AI Voice Agents
A phone line that answers every call, day or night, and sounds like a person.
What is AI voice agents?
An AI voice agent is an automated telephone system that holds a natural spoken conversation with a caller, answers common questions, captures their details, books appointments, and transfers to a member of staff with a summary of what was already discussed.
A missed call is usually a lost customer. Most small businesses miss them at lunch, after closing, and during any period when everybody is already occupied with somebody else.
A voice agent answers every time. It handles the routine questions, takes the caller details, books the appointment if there is one to book, and sends you the message immediately.
It works in English and Swahili, including callers who move between the two mid-sentence. Calls it cannot handle transfer to a person, with a short summary attached so nobody has to ask the caller to start again.
Why callers are told at the start
Duara deploys voice agents from Nairobi, most often for appointment businesses and service companies losing bookings to unanswered phones.
We recommend telling callers at the start that they are speaking to an automated system. Some clients resist this, believing it reduces engagement. In our experience the opposite is true: callers who discover it midway feel deceived, and that is a harder problem than a slightly shorter call.
Who this is for
- You miss calls during busy periods or after hours
- A call centre costs more than the calls are worth
- Most of your bookings arrive by phone
- Reception is one person who also does other work
Industries we serve
- Clinics and dental practices
- Garages and workshops
- Salons and spas
- Property management
- Logistics and courier
- Professional services
Problems this solves
Voice work is usually commissioned after somebody counts the missed calls for the first time.
- Calls missed at predictable times. Lunch, evenings and weekends account for most unanswered calls, and those callers rarely ring back.
- A call centre that costs more than it returns. Staffing a phone line properly costs more than the enquiries justify at small and medium volumes.
- Callers repeating themselves. When a call finally reaches a person, the caller explains everything again because nothing was captured earlier.
- Reception doing three jobs. The person answering the phone is also serving customers in front of them, so one of the two suffers.
What we deliver
A phone number that answers every call in English or Swahili, handles the routine enquiries, books into your calendar, captures details, and transfers to a person with context attached.
A phone number that answers every call, at any hour
English and Swahili, including mixed conversation
Caller details captured and sent to you after each call
Appointments booked directly into your existing calendar
Transfer to a person with a summary of the call so far
Recordings and transcripts of every conversation
Features
- Natural spoken conversation. The agent handles interruptions, corrections and callers who change their mind partway through a sentence.
- Bilingual by default. English and Swahili in the same call, because that is how people in Kenya actually speak on the phone.
- Live calendar awareness. It offers only genuinely free slots by reading your calendar, so double bookings do not happen.
- Context-carrying transfer. When a person takes over they receive what has already been said rather than a cold line.
- Full transcripts. Every call is recorded and transcribed, which is useful for training and for resolving disputes.
Benefits
- No call goes unanswered. The lunchtime and after-hours calls that currently vanish become captured enquiries.
- Bookings continue overnight. Appointments get made while the business is closed, which is when a good share of customers are free to call.
- Reception stops being interrupted. The person serving customers in front of them is no longer pulled away by the phone.
- You can hear what customers ask. Transcripts show the real questions, which frequently differ from what the business assumes they are.
Technologies we use
- Speech recognition tuned for local accents. Recognition tested against Kenyan English and Swahili rather than assumed to work from generic benchmarks.
- Low-latency response. The gap before the agent replies is engineered down, because delay is what makes a voice system feel artificial.
- Telephony integration. Connection to your existing number so callers dial what they already know.
- Calendar connection. Live availability from Google Calendar, Outlook or your booking system, rather than a separate diary.
- Transcription and storage. Recordings and searchable transcripts retained under your own data rules.
Voice quality and latency matter more than model sophistication here. A slow reply feels wrong on a phone call in a way it does not in chat.
How the process works, step by step
A voice agent usually takes four to six weeks, with most of that spent tuning recognition against real calls.
-
Step 1
Free call review
We listen to a sample of your recorded calls, or discuss the common ones, to see what proportion could be handled.
-
Step 2
Script and boundary design
We agree what the agent handles, what it must transfer, and exactly how it introduces itself.
