Pay Per Meeting Booked – the Model That's Hard to Copy

01 /The Concept

Throxy finds potential customers for companies – and takes the process all the way through to booked meetings with sales reps, filling up their calendars.

The platform currently serves around 50 companies, for whom it has increased the number of scheduled prospect meetings by 3.5x – generating $63M in new pipeline.

Because the deliverable is meetings, Throxy doesn't charge a subscription or a usage fee. It charges per booked meeting.

Throxy graduated from Y Combinator last spring and has already reached $1.5M in annualized revenue with a team of three. That trajectory attracted investor attention, and the company just raised $6.2M – which it plans to use partly to grow the team to 15.

02 /Why It Matters

The most interesting thing about Throxy isn't what it does – it's where it does it.

Throxy made a deliberate choice to focus on markets where potential customers are hard to find through the standard channels everyone else uses: LinkedIn, ZoomInfo, and similar lead databases.

To cover those markets, Throxy builds and maintains its own proprietary data on potential buyers. That proprietary data is how it reaches and books meetings with prospects other platforms can't.

Examples of these underserved markets: sellers of industrial equipment to factories, or logistics companies looking for distributors with consistent high-volume shipments, including import/export activity.

The key insight is that alternative data sources do exist in these niches – product compliance certification databases, importer/exporter registration records, company directorship filings. Each of these can be a proxy for a qualified buyer, if you're willing to find, extract, and interpret them.

Throxy has built that out at scale. LinkedIn contains profiles on about 1.1 billion people and companies. Throxy's database holds information on 4 billion contacts. For the UK alone, Throxy has profiled 6 million companies – most of which don't appear in any mainstream lead generation platform.

Raw data from non-standard sources is just a starting point, though. It can be incomplete, generic, or outdated. Throxy enriches and validates every contact by analyzing the company's website, news coverage, social presence, and employee activity before any outreach is sent.

A second distinguishing feature: Throxy uses human reviewers in its data collection and validation pipeline. AI agents handle the initial data gathering, but human freelancers verify and clean the output before it's treated as actionable.

Implementation takes three weeks:

- Week 1: Throxy works with the client to define the ideal customer profile.

- Week 2: Throxy builds the lead lists, using its existing database and expanding into new data sources as needed for the specific market.

- Week 3: Automated outreach begins, with early responses (declines and booked meetings alike) used to calibrate targeting and message copy.

Using humans in data work raises obvious scalability questions. Throxy's founder addresses this by pointing to the outcomes: $1.5M in ARR with three full-time employees – acquired entirely through their own platform.

The human-in-the-loop approach is unusual in this space. In the same YC cohort, 11x ([related review](/review/rynok-jeto-kogda-mozhno-shtampovat-i-nanimat)) built a fully automated AI sales agent operating on fully automated lead databases.

Which model performs better in practice is an open question. But there's a key structural difference: Throxy's pricing model is performance-based – payment for booked meetings, potentially extending to closed deals. That alignment changes everything.

A standard AI platform might send a million outreach emails, charge pennies per email, and profit from volume. Throxy's incentive is the opposite: if it sends a million emails and books one meeting, it bears the cost of a million emails and earns one small fee.

That's the economic logic behind the human verification step. Throxy uses human reviewers not to add overhead, but to maximize precision – because precision is what its business model rewards. A sharp alignment with client interests.

It's worth noting: Throxy built for markets where standard prospecting fails. That deliberate constraint – committing to a hard problem – is also what protects it. The founders who want a clean AI-only solution will build something else. The founders who go after the same problem will face higher barriers to matching Throxy's data depth.

03 /Opportunities

The general direction is building AI platforms in areas where "pure" AI alone can't yet deliver the quantity or quality of results needed – and where closing that gap requires human judgment in the loop.

This approach does two things at once. It filters out the large pool of founders who are philosophically committed to fully automated systems. And it lets you outrun the ones who do go after the same problem, by setting a higher bar on output quality that takes time and effort to replicate.

The model appears across domains.

GrowthX ([related review](/review/novaja-biznes-model-dlja-bystrogo-i-pribylnogo-rosta)) built an SEO platform under the premise: "AI can't do it alone, experts don't scale – so combine them." Human strategists define the brief with clients, AI generates the content, human editors review and polish. Clients saw 10–20x organic traffic growth in a short time frame by SEO standards. Revenue went from zero to $1M ARR in nine months. GrowthX raised $12M in its first institutional round in May.

Crosby ([related review](/review/dlja-odnih-jeto-povod-dlja-rasstrojstva-a-dlja-tebja-sposob-zarabotat)) raised $5.8M in its first round in June for a legal AI platform where AI generates legal documents on demand – but licensed attorneys review, verify, and take final responsibility for the output.

The question worth answering for your own startup: in which domain, for which customer segment, and for which types of tasks does involving humans in the AI workflow lead to meaningfully better results – and leave enough competitors behind to make it worth building?