the date is published. TELUS PS Suite is the largest single EMR in Ontario with 9,254 seats. In New Brunswick its certificate was suspended on 31 March 2025, because TELUS told the province it did not intend to implement the electronic laboratory results specification introduced in June 2023. The suspension runs a 24-month transition, after which certification is withdrawn on 31 March 2027; funding incentives and access to provincial integrations for those clinicians end on the same date.

The revenue at stake is 32 seats, which is nothing. The mechanism is not nothing. It is the only dated, involuntary displacement visible in Canadian primary-care EMR documents, it runs on interoperability compliance rather than discounting, and 31 March 2027 falls inside the same renewal window as the funded scribe licences this report’s investment chapter is built around. A vendor that will decline one province’s specification can decline another’s, and the province that says so puts a date on it.

The paradox in the add-on layer

You would expect the Scribe market to be contested by AI specialists. In Canada it is being absorbed by the seat owners instead, and the clearest evidence is a product line rather than a press release.

Accuro — 7,784 seats, the second-largest base in the country — now ships Accuro AI Scribe, powered by Heidi, integrated directly inside the EMR, alongside dictation and transcription tools in its own clinic tools library.

That is the competitive structure of this market in one product. The specialist supplies the engine; the platform owner supplies the distribution, the integration, the billing relationship and the brand. The clinician does not choose a scribe and then connect it — they choose an EMR and receive one. The global EMR platform is the distribution channel for the add-on, not its competitor.

Which means the correct competitive question in this market is not “which scribe wins”. It is “who already owns the seat”, and the answer is a four-line table.

What the incumbents are structurally bad at

Integration is a moat, but it is a moat around the wrong castle. Nothing in the platform owners’ position helps them with the thing buyers started demanding in 2026: evidence that the note is correct. An EMR vendor can ship a scribe in a quarter. It cannot conjure a validated quality regime, third-party audit reports and enforced clinical attestation out of its integration layer — and on the evidence in the market risk chapter, at least some vendors have not been producing those artefacts even when procurement asked for them.

That is where a specialist can still win share without winning seats: not by selling documentation, which is becoming a feature, but by selling the verification, audit trail and liability cover that makes documentation defensible. The competitive boundary is moving from the note to the proof of the note.

Value Chain

The Add-On Is Priced by the Clinician, Sold by the Province, and Delivered Behind a Brand

The Canadian EMR value chain has an unusual shape, and it explains both why the add-on layer grew so fast and why it is now exposed. Three parties touch every dollar, and in 2025 two of them were not the customer.

Layer 1: the seat, which the province has historically paid for

Canada does not have a private-practice software market in the American sense. It has a publicly subsidised one. The clearest illustration is Ontario’s own programme history: the EMR Adoption Program funded community physicians up to CAD 27,100 for a local EMR or CAD 29,800 for an ASP offering, and layered a CAD 400-per-month Enhanced Use subsidy on top to keep the software current and connected to provincial laboratory and hospital-report systems.

That CAD 400 a month — CAD 4,800 a year — is the most useful public anchor for the core layer’s price. It is roughly the amount a provincial payer has been willing to spend per physician per year to keep an EMR seat current. Our core-layer construction of CAD 2,800 per seat sits deliberately below it, because the subsidy bundles connectivity and hospital-report integration that a software-only measure should not claim.

Layer 2: the scribe engine, priced per clinician and indistinguishable on features

The add-on layer prices very differently: monthly, per clinician, at retail. Tali AI’s published Canadian pricing puts dictation at CA$50 to CA$68 a month and the unlimited AI scribe tier at CA$100 to CA$150 a month depending on term and promotion, with enterprise pricing quoted separately.

That is a list price of roughly CAD 1,200 to 1,800 per clinician per year, and it is the basis for the CAD 1,500 per attached seat used in this report’s model. Note what the price is not based on: output quality, accuracy, or liability. Every vendor in the market prices the same three bundles — dictation, unlimited scribing, enterprise — because the underlying speech and summarisation capability is now a bought-in commodity rather than a differentiator.

Layer 3: the funder, which in 2025 was the state

Here is the structural break. The Infoway AI Scribe Program did not subsidise a purchase; it replaced it. Fully funded one-year licences, allocated through a national pre-qualification of vendors, with provinces running implementation and clinicians choosing from the qualified list. The portal terms cap the obligation precisely: baseline functionality only, twelve months only, everything else at the clinician’s cost.

So the 2025 value chain ran: the funder selected the vendors, the platform delivered the integration, the clinician consumed the tool, and money moved from public purse to vendor without passing through the practice. That is an efficient way to build an installed base quickly and a poor way to prove a market exists.

