What is Voice agent?
A system that answers the phone and completes the transaction. Materially harder than a chat widget, and the failure modes are specific.
Two things make voice a different engineering problem. The first is that a misheard digit across an authentication boundary does not degrade the answer, it ends the call - and the customer restarts with a person, which is worse than never having offered the automated route. First-pass capture of spoken numbers is therefore the metric that decides whether a voice agent is deployable: on a delivered system it went from 84% to 98.6%.
The second is dialect. Gulf Arabic is not one accent, and a system that performs well on average can perform badly on one region's speech - which means it has failed for a whole governorate while the dashboard looks fine. On a delivered transcription system, word error rate went from 31% to 12.7% and the spread between the best and worst dialect narrowed from 19 points to 5. The second number is the harder one and the one to ask about.
| On this | Voice agent | A chat widget on the website |
|---|---|---|
| Who it reaches | The caller who is on a plant, in a vehicle or out of hours | The person already at a browser |
| Cost of a mistake | The call ends and the customer is worse off than before | The user rephrases |
| What has to be measured | First-pass number capture, and error spread across dialects | Containment rate |
- When it is the right answer
- When a high volume of repetitive enquiries can be answered from a system of record and the caller is often not at a desk.
- The service line that delivers this
- AI Voice Agents
The caller has to read out an account number, and if the system mishears one digit the whole call is wasted.
Start here
If a term here is the one your board paper turns on, ask us about it.
We will send the entry, the evidence behind it, and the honest note about where it does not apply.
You get a reply within one working day, from the engineer who would do the work - not a sales sequence.