A go-to-market strategy for Threshold — an agent that carries a patient's enquiry through to a visit. The gap isn't that clinics are slow: a hungry solo owner answers Instagram in 2–10 minutes. The gap is coverage — after hours, on the phone, at peak, on the second and third channel. The product sells the consistency of that same response, not filling a void.
Seven findings. high — several independent sources, medium — one, estimate — the author's calculation.
An MIT study of 15,000 leads: the odds of contact fall a hundredfold when you respond after thirty minutes instead of five. Harvard Business Review, on 2.24m leads: responding within an hour gives a sevenfold better chance of qualifying.
A solo owner answers a DM in 2–10 minutes — it's their main channel on a phone with push. But 40–60% of enquiries arrive evenings and weekends when the front desk is gone; on the phone, small US practices miss 25–40% of calls; at peak a DM waits 30–90 minutes. The "63% don't respond" from Chili Piper is about web forms in B2B sales, not a beauty DM. estimate transfer.
81% of US medspas are single locations; about 839 of 10,488 are in chains. At an average chain of six, that's roughly 140 groups nationwide and a revenue ceiling of about $1.1m.
The five clinic-management systems aren't the threat — they sell records. The threat is companies like EliseAI, with a $250m round at a $2.2bn valuation, and Assort Health, with $120m at $1.2bn — working in patient scheduling and able to come down into the niche.
Vertical AI agents are the fastest-growing segment, at about 62.7% a year. Healthcare leads on adoption at 68%, and the driver is direct: a staff shortage that horizontal AI can't close.
Inference eats about 23% of revenue at growing companies in the category. We assume 65% instead of the industry's 75% for pure software, and that changes the whole unit economics.
Pay-for-outcome requires a control group; a control group requires a volume of enquiries. A single clinic doesn't have it. The pricing decision and the segment decision turn out to be one decision.
A section that wasn't there. All of the product's value rests on the claim that response speed changes conversion — and that claim has to be proven with someone else's data, not our own words.
Three studies converge. The MIT–InsideSales work covered six companies, over 15,000 enquiries and more than 100,000 dial attempts: the odds of contact fall a hundredfold if the response comes after thirty minutes instead of five. An enquiry handled within five minutes qualifies twenty-one times more often than one handled after half an hour.
Harvard Business Review analysed 2.24 million enquiries: companies that responded within an hour qualified a lead almost seven times more often than those who waited longer. A delay of a day or more cut the odds by about sixty times.
And here is where the business appears. This is not "clinics are slow." In business hours, on the main channel, the owner answers in minutes. The gap is where they're absent: after 7 pm, weekends, at peak, on the phone (a missed call is a lost high-intent patient), in a second messenger no one monitors. Threshold closes those windows with the same level of response — it doesn't "replace a live receptionist."
Overlay this on the economics of the niche from the market research: gross profit per patient is about $3,100. An enquiry that arrives out of hours and gets no reply by morning has most likely gone to a competitor who did reply. A lost evening stream is a lost slice of the ad budget that was spent making that enquiry appear in the first place.
The second thing that needed to be found out, and the one that overturns the obvious strategy: there are almost no mid-sized buyers.
More than 90% of clinics are owned by independent operators. About 839 locations are in chains and fund platforms — at an average chain of six, that's roughly 140 groups nationwide.
Selling to them at $28,000 a year hits a ceiling of about $1.1m even at 28% penetration. You can't build a company on that.
But the structure is changing: the industry association forecasts a record year for M&A, and the number of publicly announced deals has grown from a handful in 2019–2020 to more than forty a year for the last three years. You sell not to the chains but to the ones assembling them. A platform that has bought twenty clinics gets an urgent task — to bring twenty different ways of handling enquiries down to one — and an investor mandate to show profit growth.
The estimate was made two independent ways and reconciled. Agreement within an order of magnitude counts as confirmation.
| Level | How derived | Estimate |
|---|---|---|
| TAM bottom-up | 10,488 locations × $12k blended annual price | $126m |
| TAM top-down | Global medspa software market $384.7m × North America's 46.15% share | $178m |
| Beachhead | ≈140 groups × 28% penetration × $28k | $1.1m |
| Adjacent verticals | Dentistry, dermatology, ophthalmology — the same enquiry structure | not estimated |
The US medspa software market is $126–178m a year across all vendors combined. This isn't the niche in which you build a billion-dollar company without going beyond its borders.
