Founder Stack Ep01: How an AI Voice Agency Actually Makes Money — Matthew Cohn on Content as the Only Free Channel, Value-Based Pricing, the Three-Problem Rule, and SOPs Before Hiring

The challenge taught him two things he now states as rules. Consistency pays: at the start you get no likes, no views and no attention, and the only way through is arithmetic. Twenty posts a month at 200 views each is 4,000 people — "if you put 4,000 people in a room you'd be absolutely amazed at how big of a crowd that looks", but on a screen 200 reads as failure. More posts, more visibility; more visibility, more connections; more connections, more opportunities. Content is king: one post reached 75,000 people, drew 2,000 comments, and signed roughly eight clients directly; over the challenge he reached 87,000 accounts and half a million impressions for zero money and only time. He then added YouTube long-form — screen recordings made with a £20-30 Amazon microphone and no editor — which produced a consistent lead flow because a viewer who watches thirty or forty minutes trusts you differently from someone who saw a thirty-second ad.

He posts three kinds of thing. Demonstrations: show the tech working — a three-minute video of him talking to his own Scottish voice-agent receptionist, or a free resource on writing voice-agent prompts. ICP problems, in the client's language, not his: "you're missing calls"; "you have leads coming in that you've not called in two days." And personable posts that simply let people know him — a selfie, a joke, his parents — which he notes are called engagement farming and rarely sell anything, but which mean that when you do reach out, they recognise you. On all of them, "the hook is the most important part", and the second effect of content matters as much as the first: when you DM someone cold, they check your profile, and a profile with nothing on it reads as suspicious.

Four failures, in sequence. The client does not value it, because they did not pay for it — "you could get a new laptop for free and you wouldn't appreciate it as much as if you paid £2,000." They have no skin in the game, so when the build hits its inevitable rough patch they say turn it off, it doesn't work, I told you AI doesn't work — and your weeks are gone. You start resenting the work, especially if it turned out harder than you agreed to; effort drops, and when effort drops quality drops, and when quality drops the result confirms the client's doubt and your own. And the dynamic that payment creates runs both ways: they acquire an expectation and will chase you to make it work, you acquire a responsibility and a little useful pressure. "Pressure makes diamonds."

He has never done a fully free project. He has been burned starting work on a promise of payment on launch — "now that I say that out loud it sounds so stupid" — and his floor is now 50% upfront. A free setup with a performance-based back end is a categorically different arrangement, because both sides have skin in the game: it is in the client's interest for you to perform. And if the objection is genuinely risk, de-risk it with a guarantee rather than a discount: charge, and refund if the project flops. His caution is that the client does not know what kind of business person you are and may have been burned before, so a guarantee still has to be sold — but it converts their hesitation into motivation to make the project work.

Builders fall into the features trap because features are what fascinate builders. He did: he found voice agents genuinely exciting, talked to them on the phone for fun, and led with how cool they were. Enthusiasm gets you in the room; it does not close. What a business owner actually asks is: does it work, will my problem be solved, will you make me more money, will you book me more appointments, will I save money, will I save time. That is the entire list.

So he rewrote his own sentence. "I build voice agents" became "I help businesses recover lost leads", and at its boldest "I help businesses make more money" — which invites the question how, at which point you get to explain the technology because they asked. "I build voice agents means nothing to the guy who runs a plumbing company." But "I help businesses answer every call 24/7" lands immediately on someone who knows they are losing weekend calls. The mechanics change too: describe the product in the buyer's language, not yours, and identify both the specific problems you solve and the specific audiences that have them.

The confusion is that niching is assumed to mean one industry: we only help roofers, we only help financial advisors. But what he actually found was a problem he wanted to solve — helping appointment-booking service businesses capture, qualify and convert more leads — and that problem's shape is identical across plumbers, roofers, solar installers, med spas and dentists. His speed-to-lead offer calls any new lead within two minutes, qualifies them and books an appointment; the sales process model does not change when the industry does.

So the niche is the problem, and the audiences are whoever has it. That keeps the specificity that makes marketing work — you can talk about missed weekend calls in detail because you know that world — without capping the market at one vertical. A vertical niche is still legitimate if you want one; the point is that it is a choice, not the definition.

His analogy: a hair salon with an average customer value of $100, and a real-estate broker at $10,000. Both miss 50 calls a month for the same reasons — staff off sick, after hours, nobody free. The salon's exposure is roughly $5,000 a month; the broker's is roughly $500,000. Same tech, similar build — he argues the salon's would actually be the harder build, because the receptionist logic is more complicated — and therefore two completely different prices. Value-based pricing simply asks: what is it worth to this business to solve this problem?

Which forces a second decision: choose who you sell to. He charges $4,000 to $10,000 and sometimes $20,000 for a system, so a salon is not a prospect at any level of persuasion. "Don't go into an industry that has razor-tight margins and won't be able to afford your service." Cheap SaaS can sell to thin-margin businesses; a bespoke build cannot. He is candid that this is not a magic wand — "I'm not gonna be that guy that makes it sound like you just throw a number out there and people say yes" — but that higher-ticket selling is easier into businesses with money.

Paul asked him to define outsized value, and the answer is arithmetic. Take the broker: 50 inbound calls, 70% would become appointments, so 35; a pessimistic 10% appointment-to-sale, so 3.5 sales; at $10,000 average value that is $35,000 a month recovered. Against a $5,000 setup and $1,000 a month. "That's outsized value." The number you charge is not compared to your time; it is compared to the gain you can evidence.