The economics that decide the renewal

The renewal case rests on one ratio, and it is strong. On the six-month evaluation, clinicians reported saving 1.11 hours a week during work hours — setting aside the near-hour saved after hours, because after-hours time is not paid time in a fee-for-service system and should not be valued at a billing rate.

Note on the hour value: CIHI reports average gross clinical payment per physician of CAD 383,000 in 2023–24, up 3.9 per cent. Dividing by an assumed 1,600 fee-generating hours gives CAD 239.38. This is an upper bound — gross payment includes practice overhead the physician does not keep — and it is a system-value figure, not a take-home one. The ratio is presented so a reader can substitute their own divisor.

An 8.15x return ought to make renewal trivial. It does not, for a reason that has nothing to do with arithmetic: the clinician who renews is not the party that captured the CAD 12,223. The evaluation found that half of clinicians redirected the saved time into other administrative work and only a fifth converted it into additional appointments. In a fee-for-service practice, an hour of paperwork moved to 6 p.m. is a personal benefit and a thin financial one. In a capitated or salaried practice it is not a revenue event at all.

Which is why the value chain’s weakest link is the one between the party that benefits and the party that pays — and why the vendors who will win the renewal are the ones who can bill the saving through the practice’s own revenue cycle rather than asking the clinician to fund a lifestyle improvement.

Where the 8x actually lands

If the saving is real and the clinician will not pay for it out of pocket, the value migrates to whoever already holds the billing relationship — the platform. That is the economic logic behind Heidi sitting inside Accuro: the platform can bundle the add-on into the seat price, capture the attach, and pass a clinician a single invoice for a capability they have already experienced for a year. The specialist’s engine becomes a component; the platform books the margin.

This is also why the healthcare revenue cycle boundary is the one to watch. Documentation, coding and billing are three steps in one workflow, and the first of the three is the one that just became free to try. Whoever joins the scribe to the claim captures the CAD 12,223 instead of reselling software for CAD 1,500.

There are two funders now, and the second one negotiates

Everything above describes a value chain with one funder: Infoway, allocating twelve-month licences through a national pre-qualification. That was the 2025 structure. It is not the 2026 one. Ontario now runs its own AI Scribe Program, established by the Ministry of Health and Ontario Health as part of the Primary Care Action Plan and aligned with Ontario Health’s Patients Before Paperwork initiative — and it buys differently. It maintains a Vendor of Record list managed by Supply Ontario, restricted to vendors that have met provincial requirements for clinical functions, privacy and secure data storage, and offers them to clinicians at prices discounted from current market prices.

The difference between a grant and a vendor-of-record list is not the money. A grant sets a quantity and expires. A list sets a price and keeps setting it: every new entrant has to clear the same clinical, privacy and storage requirements, and every clinician buying through the programme buys at the negotiated rate. A programme written into a plan to connect every Ontarian to a family doctor by 2029 is a procurement apparatus that outlives any individual funding envelope.

For this report’s model the consequence is specific, and it runs against the price assumption. The renewal has been treated as a binary — list price or nothing — and the price path was built upward, from CAD 1,500 to CAD 1,800, on the argument that scope expands faster than scribing commoditises. A standing provincial list at a published discount is the mechanism by which the opposite happens, and it is the only price-setting channel in this market that exists today rather than in forecast. If Ontario’s discounted rate becomes the reference price for the provinces that follow, then for the publicly bought share of this market the price variable compounds down, not up.

Which is a limit of the band rather than a refinement of it. The low case prices a 2033 seat at CAD 1,650 and the base case at CAD 1,800, both above today’s CAD 1,500 list. A vendor-of-record Canadian price would sit below even the low case, and nothing in this model produces that outcome. A reader who expects public purchasing to set the price here should treat 4.6 per cent as an upper bound rather than a floor.

The smallest province, and the two rules inside it

Nova Scotia’s AI Scribe Pilot is funded by Infoway and administered by the provincial Department of Health and Wellness, and it has named a single vendor: Mikata Health, whose Mika AI Scribe integrates with provincial EMRs including Med Access. Two lines in its eligibility rules are worth more than its seat count.

The first: to take part, a clinician must agree to use the scribe at least ten times a month. That is the same threshold the national evaluation used to define an active user — the 7,400 of 12,019 this report’s renewal scenarios lean on. The measurement has become a condition. Which means the habitual-use rate is not a passive statistic the market happened to produce; it is being written into programme terms, and provinces will be able to report conversion on the same definition.