The fund-platform beachhead yields about $1.1m of revenue and hits a ceiling. It exists not for the money but to earn the right to the mass segment and to the adjacent verticals. Without going beyond medspas the strategy makes no sense, and that has to be built into the plan from day one, not discovered in year three.
The section whose absence was the main mistake of the first version. A product with a language model inside competes not with records systems but in the vertical-AI-agent category.
| Category metric | Value | What it means |
|---|---|---|
| Growth of vertical AI agents | ≈62.7% a year | The fastest-growing segment by solution type |
| Conversational AI in healthcare, 2026 | $21.6bn | From $17.2bn in 2025; forecast to $212.9bn by 2036 |
| Voice agents in healthcare | $468m → $3.2bn | From 2024 to 2030 |
| Agent adoption in healthcare | 68% | First among all industries |
Healthcare leads on AI-agent adoption not out of a love of technology. The driver is stated plainly: a staff shortage that horizontal solutions can't close, legally or clinically. It's the same shortage that the market research names as the bottleneck of all aesthetic medicine.
The practical positioning conclusion: sell the product not as a marketing tool but as a solution to a staffing problem. A clinic's marketing budget is limited and contested; its budget for missing hands isn't.
The first version named five clinic-management systems as competitors. That was a mistake: they sell records and work in a different budget.
| Who | What they do | How dangerous |
|---|---|---|
| EliseAI | Patient scheduling and communication. A $250m round at a $2.2bn valuation in 2026 | The most dangerous. Capital and product let them come down into the niche in a quarter |
| Assort Health | Voice agents for medical practices. A $120m round at a $1.2bn valuation | Highly dangerous. An adjacent vertical, the same class of task |
| Voice platforms: Bland, Vapi and similar | Infrastructure for voice agents | Not competitors but a potential layer beneath the product. A partnership rather than a competitive scenario is possible |
| AestheticRecord, AestheticsPro, Boulevard, Moxie | Records, scheduling, charts, POS | Low danger as competitors, but dangerous as a channel: they could build in a similar module |
A company with a $250m round can enter the medspa niche in a single quarter and subsidise its price longer than a startup will survive. So the defence isn't in development speed or in price.
The defence is in two things money can't buy. The first — a measured lift in visits against a control group at specific clinics: evidence accumulated over time, which can't be sped up with capital. The second — a pay-for-outcome model that a large player with subscription revenue won't switch to, because it would collapse their own revenue reporting.
From this the plan's urgency changes: the Q2 gate matters more than any sales. Until there is a measured lift, the company has nothing that money can't buy.
An agent that takes a patient's enquiry from the first message to a confirmed booking. The name means threshold: the product is responsible for everything that happens before a person crosses the clinic's threshold.
Instagram DMs, WhatsApp, the website form, missed calls. One inbound stream instead of four scattered ones.
Responds within a minute, addresses specific fears from a script, states the price out loud and offers a time. The lead reaches the receptionist with context.
The count of extra completed visits, measured against a control group. Not messages, not minutes, not features.
| Version | What's included | Why exactly this |
|---|---|---|
| First | One channel — DMs. A script that addresses four barriers, price publication, handoff to the receptionist, a control group | The minimum on which a lift can be measured. Everything else distracts from the proof |
| Second | Missed calls, the web form, a control-group report for the client | Fullness of the stream, and the client checking the result themselves |
| Third | Recovering those who didn't complete, working with no-shows, integration with the records system | Expansion into the client: keeping revenue above 100% with no manual upselling |
Compliance requirements aren't a detail — they're a deadline. A medical context means working with health data. Certification and security reviews add two to four weeks to every deal. This raises the cost of selling and, at the same time, serves as a barrier to small competitors.
Last-touch attribution doesn't work: the enquiry came from the clinic's advertising, and the product only handled it. Counting every booking as your own achievement is a deception that unravels at the first review.
The solution is a random control group. Fifteen percent of enquiries go through the clinic's old process; the rest go through Threshold. Once a quarter, the share that completes a visit is compared. The invoice is issued only on the difference, and the calculation is published in the invoice.