Model one, custom builds: a setup fee plus a monthly retainer for management and optimisation. This is where he started and what most people start with, and he is explicit that the retainer is the point — "I've charged a massive setup fee, and it feels really good in the moment, but then after 2-3 months that slowly dwindles away, and your recurring is really small, and you're like, oh wait, I need to just constantly be on this hamster wheel of closing more deals." When a client resists the retainer, the script is the staffing analogy: you would not hire someone and expect them never to improve, never to stop repeating their first-week mistakes. And a client who refuses outright is one he declines.

Model two, pay-per-result: reactivate a dormant lead database with voice and text agents, and charge per qualified appointment booked. He launched it at zero setup fee because he did not know whether anyone would take it, could get a client live in ten days, and got paid as results landed. After fifteen clients proved the offer he added a £1,500 setup fee — and discovered the fee is also a negotiating instrument. He will trade it down against a higher per-appointment rate for a business with plenty of leads and a good offer, because he can now look at the lead volume and the offer and predict the outcome. "If you can tie pricing to results, it removes their biggest objection."

Model three, coaching and mentoring other people building voice agencies — consulting on prompting and frameworks, or handing over the full roadmap.

His biggest early mistake was instinctive: a prospect mentions a problem, and he immediately explains how he would fix it. It feels helpful and it is how builders think. But solving problems one at a time produces "small wins or small hits" and makes the service as a whole feel thin. Hold them instead: extract at least three problems before presenting anything, and then present a single solution that answers all of them. "That hits way harder, because it makes them feel like you're understanding everything that they're going through."

That requires two calls. Call one is discovery, framed out loud so the prospect knows they are not being sold to — which also sets the expectation that they will be answering questions. Call two is the proposal. Between them, stay in contact. And open call two with what he calls opening the wounds back up: restate the pains from call one and have the prospect confirm them. This does two jobs — it proves you listened, and it brings the discomfort to the front of their mind immediately before the solution appears.

The other output of call one is arithmetic. Get the lead volume, the lead-to-appointment rate, the appointment-to-sale rate and the average sale value, because the proposal is built by reverse-engineering those numbers. He also collects the owner's ambition: "it doesn't need to be all doom and gloom", and weaving where they want to take the business into the story helps close.

The discovery numbers were: 500 leads a month, a 15–20% lead-to-appointment rate, a 35% close rate, and roughly £3,000 average treatment value. So 75–100 consultations and 26–35 treatments a month. He had already taken a comparable dental practice running the same offer from 18% to 29%, so he applied 29% to these numbers: 145 consults, 50 treatments, 45 extra consults, 15 extra customers, +£45,000 a month — "and we're not increasing your ad spend, because it's the same amount of leads that you already have, and you're not having to hire more staff." He sold it at £6,500 setup plus £1,000 a month, and closed it.

The document's structure is worth copying. Headline: their stated goal, verbatim — "more deposit-paid appointments on autopilot". Then the wounds, reopened in the client's own words: the ads work, the clinic closes well, the break is in the middle. Then why, again in their words: inconsistent speed of follow-up, after-hours and weekend leads, and no visibility into which leads were missed. Then a single focus — "we're laser-focused on one thing" — with the mechanism under it and the line "this takes 90% of the follow-up burden off your team". Then a map of the actual solution, sub-agents for each service line, a screenshot of what the CRM looks like populated (taken from the comparable client's real account), a live callable demo built for them, the numbers, future-pacing ("if you doubled your leads the system runs the same"), what to expect, two case studies, and finally two priced options with one recommended. They took the recommended one.

Paul asked the question that guards the whole method: how do you know you can actually achieve those results? The answer is a track record plus diagnostic reasoning — if the owner has no dedicated five-minute caller, does not work weekends, and has no structured follow-up, then he knows they are not doing the things that raise the rate, and he has the case studies to price the uplift. He also deliberately under-promises: where he can see 30–40% is achievable he quotes 25%, "and then when we start working with them, we over-deliver."

Two claims, in order. First, you must understand fulfilment well enough to do it yourself before you hire — he fulfilled every project personally for roughly the first year — because "if you don't know what good looks like, you can't properly evaluate whether someone else can do it." His hiring process proved the point: plenty of applicants claimed extensive voice-agent work, and when asked to show a system they had built, could not. "You're lying, you know? So many people just talk a big game and can't actually deliver, and that is definitely not someone that you want on your team, because then when they can't deliver, it's your reputation that gets burned." He is directly contemptuous of the online narrative that you can start an agency and hire out all the work.

Second, once you do understand fulfilment, get help quickly — doing sales, marketing, admin AND fulfilment makes scaling impossible, and he admits he held onto fulfilment too long and "lost my way". But the transition has a prerequisite: "when you're solo, you can coast in chaos. The moment you bring someone in, your chaos quickly becomes the whole bottleneck of the business." He brought on a developer with close to zero SOPs, and had to walk every step manually — which defeated the purpose, ate the time he should have spent growing, and slowed the developer's learning because there was nothing to read. A mentor named it. He then spent three ten-hour days documenting every fulfilment step, which he says accelerated the developer's learning tenfold and has freed hundreds of hours in the eight months since.

His advice is to start SOPs now even if it is only you and even if the process will change, because a changed process means editing a document that is 60–70% right, not starting from nothing.

He does not need anyone to switch. Most business systems publish API documentation, and where they do, the voice or text agent can be integrated into the system the business already runs. He lists what he has worked with: HubSpot, Zoho, GoHighLevel, Jobber, and "this other really weird one called Simpro". The test reduces to one question — does the incumbent system have detailed API documentation? — and if it does, the switching objection disappears, because there is no switch.

The companion point, raised by a cohort member whose client's owner had approved but whose staff kept hesitating, is that buy-in is two separate pitches. To the owner you sell time and money. To the staff you sell their own day: you will not have to follow up with hundreds of leads, you will not waste time on calls nobody answers, you will not stay late clearing a backlog — more of the work you enjoy and less of what you do not. The deployment stalls without the second pitch.