The second: clinicians who are not allocated a funded licence are offered preferred pricing on the same product. That is the mechanism by which a subsidised cohort becomes a commercial funnel, and it is the cleanest available answer to the question the renewal poses. The pipeline does not end when the licence does. It converts into a discount relationship with the vendor the province already validated.

The envelope that ran out before the clinicians did

One figure from 2025 deserves restating, because it is the best evidence in the public record about what happens after a funded year. Infoway’s launch announcement in June 2025 offered up to 10,000 fully funded, one-year licences. By February 2026, 12,019 clinicians had enrolled — 120.2 per cent of the announced envelope, or just over a fifth more licences than the programme said it had.

Either the envelope was expanded or provinces added licences on top of it; the published record does not say which. Either way the investment chapter’s observation stands and its implication shifts. No 2027 envelope has been announced, and that remains true. But the strongest argument that a second one appears is not an inference about politics — it is that the first one exhausted its allocations before it exhausted its demand, by a margin large enough that somebody had to find the difference. A funder that overruns its own envelope by a fifth has demonstrated both the appetite and the administrative capacity to run the programme again.

Investment Activity

The 2026 Renewal Is the Whole Investment Case

There is one number that decides whether this market’s 19.1 per cent add-on growth is real, and it is not a forecast. It is 12,019 — the primary care clinicians enrolled in the national AI Scribe Program between June 2025 and February 2026.

The re-pricing event

Every one of those licences was fully funded for twelve months, with agreements required by 31 December 2025. On this report’s CAD 1,500 per-seat assumption, the funded cohort represents roughly CAD 18.0 million of annualised add-on revenue that has no contractual reason to exist after its first year.

Now run the conversion. The evaluation’s most useful operational figure is not satisfaction — it is active use: 7,400 of 12,019 clinicians, 61.6 per cent, were using a scribe ten or more times a month by February 2026. Habitual use is the best available proxy for renewal intent, and it is the only number in the public record that lets anyone model this.

The spread between those rows — CAD 5.6 million to CAD 11.1 million — is roughly a quarter of the entire 2025 add-on layer hanging on a decision that no clinician has yet had to make with their own money. This is why the market’s growth is best described as forecast rather than observed.

What capital has actually been willing to pay

Canada has produced very few transactions that price a clinician seat, which makes the ones that exist unusually informative. WELL Health’s acquisition of Juno EMR and ClinicAid from CloudMD closed at approximately CAD 5.75 million for assets serving more than 2,500 practitioners and contributing more than CAD 9 million of topline.

Work that through and you get roughly CAD 3,600 of revenue per practitioner and CAD 2,300 of purchase price per practitioner — a revenue multiple of about 0.64x. That is the market’s own valuation of an EMR seat bundle: low, because the revenue is recurring but barely growing, and because the buyer is principally acquiring a base to which other products can be attached rather than a growth engine in its own right.

That framing is the investment case in miniature. Nobody buys Canadian EMR seats for the subscription. They buy the distribution layer and then attach things to it — which is precisely what WELL said at the time, describing the acquisition as giving users access to its Apps.Health platform of third-party tools including an AI-powered assistant.

Where the strategic money went early

The AI attach was anticipated years before it was funded. WELL took a minority position in Tali.ai in April 2022 through WELL Ventures and signed an alliance to commercialise Tali’s technology across its EMR portfolio, at a time when its network covered more than 15,000 doctors and other providers in over 2,800 clinics.

Read that release again with 2026 hindsight and the strategy is legible: own the seat, invest in the engine, control the marketplace through which the add-on reaches the clinician. The most recent iteration — a scribe built into the EMR under the platform’s brand — is that strategy executed rather than announced.

What is conspicuously not happening

There is no consolidation wave in the add-on layer. No Canadian platform has acquired a scribe outright at a disclosed price; the integrations are commercial rather than structural. Two explanations are possible and they lead to opposite forecasts. Either the platforms are waiting to see whether the funded cohort renews before committing capital — which would imply the integration announcements are options, not commitments — or the scribe has already been priced as a commodity component whose supplier does not need to be owned.

The evidence available favours the second: the deals are structured as integrations and marketplace listings, vendors publish interchangeable three-tier price lists, and the engine brand is subordinate in the product name. But this is an inference from absence, and it is the weakest link in this report’s investment argument. A single disclosed acquisition of a Canadian scribe vendor would settle it.

