Skrypt HealthA patient calls to move a cleaning. She has called this clinic for years. She knows appointments is 1, so she presses 1 before the greeting finishes. The next caller has never phoned before, has a question that fits no menu option, and just wants to explain it.
Skrypt's first plan served only the second caller. We pitched open speech as a replacement for the clinic IVR. Clinic teams pushed back: plenty of their patients like the menu. So we kept it, added open speech next to it, and put the same AI behind both. Given the choice, 52% of callers spoke and 48% pressed a key.
The takeaway is simple: neither group is an edge case. If we had shipped our first idea, it would have failed almost half of callers.
How the feature came about: we listened, and changed course.
Every clinic we met already had an IVR. We saw it as the thing standing between the patient and a real answer, so we planned to remove it.
Front-desk teams answer these calls every day. More than one of them told us that regular patients rely on the menu because it is fast and predictable.
A hybrid greeting: callers press a number or just speak. The same AI answers, books and hands off, whichever way they start.
The caller's first input after the hybrid greeting. If the first input was speech, the call counts as spoke. If it was a keypress, it counts as keypad.
Inbound patient calls from Skrypt's production records at clinics running the hybrid greeting. We removed test calls, duplicate records, and calls that ended before any input.
Both paths lead to the same AI, with the same booking rules and the same hand-off to staff. Callers were not nudged toward either option; the greeting offers both in one sentence.
Clinic names are withheld. Results are reported as a single split across all participating clinics.
4.1 The split: almost exactly half and half.
Given the choice at the greeting, 52% of callers started by speaking and 48% started by pressing a key. Neither option was a fallback. Each was the first choice for roughly one caller in two.
4.2 Two ways in, one way through.
The caller picks how to start. After that, the experience is the same: the AI understands the request, books directly into the clinic's schedule, and passes anything it shouldn't handle to a person on the team.
| Path | Share | What the caller does | Why clinics said it matters |
|---|---|---|---|
| Speech | 52% | Explains the request in their own words | New patients and unusual questions that fit no menu option |
| Keypad | 48% | Presses the option they already know | Regular patients who want the fastest, most predictable route |
It is easy for a voice-AI company to treat the phone menu as the problem. We did. The people who corrected us were the ones closest to the calls: front-desk teams who hear every day which patients press 1 and which ones want to talk. A near-even split means both groups are large. A product built for only one of them would have made the other group's calls slower or more frustrating, and the clinic would have heard about it first.
That is the lesson we are taking into the rest of the product. Skrypt is built in clinics, with clinic teams, and the best changes often start as a correction.
- 01Pooled across clinics. The split is one figure across all participating clinics. Individual clinics, patient mixes and regions will differ.
- 02Preference, not outcome. This study reports which path callers chose. It does not compare booking or resolution rates between the two paths.