The agency is the product

Ask what a small business owner bought the last time they paid for AI, and the answer rarely sounds like software. It sounds like a result: the phone gets answered after hours, the product is ready for the grocery shelf, the shifts are covered. The pattern running through the ideas and episodes collected this fall is consistent. The businesses with traction sell a finished outcome to plumbers, restaurants, and food brands, and the person doing the selling looks less like a software vendor and more like an agency. The AI does the work backstage. What the customer pays for is that the work is done.

The phone is the storefront

Start with the clearest case. HVAC shops answer only 34% of their calls, according to Invoca figures cited in the IdeaBrowser entry. A human receptionist averages $37,000 a year, which a solo contractor cannot justify, so two out of three calls ring out. Some of those calls are worth real money: a ServiceTitan case study cited in the same entry puts one recovered call at a $1,000 job.

The product built on that math is sold as a service, not a subscription to a tool. Setup costs $500, maintenance runs $299 a month. The owner hands over their rules once: where they work, which jobs they want, their minimum price, their hours. The AI then answers, collects the caller's information, and books or escalates. Spam, wrong zip codes, and the $80 jobs the owner does not want get turned away or referred out. The owner never logs into anything. Demand signals point the same direction: searches for "AI receptionist" have doubled while searches for "virtual receptionist" fell sharply. Buyers are searching for the outcome, and the winning offer is priced against the $37,000 salary it replaces, not against other software.

The granola maker does not want a dashboard

Food brands show the same shape from the other side of the counter. Small brands drove 86% of food and drink growth last year while holding only 24% of sales (BCG, cited in the entry). Growth is not their problem. Operations are: landing a new grocery store triggers testing, barcodes, and label work that a two-person brand has no staff for.

The proposed product is a background agent that handles the retail-readiness work while the founders make the product. Nobody pitches the brand a console to learn. A working version already exists and it is priced like a service: Retail Ready charges $1,999 for a retail roadmap and has 3,500 brands enrolled. The go-to-market in the entry is deliberately unscalable at first: walk the local aisle of a grocery store, write down every small brand on the shelf, call them, get friendly before selling anything, then set up agents around the problems they actually name. Distribution is relationships first, tooling second.

The webmaster is back

The October 6 IdeaBrowser entry names the pattern in its trend note: be the new-age webmaster. In 2000, small businesses did not buy web servers; they paid a person to build the website and keep it running. The entry argues 2026 is the same arrangement with agents in place of HTML. Its evidence for the gap: Claude for Small Business has passed 900,000 installs since May, yet most small businesses use AI for little beyond writing emails. Access to the tool is not the constraint. An operator who sets the agents up, points them at a real workflow, and charges monthly to keep them running is the constraint, and that operator is an agency in everything but name.

The Koerner Office pitch episode, recorded in front of a live audience, supplies the agency economics from a business that predates the AI version of the model. A marketing agency for home-service businesses reported roughly $1.5M in annual recurring revenue at an average of $450 per customer per month, with 78% net revenue retention at 12 months. The founder built it alongside a full-time job for nearly half of its three years. The design is deliberate: a low ticket, spread across enough clients that no single one matters much. That is the template the AI operators are copying, with agent labor swapped in for some of the human labor behind the same monthly fee.

The same episode underlines what buyers reward. The room's biggest earner was a veterans' massage business running about 750 sessions a month for $68,000 a month in revenue and roughly $20,000 a month in profit. Those figures are self-reported pitch claims, and the outline flags them as such. Still, the ordering is instructive: a plain services business with paying customers out-earned every idea-stage pitch in the room.

What the buyer is actually paying for

Across the three sources, the purchase has the same four parts:

• A named outcome with a dollar value attached: a recovered $1,000 job, a store-ready product, booked work from marketing that runs every month.
• One setup conversation. The owner states their rules, prices, and boundaries once; the operator absorbs everything after that.
• A monthly fee for keeping it running, priced against a salary or an existing agency bill rather than against software seats.
• No new behavior from the owner. No dashboard to check, no prompts to write, no model settings to learn.

The corollary is that the AI itself is close to unsellable to this buyer. The receptionist entry leads with call-answer rates and a salary comparison, not model quality. The food brand entry leads with shelf access. The tool stays backstage because the buyer never asked for a tool; they asked for the shift covered and the phone answered.

The honest limits

Low ticket cuts both ways. At $299 or $450 a month, an operator needs dozens of clients before the business pays a salary, and the Koerner Office agency's 78% net revenue retention means roughly a fifth of the revenue base has to be replaced every year just to stand still. The receptionist math also depends on call volume: a contractor who misses two calls a month is not losing $1,000 jobs at a rate that supports any fee. And the strongest numbers here, the massage business revenue and the agency's recurring revenue, are claims made on a pitch stage, not audited results. The pattern is consistent across independent sources; the individual figures should be treated as directional.

What this means for a solo builder

• Sell the outcome in the buyer's unit: recovered jobs, answered calls, shelf-ready products. Price against the salary or invoice it replaces.
• Charge a setup fee. The $500 and $1,999 data points both filter for buyers with a real problem and fund the configuration work the service model depends on.
• Keep the monthly fee boring and legible. $299 to $450 a month is a line item a small business already understands from its agency and bookkeeping bills.
• Take the first clients in person. The food brand playbook starts with a walk down a grocery aisle, and the home-services agency was built part-time over three years. Local trust is the distribution channel at this ticket size.
• Expect to be called an agency, and accept it. The retention economics, the setup labor, and the monthly fee all say agency. The margin advantage comes from how much of the delivery the agents carry, not from pretending the service is software.

What would change this conclusion

If small businesses start buying AI tools directly and running them without an operator, the agency layer becomes a temporary scaffold and the margin moves to the platforms. The install numbers show the tools are already in their hands; the usage pattern, email drafting and little else, says the operator is still the missing piece. Until owners demonstrably run the workflows themselves, the product worth building for Main Street is the agency, with AI doing the work behind it.

All articles · Research home