The bolt-on tax: why many EHRs price AI as an add-on (and what that signals about their architecture)

Stacy runs a three-therapist group practice in Austin, Texas. Last Tuesday she asked me a question I've now heard some version of from eleven clinic owners in six weeks. "If I add AI notes to my SimplePractice subscription, what am I actually paying?" She'd already done the math. I just confirmed it. Her SimplePractice bill goes up by $105 a month — $35 per clinician for the AI Note Taker add-on. That's $1,260 a year, for a feature the marketing calls "the future of charting."

What bothered her more than the number was the sameness. Same software. Same login. Her team already paid for the plan. Turning on AI notes didn't move her to a new product. It just added a line to the invoice.

I call this the bolt-on tax.

The bolt-on tax, defined

The bolt-on tax is the surcharge you pay when an EHR vendor has not absorbed AI into its core product, so it ships AI as a separate SKU, a higher tier, or a usage cap. You are paying extra because the vendor's product economics have not made AI part of the base subscription. That is a signal about architecture and margin structure, not a moral judgment about their team.

I want to name it because the name matters. If you see "$15/practitioner/month AI Scribe" on a pricing page, that's not just a feature announcement. It tells you AI has separate entitlement logic, separate billing logic, and probably a separate cost model inside the company. The product may still be good. The invoice is telling you where the new capability sits.

Plenty of established EHRs have some version of this. The specifics change, but the pattern keeps showing up.

The numbers, vendor by vendor

Let me lay out the exact costs for a three-practitioner clinic so nobody thinks I'm cherry-picking.

Jane App. Jane's AI Scribe includes five free notes per month, then unlimited AI Scribe costs $15 CAD per opted-in practitioner per month. The base plans are $54, $79, and $99 CAD; for a three-practitioner clinic on Practice, Jane's calculator data puts the base at $79 for the first full-time practitioner plus $35 for each additional full-time practitioner. So the AI line is still the clean part of the math: three therapists pay an extra $45 a month just for unlimited AI documentation. $540 a year. For a team of five, it's $900. That money buys AI-generated notes. It does not buy additional scheduling. It does not buy additional intake. It buys one capability clinicians already use as if it were core.

SimplePractice. SimplePractice's AI Note Taker is $35 USD per clinician per month after the 30-day trial, and their support page says it is available across Starter, Essential, and Plus as an optional add-on. Three clinicians: $105 a month. Five clinicians: $175. Over twelve months a five-therapist practice is spending $2,100 on AI documentation alone. For a three-clinician group, the base math is not $79 times three. SimplePractice says additional team members are only available on Plus, with Plus starting at $99/month and additional clinicians for 2-5 clinician practices at $74/month each. That makes the three-clinician base $247 before the $105 AI add-on.

Carepatron. Carepatron does something more interesting, and honestly, I almost admire the design. Their Free tier lists one million AI tokens per month. Plus and Advanced list AI tokens as unlimited. That means Carepatron is not selling AI as a separate add-on the way Jane and SimplePractice do; it is making AI usage visible through a cap on Free and unlimited usage on higher paid tiers. Public token math does not support treating that cap as only a handful of sessions. OpenAI's public token rule of thumb puts one million tokens closer to hundreds of thousands of words before you account for prompts, outputs, retries, and whatever internal pipeline Carepatron uses. The honest version is less punchy: the free plan is capped, the paid plans remove the cap, and the public page does not tell clinicians exactly how many complete notes that cap buys.

Across those three vendors, the pricing models differ, but the signal is the same: AI usage is separately visible, whether as an add-on, a per-clinician SKU, or a token cap.

What a three-practitioner clinic pays per month

I compared 11 platforms for solo practitioners in a pricing breakdown last month. The pattern there was the same pattern here. The more legacy feature layers a platform has, the more line items its invoice tends to carry.

Why the line item exists

Here's where most articles on EHR pricing stop. The number is the number. Pay it or don't.

I want to keep going, because the number is the output of something more interesting. When a vendor prices AI separately, they are telling you, without meaning to, how the product economics work. Jane's current AI Scribe page says the tool is built into Jane, and its FAQ says Jane sends the recording to a trusted compliant third party to create the chart note. That's not a scandal. It is the kind of implementation detail that explains why unlimited AI Scribe has its own price. Somebody had to connect the workflow, the consent model, the third-party processing, the storage rules, and the billing toggle. Somebody on the finance side had to price that bundle so it didn't run at a loss.

This is the engineering economics part. If AI is core to the platform's base promise, the cost of inference can be amortized across the subscription the way hosting and storage are. You don't pay extra for "database access" or "HTTPS." You pay once, and the product includes the infrastructure required to run. When AI gets its own line item, it means the vendor has chosen not to absorb that infrastructure into the base product. It may mean a separate vendor contract, separate usage controls, separate margins, or simply a pricing strategy designed to keep the base plan lower for people who do not use AI. The invoice does not prove the codebase shape by itself. It does prove the company is tracking AI as a distinct cost centre.

There's no dishonesty here. The teams building these products are smart and working hard. The ring-fencing is rational. You can't cross-subsidize inference costs indefinitely without pricing pressure on the base plan. SimplePractice says this part out loud in its Note Taker FAQ: AI computing has meaningful costs, and some customers prefer optional add-ons so the base subscription does not rise for everyone. That is a fair argument. Inference is not free. Someone pays. The question is whether the product treats it as a utility bill or as a core capability.