The path the 19.1 per cent has to be true of

An endpoint CAGR is a slope, not a forecast. Compounding each layer out of its own 2025 base gives the annual path, and the shape of that path is the part a practice, a vendor or a buyer can actually plan against.

The layer columns compound at 3.13 and 19.07 per cent — the exact rates implied by each pair of endpoints — and the total at 5.96 per cent. Each column is rounded independently, so the layers will not sum to the total in every mid-year, and the 2025 components (189.8 and 24.0) round to 213.8 against the published 213.9. The endpoints are the published figures; the mid-years are derived.

Read the columns against each other. The core layer adds CAD 53.0 million of revenue across eight years — an entire country’s primary-care software market growing by a little over six and a half million a year. The add-on layer adds CAD 73.0 million in the same window from a base an eighth the size. One of those is an annuity with a renewal cycle attached; the other is the growth story, and it depends entirely on an attach decision taken practice by practice after the funded licences lapse.

The composition also moves later than the headline implies. The add-on layer is 11.2 per cent of spend in 2025, 18.3 per cent in 2029, 20.6 per cent in 2030 and 28.5 per cent in 2033. Anyone buying for the shift should notice that it becomes visible in the total only after the renewal question is already answered.

The first time anyone marks the distribution layer

Until 2026, every valuation of a Canadian EMR distribution layer was private and therefore arguable — the Juno and ClinicAid transaction priced at roughly 0.64x revenue, above, being the only real datapoint. WELL Health’s second-quarter 2026 disclosure puts a public market around that layer for the first time: WELLSTAR, which owns the OSCAR Professional and Juno EMR bases plus billing and connectivity, reported annual recurring revenue of CAD 80.1 million, up 37 per cent from CAD 58.4 million a year earlier, which was itself up 45 per cent from CAD 40.2 million in Q2-2024. SaaS segment revenue, which includes CYBERWELL, was CAD 25.1 million in the quarter, up 29 per cent. The technology supports more than 45,000 unique providers.

The structural event is the listing. WELLSTAR is being separated onto the TSX Venture Exchange through a reverse takeover with a concurrent CAD 50 million subscription-receipt financing — CAD 36.2 million treasury and CAD 13.8 million secondary — against roughly CAD 95 million of expected 2026 revenue, and the stated uses of proceeds include AI-driven product innovation. WELL remains a majority, long-term controlling shareholder and, in its own words, a growing customer of the entity after the listing.

Two readings, and the report needs both. The multiple on offer is a growth-software multiple applied to a business whose recurring revenue is growing 37 per cent — a re-rating no Canadian EMR seat has ever received while sitting inside a clinic-and-services group. The scale check is a caution: CAD 80.1 million of ARR against this report’s CAD 189.8 million core layer is 42.2 per cent, but WELLSTAR’s ARR spans US billing and connectivity as well as Canadian EMR subscriptions, so that is an upper bound on its Canadian share, not an estimate of it.

What the spin-out does settle is the logic the value-chain chapter argues from the other end. The distribution layer is worth more to its owner as a separately listed vehicle than as a division, which is exactly why attaching a 19 per cent-growth product to a 3 per cent-growth seat base is rational even when the attachment is powered by someone else’s engine. It weakens — it does not close — the inference-from-absence weakness flagged above: capital is now being raised against the assumption that the attach works.

The capital the funder has not committed

The one thing that would de-risk the entire layer is a second funding envelope, and there isn’t one. Infoway’s published conclusion is deliberately open: the evaluation will “help inform continued work with jurisdictions, vendors and health system partners”, and clinicians themselves identified “the need for flexible funding models to support equitable access across practice settings.”

That is a funder describing a gap, not closing it. Until a 2027 envelope is announced, the honest base case for this market is conversion at market price, and the honest bull case is a second public programme that turns another cohort of clinicians into free-year users. Only one of those is in the forecast.

The three variables inside the 19.1 per cent

An add-on CAGR is one number standing over three different things, and separating them shows what the forecast is actually a bet on. Add-on revenue is seats multiplied by attach multiplied by price per attached seat, and each variable compounds at its own rate between 2025 and 2033.

The three rates multiply to 1.1908 — 19.08 per cent — reconciling with the 19.07 implied by the published CAD 24.0 million and CAD 97.0 million endpoints. The difference is rounding in the opening attach share: 24.0 divided by (67,800 × CAD 1,500) is 23.60 per cent.

Read the columns rather than the total. More doctors contribute 1.6 points, a modest price rise contributes 2.3, and attach contributes 14.56 — 76.3 per cent of the growth in this market is clinicians deciding to keep using a tool they were given for free. This is not a demographic forecast or a pricing forecast wearing a growth label; it is a behaviour forecast. Which is why the 12,019 renewal matters more than any other variable in the model, and why a reader who doubts the behaviour should not use the base case at all.

