Every AI voice company demoes the same moment — the bot picks up, sounds human, handles the question. It's a good demo. It's also not the hard part. The hard part is everything that happens before it: the menu.
Every AI voice company demoes the same moment: the bot picks up, sounds human, handles the question smoothly enough that you forget it's software. It's a good demo. It's also not the hard part. The hard part happens before any of that — the moment the call connects to a machine, not a person, and the machine wants you to press a number.
What actually happens before the conversation
Call a hotel, an airline, almost any real business at scale, and you don't reach a person first. You reach an IVR — press 1 for reservations, press 2 for the front desk, press 3 to repeat these options, sometimes three menus deep before a human ever picks up. Every AI voice tool we tested handles the conversation that happens after that. Getting through the menu itself was, as far as we could find, something nobody had actually solved.
Where this problem actually came from
We didn't find this problem by researching competitors. We found it running our own re-verification calls, at real volume, for years. Before any of it was automated, re-verifying a booking meant a person — or a whole team of them — manually calling a hotel, working through whatever menu that property happened to have, and hoping the language on the other end matched someone on the team. Every new hire had to be trained on the same process. Every busy week put the same team under the same load, and load is exactly when a manual process starts quietly failing.
We identified this more than a decade ago, and built a small app as a proof of concept. The idea worked. Turning it into something that could run at scale took a great deal longer.
Why third-party tools couldn't solve it
Most AI calling companies build on top of a rented, third-party telephony layer — which is a perfectly reasonable way to launch fast, but it puts a ceiling on what you can actually automate around an IVR, because you don't control the infrastructure the call is running on. rePUSHTI built its own calling infrastructure instead. That's the unglamorous, unsexy reason IVR navigation was solvable for us: we weren't trying to bend someone else's platform into doing something it wasn't built for.
If it were easy, someone else would already have built it.
What “handles both sides” actually means
Almost all AI voice calling is built for one direction: a customer calling a business. rePUSHTI's Voice Calling handles that — and the direction almost nobody builds for, which matters enormously in travel: a business calling another business. A B2B travel agent's AI calling a hotel directly, working through that property's own IVR the same way a person would, and reaching the right department without anyone doing it by hand.
Built on a million bookings' worth of real calls
None of this was trained in a lab against synthetic data. It was refined against rePUSHTI's own operational history — more than a million bookings and two million calls, each one a real conversation with a real hotel, in whatever language that hotel actually answers in. That volume of real re-verification is what actually taught the system which menus look like what, and how to get through them.
What's live now, in beta
- Every call made, listened back to, and attached to the booking it was about
- AI summaries of each re-verification, generated automatically, no manual note-taking
- Multiple languages, so re-verification isn't limited to wherever your own team happens to speak
- Runs agentically in the background — bookings get re-verified day and night, not just during business hours
We didn't set out to build another AI caller. We set out to fix the one part every AI caller still couldn't do.
rePUSHTI is already the world's first booking re-verification platform built on AI and automation. Solving IVR navigation on top of that is the same instinct, applied one layer deeper — the small, unglamorous piece of the problem that decides whether a whole call actually completes, or quietly needs a person after all.