The mechanic chooses the segment by itself. A control group needs volume: at two hundred enquiries a month per location, fifteen percent yields thirty observations, and over a quarter about ninety accumulate against five hundred and ten. That's enough to see a difference of eight to ten percentage points.
A single clinic doesn't have that volume. So pay-for-outcome only applies where volume aggregates — in groups and platforms. Single clinics get a subscription with no outcome component, and that's not a concession but a consequence of the maths.
The expected size of the lift. The effect is expected not from "the clinic is slow" but from the closed windows: evening and weekend volume, missed calls, the second channel. The sevenfold multiples from the studies can't be transferred: they measured lead qualification in B2B, ours is a completed visit, and part of the effect is eaten by the clinic calling back later anyway. A realistic Q2 gate is eight percentage points, deliberately conservative; if a client already answers fast on every channel around the clock, the lift will be lower — and that client isn't ours.
A hybrid of a base fee and an outcome component is the current market standard: hybrid models hold 41–43% and are forecast to reach 61% by the end of 2026.
| Tier | For whom | Base | Per outcome | How we sell |
|---|---|---|---|---|
| Solo | Single clinic | $350 / mo | none | Self-serve |
| Group | 2–9 locations | $400 / location / mo | $40 per visit | Inside sales |
| Platform | 10+ locations | $350 / location / mo | $30 per visit | Field sales |
The outcome component exists for retention: companies with it show 31% higher customer retention and 21% higher satisfaction. That is what pushes net revenue retention above 100% with no manual upselling.
The benchmark for a per-outcome price is taken from adjacent categories, where handling an enquiry costs from fifty cents to a dollar. Here the unit is an order of magnitude more expensive, because one completed visit brings the clinic about $3,100 of gross profit over eighteen months. Forty dollars is 1.3% of the value created.
The margin rationale. The 2026 benchmark for pure software is 75% and above. But at growing companies with a model inside, inference eats about 23% of revenue, public AI companies show margins 10–17 points below their previous level, and AI-product builders expect about 52% on average. We assume 65% — the top of the AI-native band, reachable with discipline on compute cost.
Deriving five-year value. Annual gross profit of $18,200 grows 15% a year at 115% net revenue retention. Sum: 18,200 × (1 + 1.15 + 1.32 + 1.52 + 1.75) = $122,711. Divide by the acquisition ceiling and you get 6.7 to 1 against a benchmark of 3 to 1.
Comparison with the mass model. Selling to single clinics at $9,000 a year with 97% retention, typical for small business, the same five-year value is $27,500 — four and a half times less.
Why a 12-month payback. The industry median is 16 months; for vertical software it's 18 against 14 for horizontal. Twelve is taken as a target constraint, not an observed figure.
Calculated per rep on 2026 benchmarks.
| Metric | Value | Source |
|---|---|---|
| Quota per rep | $500,000 | Planning assumption |
| Quota attainment | 45% | Mid-market median, band 40–55% |
| Actual revenue | $225,000 | Calculation |
| Deals per year | 8 | 225,000 / 28,000 |
| Win rate | 24% | Mid-market median, band 20–28% |
| Qualified opportunities per year | 33 | 8 / 0.24 |
| Cycle length | 75 days | Middle of the 60–90-day band |
| Ramp to capacity | 6 months | Middle of the 4–12-month band |
| People on the buying committee | 6.8 | Average size, up from 5.4 |
The model produces an unpleasant arithmetic: one rep closes eight deals a year and needs thirty-three opportunities. The whole market has about 140 suitable groups. Two reps would exhaust the list in eighteen months.
This isn't an argument against the strategy but its main constraint: the sales team is built not for growth but to obtain proof. Eight to twelve platforms with a measured lift is the asset the rep is retained for.
Note separately: cycles have lengthened 22% since 2022, CFO involvement is up 40%, and security reviews add two to four weeks.
The sequence is essential. Launching the mass motion first means entering a segment with 97% net revenue retention and no proof of value. That is exactly the scenario that makes vertical software in small business unprofitable.
If a gate isn't passed, the next quarter doesn't begin and the strategy is reviewed. This isn't a figure of speech: the Q2 gate can close the project.
