Voice agents
Start with the missed call. Build the talking agent where the numbers make sense.
Calls to a Zimbabwean mobile cost real money per minute and there is no local Twilio number to answer on. So stage one turns every missed call into a WhatsApp conversation within a minute. Stage two puts a voice agent on a number that can actually be routed.
The Zimbabwean situation
What usually goes wrong
The phone is a mobile in someone's pocket. It rings during a delivery, at lunch, and after 17:00, and the caller does not leave a message; they call the next shop. Nobody knows how many calls were missed because nobody counts.
Two Zimbabwe-specific facts shape the fix. First, Twilio, the usual telephony layer for voice agents, offers no voice-enabled Zimbabwe numbers, and charges USD 0.8601 per minute to call a Zimbabwean mobile. Second, the same caller almost certainly has WhatsApp, and POTRAZ's data shows it is the most used platform in the country. The economical move is to answer the missed call on WhatsApp.
What we build
What gets built
- Stage 1 — missed-call recovery. Busy/no-answer forwarding or a call-log check triggers a WhatsApp template within a minute: "We missed your call. What do you need?" The conversation captures the need and books, quotes or routes it, exactly like the WhatsApp automation.
- Stage 2 — inbound voice agent. On a SIP-routable number: disclosure and consent, answers to the top questions, booking against the calendar, and transfer to a person for anything else. Transcript and summary to the CRM.
- Outbound reminders, selectively. Reminder calls only where a WhatsApp reminder failed and the appointment value justifies the per-minute cost.
- Counting. Missed, recovered, booked, and cost per handled call, reported monthly.
How it works
The flow, end to end
What it needs from you
- Which number, who owns the SIM, and whether it can forward on busy/no-answer.
- One week of call logs (or just start counting: we can do it for you in the pilot).
- Business hours and who takes the transferred calls.
- Your disclosure and recording-consent wording, or agreement to use ours.
- For stage two: a provider that can route the number over SIP, or the willingness to get one.
Timeline shape
| Phase | Typical span | What happens |
|---|---|---|
| Scoping | 1 week | Count missed calls for a week, check what your number and provider can do, decide stage one or two. |
| Stage 1 build | 1–2 weeks | Missed-call detection, WhatsApp template within a minute, capture of the need, booking or routing. |
| Stage 2 build | 3–5 weeks | SIP routing, voice agent scripts, disclosure and consent, transcript to CRM, escalation to a person. |
| Pilot | 2 weeks | Live on a subset of calls (after hours first), every transcript reviewed. |
| Run | ongoing | Monthly review of failed calls, cost per handled call, script changes. |
Spans are typical shapes, not quotes. The process page explains what moves them.
What it costs
What drives the price
Stage one is priced like WhatsApp automation. Stage two adds per-minute telephony and speech costs that scale with call length and volume, plus the routing contract. We show the per-call estimate before you commit.
We do not publish package prices because the drivers below move the number by multiples. The pricing page lists every driver with sourced platform costs.
| Driver | Effect on cost |
|---|---|
| Per-minute telephony | Carrier fees per minute (Twilio: USD 0.405 landline / 0.8601 mobile to Zimbabwe) times call length times volume. |
| Speech and model minutes | Speech-to-text, model and text-to-speech are billed per minute or token; long calls cost more than short ones. |
| Number and routing | Reusing an existing forwardable mobile is cheapest; a SIP-routable line from a local provider is a separate contract. |
| Languages and accents | English only is standard; testing and tuning for Shona/Ndebele adds pilot time. |
| Recording retention | Storage and access controls for call recordings, set by your data-controller obligations. |
Questions we get asked
Can we get a Zimbabwe (+263) number for the voice agent?
Not from Twilio: its Zimbabwe page states it has no voice-enabled local numbers, only routes to reach Zimbabwe from numbers in other countries. A local number means working with a licensed Zimbabwean provider that can hand calls to us over SIP. We check what your current number and provider can do at scoping, and we do not promise a +263 number until it is confirmed.
What does a voice call cost to run?
Carrier fees alone: Twilio lists outbound calls to Zimbabwe at USD 0.405 per minute to landlines and USD 0.8601 per minute to mobiles. Speech-to-text, the model and text-to-speech add a few cents a minute on top. A three-minute mobile call is therefore roughly USD 2.60 in carrier fees before the AI, which is why missed-call-to-WhatsApp is the first stage for most businesses.
Will it understand callers speaking Shona or Ndebele?
English is reliable; Shona and Ndebele speech recognition is weaker with current models and we test it with recordings of your real callers before promising anything. The recovery path is always there: a caller the agent cannot understand is offered a WhatsApp follow-up or a human callback.
Do we have to tell callers they are talking to an AI, or being recorded?
We disclose both at the start of every call. Recording and transcription create personal data under the Cyber and Data Protection Act, so the consent line and your retention period are set at scoping.
What if our line is a mobile phone with no PBX?
That is the common case. Stage one works with plain call forwarding on busy/no-answer and a call-log check; stage two needs a number that can be routed over SIP, which may mean a new line.
Next step
Scope it before you book it
Fill in the scoping worksheet (five minutes, no sign-up). It produces a summary you can paste into WhatsApp or email, so the first call starts with the actual work.