The same decomposition gives the unit path, which no table in this report has shown until now:

Attached seats are revenue divided by the stated price at each year, so the mid-year row carries the price-path assumption explicitly rather than hiding it. Attach rises through the renewal window: the model’s own compounding puts it at 40.6 per cent by 2029. Attached seats multiply 3.37 times while the underlying seat base grows 13.5 per cent.

Which makes the exposure asymmetric in a way the band does not show. A price disappointment is bounded: a fifth off the 2033 price takes CAD 19.4 million off the layer. An attach disappointment is not. Falling ten points short of 70 per cent costs 14.3 per cent of the 2033 layer, and landing on the low case’s 50 per cent costs 28.6 per cent of it. The forecast’s risk sits in one variable — and it is the one the market will shortly report on directly, because at least two provinces are now collecting habitual-use data on the same ten-uses-a-month definition this model is built from.

Technology & Innovation

The Scribe Is Solved Enough to Sell and Not Solved Enough to Trust

The technology question in this market is not whether a machine can write a clinical note. It is what happens to the error rate when the note stops being a document and becomes an action.

What is now reliable

The measurement is unusually good, because the national programme was independently evaluated rather than vendor-reported. On the six-month results from the Centre for Digital Health Evaluation at Women’s College Hospital, the operational gains are consistent and large.

Two of these deserve more weight than they usually get. The timeliness shift — same-day completion up 19 points, delayed documentation cut by more than two thirds — is the change that actually matters to a health system, because a note written the same day is a note that can be used for care rather than for record. And the completeness gain, taken from a retrospective chart review rather than clinician self-report, is the only measure in the public record that tests the note against the chart instead of against the clinician’s impression.

Gains appeared as early as three months. That matters commercially: it means the value is demonstrable inside a one-year funded licence, which is precisely the window a renewal decision falls in.

What is not reliable, measured in the same quarter

Against that, the Ontario Auditor General’s procurement review tested 20 systems and found a defect pattern that no satisfaction survey would surface.

Both documents are true. They are measuring different objects: whether the clinician’s working life improved, and whether the artefact is correct. A tool can deliver four hours a month and still write down the wrong drug.

The technical claim worth making

The failure modes above are not evenly dangerous, and the gradient points at the layer’s next product cycle.

A hallucination inside a narrative note is a documentation defect. It propagates to the chart, to the next clinician, and to the billing record — bad, and recoverable by a physician who reads what they sign. A hallucination that proposes a referral or orders a blood test the physician did not request is a different class of event: it crosses from text into a clinical action, and the physician’s review is no longer a proofread but a veto of something that has already been generated.

That distinction is already shaping the roadmap, and the clinicians in the evaluation said so explicitly. Asked what the next generation should do, they named referral drafting, inbox management and clinical decision support — three functions that all sit further along the action gradient than transcription. The market is walking directly toward the failure mode that just scored 45 per cent.

Where the engineering constraint actually sits

It is not in speech recognition, which is commoditised, and it is not in summarisation, which every vendor in the market performs at a similar level. It sits in context. The 2024 national survey found 73 per cent of physicians citing poor system integration or multiple unconnected systems as a major barrier, and 68 per cent spending an hour or more a day looking for patient information they need.

A scribe that hears only the room is working from the thinnest possible context. It cannot know that the patient’s metformin was changed last week, because that fact lives in a hospital report that never reached the EMR — which is the same interoperability failure the survey identified, now inherited by the AI layer. The accuracy ceiling on this technology in Canada is set less by model quality than by how much of the patient’s record the model can see.

That is also why the integration ownership question is a technology question and not just a commercial one. The party that controls the record controls the context, and context is what separates a note that reads well from a note that is right.

The assurance layer is the real product

Which brings the technology chapter to the same conclusion as the commercial one. The next defensible thing to build is not a better scribe. It is the apparatus that makes a scribe’s output verifiable: confident extraction of every clinical assertion, an audit trail that distinguishes what the physician said from what the model generated, and an enforced attestation that a human read it. The general AI stack supplies none of that, because it is domain-specific and unglamorous — and it is what a buyer with a 45 per cent hallucination rate in the public record will actually pay for.