Many legacy EHRs treat it as a utility bill. Newer AI-first EHRs try to treat it as a core capability. That choice is usually made long before the first AI toggle appears in a billing screen.

There's a subtler problem with the fixed-price add-on that most people miss. When every user pays the same $15 or $35 per month regardless of usage, the vendor's margin is highest on the practitioners who use AI the least. The therapist who runs AI Scribe once a week is subsidizing the one who runs it thirty times a day. The vendor knows this, and the incentive it creates is uncomfortable: the less you use the feature you're paying for, the more money the company makes. That is a pricing model that quietly roots against your adoption.

It also creates a structural tension inside the product team. Every product improvement that makes AI more useful — better prompts, richer outputs, longer context windows — increases token consumption per user. But the price is fixed. So the vendor is trapped between two bad options: keep raising the fixed price to cover growing inference costs, or quietly limit the AI's capability to manage token spend. Cap the note length. Reduce the model quality. Throttle how many times a day you can use the tool. The customer doesn't see the constraint as a pricing decision. They see it as the AI being "not that good." The fixed price made that outcome inevitable.

What happens to platforms that can't absorb AI

Look three years ahead. Today the bolt-on is AI Scribe. Tomorrow it will be AI treatment planning. Then AI intake, AI billing optimization, AI patient messaging. Each one lands in the product as a discrete feature built by a different team, with its own vendor contract, its own usage metering, and its own line on your invoice. That is the structural trajectory of treating AI as a set of individual features to be priced rather than a capability woven through the product.

Legacy EHRs are walking into a multiplication problem. Every new AI feature they ship has to be individually scoped, individually billed, and individually justified to the customer. Five AI features means five SKUs, five pricing decisions, five internal teams negotiating for margin. The vendor's roadmap says "deeper AI integration." The invoice says "more line items." Those two things are in direct tension, and the only ways out are raising the base price, collapsing the SKUs into a bundle that still costs more, or building a true AI-first replatforming behind the scenes that customers never see.

The difference with an AI-first architecture is not just pricing — it's organizational. When AI is the product's core layer rather than a collection of bolt-ons, one unified AI system powers scribe, intake, overview, charting assistance, and whatever comes next. The same credit buys any AI action. There are no five separate teams building five disconnected AI features with five separate billing toggles. The AI is harmonized because it was designed as one thing, not assembled from parts.

Stacy, the Austin practice owner, put it more cleanly than I would have. "I don't mind paying for software. I mind paying three times for the same software." That's the thing the bolt-on tax will keep signaling. Not that the vendors are unfair, but that their margin structure is fighting the product they want to ship.

A few of them will get through it. Some will be outpriced by newer platforms whose architecture treats AI as infrastructure — one layer, one credit, one price — instead of a feature catalogue with a separate checkout for each item. Five years from now, I expect "pay extra for AI notes" to read the way "pay extra for email" reads today.

The Oli position

I'll close with our number because the brief I wrote for this article said to. Oli is $19.95 per practitioner per month, and every core AI feature is in the base price. AI clinical notes. AI-assisted patient intake. AI patient overview. Each practitioner gets two million Oli AI Credits included monthly — worth $20, which effectively pays for the subscription itself. Beyond that free tier, additional credits cost $1 per 100,000, or 1c per 1,000 credits.

The usage-based model is the fairness argument the flat-fee vendors can't make. A chiropractor generating quick adjustment notes has fundamentally different AI needs than a naturopath dictating hour-long intake narratives, who has different needs than a social worker documenting complex case histories, who has different needs than a therapist running fifty-minute sessions five days a week. Charging all of them the same fixed $15 or $35 add-on means low-usage practitioners are subsidizing high-usage ones. That's not a pricing plan. That's a cross-subsidy hidden inside a line item. With Oli, if you generate one note a day, you pay for one note a day. If you generate fifty, you pay for fifty. The pricing page FAQs publish rough estimates: a one-hour continuous call with transcription, scribe, and notes runs around 50,000 credits. An AI overview or summary is about 10,000. A simple question-and-answer exchange is under 1,000. Most solo practitioners won't touch the 2M free ceiling.

There's another advantage that gets overlooked. Because every AI feature draws from the same credit pool, every feature is unlocked from day one. You can try AI Scribe, test AI patient overview, experiment with AI-assisted intake — all of it — and you only consume credits when you actually use a feature. If AI overview turns out to be something you run twice a month, you've spent roughly 20,000 credits total. Not $35. Not $15. Twenty thousand credits out of two million free. Compare that to a fixed add-on model where you're paying the same monthly fee whether you use the tool every session or forget it exists after week two.

Calling that price generosity would be dishonest. The architecture made it the natural number. We built the product in 2025 with modern AI models already assumed in the critical path, so there's no older AI-free product line to protect. The result is that we don't need a separate SKU to charge for AI, because running AI inference is a cost we already priced into the base plan. Whether you generate one note a month or fifty a day, the per-credit cost is the same — and that cost goes in one direction as inference gets cheaper.

If you're comparing platforms right now, the test I'd run is simpler than a feature matrix. Pull up the pricing page. Count the AI SKUs, token caps, and tier gates. If AI is priced separately, the platform is managing it separately. That might still be the right choice for your practice. A lot of clinics on traditional platforms have years of tenure and real switching costs. I don't think they should move for the sake of moving. I do think they should know what the line item means.

Frequently asked questions