The Next Wave of AI Companies: Selling Results, Not Tools

Deep thoughts on AI and aspirations —— ByteDance Deep Thinking Circle

There’s a company in the US called Zoca doing something that sounds almost audacious: it serves local businesses like nail salons, massage parlors, and fitness studios, charging no subscription fees—only getting paid when it actually brings in bookings. According to publicly available information, one nail salon owner went from three or four clients a day to a waitlist and turning customers away after 30 days. Since launching in 2024, Zoca has served over 1,000 local businesses, completed 120,000 bookings, and generated over $10 million in revenue.

These numbers need more time to validate, but the direction it’s betting on deserves serious attention: the AI industry’s business model is quietly shifting from “selling tools” to “guaranteeing results.”

What Really Crushes Small Businesses Is No One Taking Responsibility for Outcomes

First, consider what these businesses are actually selling. Local service businesses sell time: an empty afternoon for a hairdresser is gone forever once it passes. You can’t stockpile it, make it up later—every unsold minute is permanent loss. So marketing efficiency is an existential matter for them.

But when you calculate time and energy, they’re precisely the ones least equipped to do marketing. Fiverr research shows 70% of small businesses spend no more than 5 hours per week on sales and marketing. It’s not that they don’t know it’s important—a hairdresser working ten to twelve hours a day serving clients simply doesn’t have time to learn SEO, social media management, and ad placement. And the toolkit on the market—booking systems, customer management, social media scheduling, ad platforms—each requires learning. Mastering one suite is practically like picking up a second trade.

So the traditional tool model has an implicit message: I’ve taken your money, the results are your responsibility. The tool company’s revenue has nothing to do with whether the shop survives. The risk is pushed entirely onto the party least capable of bearing it—small shops pay up, time passes, and if there are no results, they can only blame their own bad luck.

“Guaranteeing Results” Really Means Risk Transfer

Zoca’s model flips this implicit message: it doesn’t promise features, it promises bookings; no bookings, no charge.

This change appears to be just a different billing method, but it actually transfers risk from buyer to seller. The tool model is “I’ll rent you the fishing rod, whether you catch fish is your business”; guaranteeing results is “I promise you a quantity, no catch means no charge.” The latter means the service provider must swallow the uncertainty of delivery themselves—only someone confident enough in their own system would dare play this way.

The incentive structure changes completely as well:

When the service provider’s revenue directly equals the client’s results, both sides’ interests align for the first time. The provider won’t pile on shiny but problem-solving-free features anymore, because piling them on doesn’t generate bookings.

There’s an easily overlooked prerequisite here: daring to guarantee results is itself an expression of technical confidence. Performance-based pricing means every failed delivery is your own cost—this business only stands when the system’s success rate is high enough to cover those failures. So the guaranteed-results model naturally weeds out players who can only demo but can’t deliver reliably, leaving the market to those who’ve actually made the closed loop work.

Why This Bet Only Makes Sense Now

Guaranteeing results isn’t new. Ad agencies doing performance splits, sales commissions—they’ve been around for decades. Why is it only now, in the AI industry, that it might become a model?

Because before AI appeared, guaranteeing results was often a losing proposition. Human delivery costs were high, non-standardized, non-replicable—success with one client might depend on a star salesperson; taking on another meant losing on another.

AI agents change the cost structure: demand monitoring, content generation, ad adjustments, pricing recommendations, customer follow-up—all can execute automatically 24/7. Serving 1,000 shops versus 100 shops has minimal marginal cost difference. Zoca’s system claims to capture neighborhood-level demand fluctuations—when demand for a certain service in a certain area rises, it automatically adjusts the shop’s exposure and pricing; raise prices when demand is hot, run promotions when it’s cold to fill gaps. Whether this execution capability can ultimately deliver remains to be seen, but the direction is clear: only when delivery costs drop to a certain level does “pay for results” have positive margin for the first time.

So I prefer to understand it this way: this isn’t business model innovation, it’s the first commercial realization of AI’s cost structure. The tool model is how software should be sold in the high-human-cost era; guaranteeing results is how AI services should look in the era when delivery costs trend toward zero.

But Guaranteeing Results Isn’t a Shield

The dark side of this model needs to be spelled out clearly.

First, results can be faked. Bookings don’t equal show-ups; low-price lead generation can inflate booking numbers. If the system only pursues booking quantity, it might fill a shop’s entire day with low-quality customers—the shop is busy but hasn’t made money. When choosing the guaranteed-results model, you must ask by what metric “results” are counted.

Second, individual cases don’t prove universal applicability. Single-shop explosive growth without control groups might be survivorship bias; company-stated revenue and booking numbers are one-sided claims. How much net growth is truly delivered to merchants needs longer observation.

Third, dependency risk. When a shop’s customer acquisition is entirely tied to one AI system, an algorithm tweak, platform policy change, or system failure hits directly at the shop’s lifeline. This risk doesn’t exist in the tool model—if the tool fails, people can step in; if the system stops, the customer source is cut off.

So, I offer four prerequisites for the guaranteed-results model: results are measurable and can distinguish real value; delivery’s marginal cost is low enough; fakery can be identified; customer repeat purchase is high enough. Missing one of these four prerequisites, “guaranteeing results” isn’t a business model—it’s a gamble.

Assessment

The real significance of this case isn’t another company’s funding story, but a competitive direction quietly changing: the AI services industry is moving from feature competition to results competition. Whoever can bear risk, whoever can prove their delivery, gets pricing power.

If you’re building vertical AI, ask yourself before launch: do I dare guarantee results to customers? If yes, it means closed-loop delivery, data feedback, calculable costs are all working; if no, they probably haven’t—so go back and get the loop running before rushing to scale.

The same logic applies for buyers: when selecting AI service providers, rather than looking at feature lists, see if they dare tie payment to results. Those who dare to link them at least show confidence in their delivery; those who only want subscription fees while being vague about effectiveness probably lack confidence themselves.

The next wave of AI companies will split into two types: those still selling tools, and those daring to guarantee results. The latter are rare now, but nearly all have touched the true pulse of AI’s cost structure.