What the burden actually was, before any of this

The gains in this chapter are large against a baseline that almost nobody in this market states clearly. OntarioMD’s own summary of the problem, citing a 2023 Ontario College of Family Physicians survey, is that primary care physicians in Ontario spend an average of 19 hours a week — 40 per cent of their time — on administrative tasks. That is the pool a scribe is drawing from, and two other numbers in circulation are worth holding against it.

Set the two savings figures against each other and the difference is a factor of about three: 3.15 times as much time returned in the smaller, self-reported pilot as in the national independent evaluation. Both numbers are quoted in this market, usually interchangeably, and they should not be. The defensible reading is that self-reported time savings shrink under independent measurement, and 1.11 hours is the figure to model on.

Against a 19-hour baseline, 1.11 hours is 5.8 per cent of the administrative burden removed. That is a real gain to the clinician who receives it and a small number against the problem as officially stated — and it reaches the same conclusion as the engineering argument above. The constraint is context, not capability, so the size of the prize is capped by how much of the patient’s record the tool can see rather than by how good the model gets.

The programme’s own results supply the middle ground between measurement and perception: nearly 70 per cent of clinicians reported reduced administrative burden and almost 80 per cent reported lower cognitive load, more than 80 per cent felt more engaged during patient visits, and 60 per cent perceived an improvement in the quality of care they delivered. Those are experience measures, and they are consistent with 1.11 hours. They also explain the enthusiasm that produced a fifth more enrolments than the envelope allowed: a clinician whose day feels lighter is not misreporting. They are reporting something else.

Market Risk

The Auditor Found What the Satisfaction Survey Could Not

This market has a strong demand case and a weak assurance case, and the gap between them is the risk. Four things could break the forecast, ranked by how much of it they carry.

Risk one: the funded cohort does not convert

This is the largest single exposure and it is arithmetic, not judgement. The 2025 add-on layer is CAD 24.0 million, of which roughly CAD 18.0 million is the national programme’s funded licences. Twelve months, baseline functionality only, everything beyond that at the clinician’s cost.

The best available predictor of conversion is habitual use: 7,400 of 12,019, or 61.6 per cent, were using a scribe ten or more times a month by February 2026. If that ratio holds at renewal and everyone pays list, the layer retains about CAD 11.1 million of the CAD 18.0 million. If platforms bundle it into the seat price at a discount, less again. A severe case — where clinicians who liked a free tool decline a paid one — takes the 2026 add-on layer to CAD 15 million and pushes the 2033 endpoint toward the bottom of the band.

The evaluation’s own finding is the reason to take this seriously rather than assume loyalty. Half of clinicians redirected saved time into other administrative work and only a fifth into additional appointments. A clinician who experienced the benefit as slightly less paperwork at 6 p.m. is not an obvious candidate to start paying CAD 1,500 a year for it.

Risk two: the platform absorbs the price

The add-on is being converted from a product into a feature, and features get priced at the margin. Accuro ships Heidi inside the EMR; a platform that owns 7,784 seats does not need to charge list for a capability it can bundle. If bundling becomes the norm across the 90.6 per cent of Ontario seats held by three vendor groups, the addressable revenue per attached seat falls well below the CAD 1,500 assumption even as attach rises.

Note the shape of this risk: it does not reduce the number of clinicians using AI. It reduces the revenue per user while increasing the users. Adoption and market size can move in opposite directions, and a forecast built on attach rates alone would miss it entirely.

Risk three: assurance becomes a cost of goods

The auditor’s findings convert directly into vendor cost. Third-party audit reports, certifications and threat risk assessments — the artefacts some vendors did not submit and four systems were approved without — are not optional once a province has been criticised for accepting their absence. Add validation datasets, clinical safety cases, audit-trail infrastructure and the insurance that follows a documented failure rate, and the gross margin on a CAD 100-a-month scribe compresses sharply.

This is the risk that most affects the 19.1 per cent growth rate, because it attacks the layer exactly where the growth is.

Risk four: the seat base stops growing

The core layer’s 3.1 per cent depends on physician supply, and the direction is not favourable. CIHI’s 2024 data show family physicians per 100,000 falling for a second consecutive year — from 124 in 2022 to 119 in 2024 — and, for the first time since the mid-1990s, family physician supply growing more slowly than the population. Specialists grew 2.5 per cent against family medicine’s 1.9.

Our model assumes 1.6 per cent annual seat growth, which is above the family-medicine trend and below the headline physician supply trend. That is a deliberate choice to avoid both complacency and alarmism, but a reader who believes the per-capita decline is structural rather than cyclical should take the low case.