Pass health-data protection certification — deals don't close without it. Launch one paid pilot with a control group. Gate: the pilot works, the control group is accumulating statistics.
Take the pilot to a full quarter of measurement. Close two or three deals. Gate: a measured lift above eight percentage points. Below that — the product doesn't work, and that has to be acknowledged, not sold further. This quarter matters more than all the others: before it, the company has nothing money can't buy.
Hire a second rep and an implementation specialist, get to eight customers. Verify that the lift reproduces across different clients. Gate: the lift reproduced at a minimum of five.
Publish the method and the results, including the failed cases. Open self-serve. Gate: net revenue retention on the beachhead above 100% and the first sign-ups with no rep.
Who to hire and when. The section whose absence made the plan unworkable.
The hiring principle. The implementation specialist comes before the second rep in importance, though they're hired at the same time. The reason is that a badly set-up control group devalues the whole quarter of measurement — and that is the company's only asset until Q4.
Calculated through all four gates. Every assumption is in the table; none is hidden.
| Quarter | Team | Customers | Monthly spend | Monthly net | Cumulative |
|---|---|---|---|---|---|
| Q1 | 4 engineers, 1 rep | 0 | $87k | −$87k | −$261k |
| Q2 | same | 3 | $87k | −$82k | −$508k |
| Q3 | + rep, + implementation | 8 | $107k | −$95k | −$793k |
| Q4 | same | 14 | $107k | −$86k | −$1.05m |
What this calculation says honestly. For a year and just over a million dollars, the company gets $392,000 of annual revenue and 10% penetration of a segment whose ceiling is $1.1m. As a standalone business this doesn't work.
It works only as buying the right to the next market. A million dollars is spent not for medspa revenue but for eight to twelve measured cases, with which a conversation opens in dentistry, dermatology and ophthalmology — verticals with the same enquiry structure and an order of magnitude more practices.
If this isn't stated to the investor upfront, an expectation conflict starts in Q3: the investor asks about revenue while the team builds proof. It has to be framed from the start: this is a round for proof, not for growth.
| Objection | Response |
|---|---|
| "We already have AestheticRecord" | And you'll keep it. That's a records system — it stores what has already happened. We work before it happens. Different budgets — operational versus marketing. |
| "Boulevard does online booking too" | Online booking serves people who have already decided. We work with the ones who wrote in and went quiet. Ask how many DM conversations a month ended in nothing. |
| "The receptionist copes" | In business hours — probably. The question is about 9 pm, Saturday, peak and the phone. Let's check with a control group: leave fifteen percent of enquiries to the receptionist, run the rest through us. If there's no difference, you pay nothing for outcome. Ask how many of last month's evening enquiries got a reply by morning. |
| "Too expensive" | The base component is less than one location's daily revenue. The outcome component is charged only on extra visits against a control group. One completed visit brings a clinic about three thousand dollars of gross profit over eighteen months — we charge forty. |
| "Patients don't want to talk to a bot" | Agreed, which is why the product doesn't impersonate a person and hands the conversation to the receptionist at the time-selection stage. We measure not chat satisfaction but the people who got to the chair. |
| "What if a big player with a billion-dollar valuation shows up" | They will. But they have subscription revenue, and they won't switch to pay-for-outcome — it would collapse their own reporting. And by then we'll have measured lifts at your peers, which money can't buy. |
The fork is decided at the Q2 gate.
By year-end, 14 customers and a published method. A conversation opens with adjacent verticals, where the number of practices is an order of magnitude higher.
The product works, but not enough to justify pay-for-outcome. The model moves to pure subscription; positioning shifts from revenue to saving the receptionist's time.
The project closes in Q2, having spent about $500k instead of a million. This is the best of the bad outcomes, and it is what the gates exist for.
A big player entering. The main risk, and it has changed from the first version of the strategy. A company with a $250m round enters the niche in a quarter and subsidises its price longer than a startup will survive. Mitigation: measured cases and a pricing model a subscription player won't switch to.
The beachhead is too small. About 140 groups is eighteen months of work for two reps and a $1.1m ceiling. If the move into adjacent verticals doesn't happen, the company hits a wall. This is a structural constraint, not an operational risk.
