AI Voice Agent and AI Receptionist Development UK.

What an AI voice agent does on a real phone line

An AI voice agent answers a phone call, understands what the caller wants, and either deals with it or passes it to the right person. For most UK small businesses the useful jobs are unglamorous:

• Answering opening hours, prices you publish, directions and other common questions.
• Taking a proper message, with name, number, reason for calling and urgency, and emailing it to the right person.
• Booking, moving or cancelling appointments in your real diary.
• Checking an order or job status from your system.
• Routing: sales to sales, accounts to accounts, emergencies to the on-call mobile.

It runs out of hours and at busy times, and the caller is never on hold to reach it. It should not pretend to be a person. We recommend the greeting says clearly that the caller is speaking to an automated assistant and how to reach a human.

How the technology fits together

There are three layers, and we choose each one for your situation.

Telephony. Your existing number is pointed at the AI through a SIP trunk or a programmable voice platform. Twilio's documentation describes two routes: ConversationRelay, where Twilio does the speech recognition and text-to-speech and sends your server text over a WebSocket, and Media Streams, which streams raw call audio to your application so you bring your own speech services.

Speech and reasoning. Either a single realtime speech-to-speech model, or a pipeline of speech-to-text, a language model and text-to-speech. OpenAI announced general availability of its Realtime API, including SIP phone calling, in August 2025, and its current telephony guide also covers outbound calls. Realtime models feel more natural; pipelines give more control over voice, cost and logging. We test both with your real call types, including regional accents and noisy lines.

Your systems. The agent's tools connect to your booking system, CRM or helpdesk through their APIs. This is the part off-the-shelf AI receptionists usually lack, and it is where we spend most of our time.

Handing over to a human

The handover is what makes callers trust the system. We design it as carefully as the AI itself:

• Clear triggers: the caller asks for a person, the agent is not confident, the topic is a complaint or anything safety-related, or the call runs too long.
• Warm transfer: the call is passed using the telephony platform's transfer feature (OpenAI's SIP guide uses SIP REFER), and the person picking up gets a short summary so the caller does not repeat themselves.
• Fallbacks: if nobody answers, the agent takes a full message and says honestly when someone will call back.
• No dead ends: pressing a key or saying "person" always works.

After each call, a structured summary lands in your CRM or helpdesk with the transcript attached, so nothing depends on someone remembering to write it up.

UK GDPR, call recording and consent

Voice agents process personal data, sometimes sensitive data, so compliance is part of the build, not an afterthought. This is general information, not legal advice.

• Tell callers. The ICO's guidance says organisations must tell people they are recording a call and why, and suggests a recorded message as good practice, with the fuller privacy information available by other means such as your website.
• Lawful basis and retention. Recording and transcription need a documented lawful basis, often legitimate interests, and a retention period. We make retention configurable and delete audio and transcripts automatically when it expires.
• Impact assessment. AI on customer calls is new processing that can be high risk, so we prepare a draft data protection impact assessment for your review.
• Suppliers. Telephony and model providers are processors; we list where audio is processed and check their data processing terms and data retention settings.
• Outbound marketing. The ICO's PECR guidance says automated marketing calls need consent that specifically covers automated calls; consent for live calls is not enough. We build outbound voice agents for service calls such as appointment reminders, not unsolicited sales calls.

Testing, costs and running it

Before any customer hears it, we run the agent against scripted test calls covering your common requests, awkward callers, people who interrupt, and attempts to make it say things it should not. In production every call is logged with outcome, duration and model cost, and calls the agent marks as uncertain go into a review list. Per-minute telephony and model costs are paid to those providers and we estimate them during scoping. Every voice agent is a fixed-price quote after a free chat; ongoing tuning can sit under a maintenance agreement.

What we deliver

• A call flow design: greetings, intents, what the agent can do and when it must transfer
• Telephony setup on your existing numbers via SIP trunk or a provider such as Twilio
• A voice agent using a realtime speech model, or a speech-to-text, LLM and text-to-speech pipeline
• Tools to check availability, book appointments, look up orders and take messages
• Warm transfer to a person or team with a spoken or on-screen summary
• Recording and transcription notices, retention settings and a draft DPIA
• A call log with transcripts, outcomes, cost per call and flagged calls for review

How it works and what it costs

Every project gets a fixed-price quote after a free initial chat and a short scoping stage. You own the code and the data.

1. Free chat

   Tell us the problem in plain English: what you do now, what goes wrong and what "better" looks like. No charge, no obligation.

2. Scoping

   We map the processes, systems and data involved, agree what is in and out, and write it down so there are no surprises.

3. Fixed-price quote

   You get a fixed price for the agreed scope, or a phased plan for bigger builds, so you can start small and prove it works.

4. Build and test

   We build in short stages you can see and try, test against real data, then go live carefully with a rollback plan.

5. Hand over and look after

   You own the code and the data. We can host it, support it and keep improving it, or hand it to your own team.

Related