The forecast and its band

The base case is attach rising from 23.6 per cent to 70 per cent with the price per attached seat rising modestly, from CAD 1,500 to CAD 1,800, on the argument that scope expands faster than scribing commoditises. If scribing commoditises faster than scope expands — the platform-bundling scenario — the price falls instead and the low case is the right one.

Three falsifiers

This forecast should be considered wrong, and rebuilt, if any of the following occurs.

One. A disclosed acquisition of a Canadian AI scribe vendor by a platform owner. That would confirm the component view and push the add-on layer’s revenue into the core line, collapsing the two-layer split that this report’s structure depends on.

Two. Renewal data showing conversion materially below the 61.6 per cent habitual-use rate. If fewer than half of active users re-subscribe at market price, the CAD 24.0 million base is a programme artefact and the 19.1 per cent add-on CAGR has no foundation.

Three. A provincial requirement that AI-generated notes be flagged, attested or separately audited. That converts an optional productivity tool into a compliance obligation with mandated functionality — which would raise price per seat and lower attach, moving both variables in the model at once.

One final honesty note. This report does not run a second construction and present it as corroboration. Any second route available would restate the same two unknowns — how many of Canada’s community physicians run what kind of seat, and what each pays — dressed as an independent series. One construction with a stated band, and the assumptions above it, is worth more than two that share the same guess.

Regulatory Landscape

Nobody Certifies the Scribe; Procurement Is Doing the Regulating

Canada has not built a regulatory regime for AI clinical documentation. It has built a purchasing regime, and in the absence of anything else that purchasing regime is doing the work — which is why the most consequential regulatory fact in this market is not a statute. It is an audit finding about a procurement process.

The de facto regulator

The national programme’s own description of its method is the clearest statement of how this market is governed: vendors were pre-qualified nationally, with requirements that solutions meet data exchange, cybersecurity and usability standards, then provinces ran implementation and clinicians chose among the qualified list. Clinicians were also required to select a vendor and sign an agreement no later than 31 December 2025.

Read that as a regulator would. A single national pre-qualification sets the technical bar that every seller must clear to reach the funded market, and because the funded market was in 2025 very nearly the whole market, the bar became the market’s entry requirement. This is exactly the kind of standards-by-procurement that works well and fails quietly, because there is no statutory duty behind it and no published register of who was refused.

And it failed at least once, visibly

The Ontario Auditor General’s report is the sharpest regulatory evidence in this market, and the finding that matters most is not the hallucination rate. It is that some vendors did not submit third-party audit reports, certifications or threat risk assessments during the procurement — and four systems were approved anyway.

That is a controls failure, not a technology failure. A pre-qualification regime whose own evidence requirements can be waived at approval is not a regulator; it is a screening step, and the auditor has now documented what happens when a screening step is treated as an approval.

The recommendation that follows is the one with commercial consequences: the province should implement IT controls that enforce an attestation from the physician confirming that the notes were reviewed. That single recommendation does more to shape the next product cycle than any privacy statute, because it converts physician review from a practice into a system-enforced event that must be recorded.

The statutes that do bind, indirectly

Privacy law is the binding instrument, and it is provincial in a way that makes national products awkward. Vendors in this market publish compliance across PHIPA in Ontario, PIPEDA federally and the various provincial health-information statutes, alongside jurisdiction-specific regimes. Those claims should be read as vendor assertions rather than audited findings — the auditor’s report is direct evidence that at least some compliance documentation in this market was missing when it was asked for.

The practical regulatory constraint is therefore not a licensing gateway but a fragmentation cost: a product built for Ontario must be re-evidenced for Quebec, re-papered for Alberta, and re-explained to every health authority that asks. That cost falls hardest on the smallest vendors, which is the opposite of what a permissive regime would produce.

The note remains the physician’s problem

Nothing in the evidence reviewed for this report displaces the physician from the record. The clinician signs the note, the college’s record-keeping obligations attach to that signature, and the auditor’s recommendation exists precisely because the province wants that signature to mean something. So the legal position is that the AI generates a draft and the human is accountable for the whole of it.

That is a stable arrangement and a commercially useful one, because it tells you what to sell. A product that makes attestation fast, granular and defensible is helping the physician discharge an obligation they already hold. A product that generates clinical content faster is adding to the pile they are accountable for.

What is not regulated, and why it matters

There is no published national standard for what a clinical documentation model must score to be sold into Canadian primary care, and no register of validated tools. The market’s assurance layer is therefore being built by contracts and evaluations rather than by rules — one national pre-qualification, one independent academic evaluation, one provincial audit, and a great many clinician satisfaction surveys.