Fast responders shrink the TAM. A hungry solo owner answers a DM in 2 minutes — for them Threshold's lift is near zero, and the control group will show it. Our real market isn't "medspas" but multi-location groups and clinics with high call volume and a real out-of-hours stream. That narrows the 140 groups further and reinforces "the beachhead is too small."
The control group shows no lift. The mechanic that makes the offer honest can refute the product. This is deliberate: better to find out in Q2 than in year three.
Compute cost eats the margin. Inference is already about 23% of revenue at growing companies in the category. If the margin falls below 55%, a 12-month payback stops adding up. Mitigation: monthly control of the cost per conversation as a key metric.
Regulatory. An automated conversation with a patient falls under health-data protection requirements. It raises development cost and lengthens deals, but serves as a barrier to entry for small players.
Expectation conflict with the investor. A round for proof looks like a round for growth if the difference isn't stated in advance. Not a technical risk, but one that can kill the company in Q3.
The size of the effect hasn't been measured in our niche. The response-speed data comes from sales in general and from healthcare as a whole, not from aesthetic clinics. Transfer by analogy is the main assumption of the whole strategy.
The number of groups is an estimate. The figure of 140 comes from dividing the number of chain locations by an assumed average chain size of six. The real distribution is unknown.
The European market isn't in the calculation. All ownership-structure and software-market-size data is for the US.
Adjacent verticals aren't assessed. Dentistry and dermatology are named as the growth path, but their size, ownership structure and willingness to pay haven't been checked.
Our own customer acquisition cost is unknown. In the sales model it's derived from quota and win rate, but the actual cost of outbound sales into this segment hasn't been verified.
Big players' reaction is unpredictable. The claim that a subscription player won't switch to pay-for-outcome is reasoning about incentives, not an observation.
| Metric | Value | Source and status |
|---|---|---|
| Response speed and coverage | ||
| Drop in odds at 30 min instead of 5 | 100× | MIT / InsideSales, 15k leads |
| Advantage of a 5-minute response | 21× | Same |
| Advantage of a response within an hour | ≈7× | Harvard Business Review, 2.24m leads |
| Loss from a day's delay | ≈60× | Same |
| Healthcare ideal | <10 min | Industry review |
| DM response, business hours | 2–10 min | Field check (beauty DM) |
| Enquiries out of hours | 40–60% | Estimate from industry patterns |
| Missed calls, small US practices | 25–40% | Ruby / Weave, 2024–2025 |
| "Don't respond" (web forms, US B2B) | 63% | Chili Piper — different context |
| The category | ||
| Growth of vertical AI agents | 62.7% | Report, 2025–2030 forecast |
| Conversational AI in healthcare, 2026 | $21.6bn | Report |
| Agent adoption in healthcare | 68% | Industry review, first place |
| EliseAI, round and valuation | $250m / $2.2bn | Press, 2026 |
| Assort Health, round and valuation | $120m / $1.2bn | Press |
| The sales market | ||
| US medspas | 10,488 | Report, 2023 |
| Single locations | 81% | Report |
| In chains and platforms | 8% ≈ 839 locations | Report |
| Estimated number of groups | ≈140 | Author's assumption: 839 / 6 |
| TAM bottom-up | $126m | Author's calculation |
| TAM top-down | $178m | Author's calculation |
| Economics | ||
| Annual contract | $28,000 | Planning assumption |
| Gross margin | 65% | Justified by inference cost |
| Inference share of revenue | ≈23% | Industry review |
| Customer value over 5 years | $122,711 | Author's calculation at 115% retention |
| Value-to-acquisition ratio | 6.7× | Author's calculation |
| Net revenue retention, small business | 97% | Industry review |
| Net revenue retention, mid-market | 108% | Same |
| Retention uplift from pay-for-outcome | +31% | Same |
| Sales and funding | ||
| Quota attainment | 45% | Mid-market median |
| Win rate | 24% | Mid-market median |
| Cycle length | 60–90 days | Industry review |
| Rep ramp to capacity | 4–12 mo | Same |
| Funding requirement for the year | $1.05m | Author's calculation |
| Same with a 30% buffer | $1.37m | Author's calculation |
| Customers by year-end | 14 | Planning assumption |