We ran a recruitment campaign for a large pizza chain. The voicebot called around 600 candidates. 490 of them answered, another 30 called back, and 155 completed the full screening conversation. The whole thing took under an hour. A recruiter doing the same work by hand would have needed roughly forty.
That number — forty hours to one — is the one people repeat. I want to talk about the part nobody repeats.
The model was never the hard part
By the time we ran that campaign, speech recognition and synthesis were commodities. Anyone could buy them. What was not commoditised, and still is not, was everything around the model: knowing which candidates to call and when, handling the person who answers while driving, routing someone who asks for a human to an actual human within seconds rather than a queue, and producing a report at the end that a recruiter can act on before lunch.
None of that is artificial intelligence. All of it is telephony, workflow design and integration work. And it is where every automation project I have seen either succeeds or quietly fails.
Why this matters commercially
There is a strategic consequence here that founders in this space often miss. If your differentiation is the model, you are renting your advantage from someone with more capital than you, and the rent goes up. If your differentiation is the integration into a customer’s operating process, you own something.
The second position is less impressive to talk about and much better to be in. It is also, I think, the reason communications automation companies can have unusually low churn once they are genuinely embedded: replacing the vendor means re-running the integration project, and the person who would have to authorise that has other problems.
The mistake I see most often
Companies in our sector sell the technology because the technology is what they are proud of. Buyers do not care. A head of recruitment does not want a voicebot; they want to stop losing candidates in the three days between an application and a first call.
The pitch that works describes the problem in the customer’s language and only then explains what does the work. It is a smaller, less exciting sentence, and I think it sells better.
What I would look for now
If I were assessing a communications automation business today, I would spend very little time on the model and a great deal on three other things: how deep the integrations go, how much of the deployment is repeatable versus bespoke, and what happens on the customer’s side when something breaks at 3pm on a Friday.
The third one tells you more than the first two combined. Automation businesses are judged in production, not in demos — and production is where the durable ones separate themselves.
Denis Shirokiy is a Venture Partner at Boxwood Group and founder of Apifonica. He previously led commercial functions at MTT.