-
Step 3
Recognition tuning
We test against real callers and accents, adjusting until recognition is reliable rather than merely demonstrable.
-
Step 4
Parallel running
It answers overflow calls first while your team monitors, before taking the main line.
-
Step 5
Full deployment
It takes the number, with transcripts reviewed weekly for the first month.
Quality assurance and how we work
The commitments specific to putting a machine on your phone line.
- Callers are told at the start. We recommend and default to disclosure. Discovering it midway feels like deception and costs more than the call saved.
- Transfer is always available. Asking for a person works at any point in the conversation, including by simply saying so.
- Recognition is tested on real accents. We tune against your actual callers rather than relying on vendor benchmarks recorded elsewhere.
- It never books a slot it cannot see. Availability comes from your live calendar, so the agent cannot promise a time that is already taken.
- Recordings follow your data rules. Retention and access match your policy, and recordings are never used to train public models.
Frequently asked questions about AI voice agents
Will callers know it is not a person?
Many will not, but we recommend and default to telling them at the start. Callers who discover it midway feel deceived, which is harder to fix than a shorter call.
What happens if it cannot help?
It transfers to whoever you nominate and passes on what the caller already said, so nobody has to repeat themselves.
Does the voice agent speak Swahili?
Yes, including callers switching between English and Swahili mid-sentence, which is how most people actually speak on the phone.
Can it book appointments?
Yes, reading availability live from your calendar so it only ever offers slots that are genuinely free.
What diary does it book into?
The common ones including Google Calendar and Outlook, plus most booking systems. We confirm yours before starting.
Could a phone booking clash with an existing one?
No. It reads your live calendar rather than a copy, so it cannot offer a time that has already gone.
How long before the voice agent takes calls?
Four to six weeks, most of it spent tuning speech recognition against your actual callers rather than building.
Do we keep our existing phone number?
Yes. It connects to the number your customers already know, so nothing changes for them.
How good is recognition with local accents?
That is exactly what the tuning phase addresses. We test against your real callers rather than relying on benchmarks recorded elsewhere.
What if the caller interrupts?
The agent handles being interrupted mid-sentence without losing the thread, which is one of the main differences from an old menu system.
Can it recognise repeat customers?
Yes, by number, so it can greet them by name using details you already hold.
Is every voice conversation recorded?
Yes, with transcripts. Retention follows your own data policy and recordings are never used to train public models.
What does a voice agent cost per minute?
It depends on call volume and minutes. We give you the monthly figure at realistic volume before you commit.
Is this the same as an IVR menu?
No. An IVR gives keypad options. A voice agent holds a conversation and can complete the task rather than routing it.
What happens during a power or network outage?
Calls fall back to your existing arrangement, which we configure deliberately rather than leaving to chance.
Can it take payments over the phone?
We do not build that into the voice agent. Payment details spoken aloud create risk we would rather avoid, and we say so.
How do we monitor quality?
Transcripts are reviewed weekly for the first month. After that, spot checks and the transfer rate tell you most of what you need.
Can it take many calls simultaneously?
Yes, which is the main advantage over a single receptionist during a busy period.
What if we only miss a few calls?
Then it may not be worth it, and the free call review exists to establish that before you spend anything.
Does it work for outbound calls too?
That is scoped separately and we are cautious about it, because unsolicited automated calling carries reputational and regulatory risk.
Glossary of terms
Voice terminology in plain terms.
- Voice agent
- An automated system that holds a spoken telephone conversation, as distinct from a menu of keypad options.
- Latency
- The delay before the system responds. On a phone call, anything noticeable feels unnatural.
- Speech recognition
- Converting spoken words into text the system can act on. Accent and line quality affect it heavily.
- Barge-in
- A caller interrupting the system mid-sentence, which a good agent handles without losing the thread.
- Warm transfer
- Passing a call to a person along with the context of what has already been discussed.
- IVR
- The older press-one-for-sales menu system, which a voice agent replaces rather than extends.
Related services in AI Implementation
Talk to us about AI voice agents
Count your missed calls for one week before calling us. That number usually decides whether this is worth doing far better than any pitch could.
Start your project