That gap is the single most important thing a buyer should understand about this market, and it is also the clearest opening for a seller. The rules that will decide who wins the 2027 renewal are being written now, in procurement documents and auditor recommendations, and the parties writing them are not waiting for a statute.

The device line, and which side a scribe is on

There is a federal regulator for this market after all. It has simply not been asked to regulate a scribe yet, and the reason is a definitional boundary rather than an absence of jurisdiction. Health Canada’s software as a medical device guidance lists software that does not meet the definition of a device at all, and a transcription tool falls inside three of its exclusions: software intended for administrative support of a healthcare facility, software that enables clinical communication and workflow, and software intended to serve as electronic patient records.

A scribe as currently sold is documentation software. That is why no vendor in this market holds a device licence, and why the opening claim of this chapter is accurate for the product on the shelf in 2026.

The boundary is drawn on intended use, and it is closer than the market assumes. Health Canada’s remaining exclusion for clinical decision support requires all four of: that the software not acquire, process or analyse a medical image or signal; that it display, analyse or print medical information; that it only support a clinician’s decision; and that it not be intended to replace clinical judgement, with the user able to review the basis of any recommendation. Where an intended use moves from informing to driving clinical management in a serious or critical situation, the classification table lands the product in Class II or Class III under Rules 10(1) and 10(2), and a licence application follows.

Which puts two facts already in this report on the same side of the line. OntarioMD’s own description of what a scribe does includes ordering tests or prescriptions by voice command, alongside generating referral letters. And the Auditor General found nine of twenty systems fabricating referrals and blood tests the physician never ordered — the same action, generated wrongly. The feature roadmap clinicians asked for in the national evaluation, referral drafting and inbox management and clinical decision support, is the roadmap that crosses Health Canada’s boundary. The first vendor to market any of those as an intended use stops being a documentation company and becomes a device applicant.

The apparatus for that already exists. Health Canada’s pre-market guidance for machine learning-enabled medical devices, in the version published in April 2026, sets out good machine learning practice, clinical validation and post-market monitoring for Class II to IV software, and introduces the predetermined change control plan — the mechanism by which a model is permitted to change after approval. Nothing in this market uses it yet. Everything in the market risk chapter’s assurance-as-a-cost-of-goods scenario would.

One province has already built the certification

The opening of this chapter says nobody certifies the scribe, and that needs one qualification — the most operationally consequential document in the market. Nova Scotia Health’s AI Scribe policy states that Mika AI Scribe is the only validated AI Scribe for Nova Scotia use at this time, that it is the only one approved for use in NSH clinics and by NSH employees, and that it has completed all required privacy, cybersecurity, legal and clinical reviews. Other tools remain usable at the discretion, liability and responsibility of the clinician.

That is a certification regime. It is not statutory, it is not national, and it names exactly one product — which is what makes it matter. A health authority has demonstrated that privacy, cybersecurity, legal and clinical review of a scribe is something a Canadian payer can complete and publish. Once one province has done it, the artefact exists for every other province to copy, and the vendors who cannot produce the evidence now have a named benchmark they fail.

The same policy sets the consent architecture the whole layer runs on, and it distinguishes two products that vendors market as one: a streaming scribe can proceed on knowledgeable implied consent, while a recording scribe requires express consent. Patients may revoke consent at any time and the revocation must be documented. And the accountability sentence is unambiguous — providers are most responsible for their chart in every event, and must always review, edit and finalize the output. That is the position the Auditor General reached from the other direction, written as clinical policy rather than as a recommendation.

The federal layer that never arrived

Why is there no statute? Because the one that was drafted no longer exists. The Artificial Intelligence and Data Act, the AI part of Bill C-27, died in 2025 when Parliament was prorogued, and the replacement industry expected has not come. Canada’s national AI strategy, AI for All, was published in June 2026 across six pillars — protecting Canadians and safeguarding democracy, ensuring AI empowers Canadians, powering adoption for shared prosperity, building a sovereign AI foundation, scaling Canadian champions, and building trusted partnerships and global alliances. Its legislative plan runs through consumer privacy legislation and existing instruments rather than through a single AI statute.

That is the structural fact underneath this whole chapter. A market with no AI statute, no device obligation for its current product and no national standard is governed by the three instruments that do exist: a national vendor pre-qualification that sets the entry requirement, an independent evaluation that sets the evidence bar, and provincial procurement — a vendor-of-record list in Ontario, a single validated product and a written clinical policy in Nova Scotia — that sets both price and permission. None of the three requires an Act of Parliament, and all three are being written now, in documents published by health authorities rather than by legislators.