The UK has no specialist list of conversational AI companies for ISPs and telecoms, so this guide is one. Providers range from enterprise voice AI vendors and contact-centre integrators to build-and-run agencies like Fiveleaf, and the right one depends on your size, your systems and who will run the agent after launch.
We wrote it, and we are in it. We have tried to be fair to everyone else, and every claim about another company comes from its own site or a public source.
What does an ISP actually need from conversational AI?
A general chatbot answers questions from a help centre. An ISP's customers rarely want that. They want to know why their line is down, when the engineer is coming, and why the bill went up. A useful agent for broadband or telecoms does these things:
Account answers after verification. Most valuable questions are about the customer's own account: their bill, their contract end date, their order. The agent has to verify who it is talking to, then read from your BSS or CRM. Without that it can only quote the FAQ.
Faults and outages. When an exchange or a section of network goes down, contacts spike within minutes. The agent should know about the incident, tell affected customers so straight away, and stop them queuing for a person who will say the same thing. We wrote about this in handling the call spike in a broadband outage.
Installs and appointments. Booking, moving and confirming engineer visits, and knowing what happens when one is missed. Missed appointments carry automatic compensation, covered in missed engineer appointment compensation.
Billing. Price rises, failed payments, why this month is different. High volume and mostly repetitive, so this is where an agent pays its way.
One Touch Switch leavers. Since One Touch Switching, the gaining provider starts the switch and the losing provider hears late. An agent can contact every leaver, ask why, make an approved offer and pass the ones worth saving to a person, without ever getting in the way of the switch. More in One Touch Switching and retention.
Ofcom's complaint and vulnerability rules. The agent has to recognise a complaint when the customer does not use the word, follow your vulnerability policy, and never stand between a customer and a person. See what Ofcom says about AI chatbots in telecoms.
Handover to people. A clean handover carries the whole conversation, so nobody asks the customer to repeat themselves.
WhatsApp as well as web chat and voice. Many customers would rather message than call, and WhatsApp keeps the thread on their phone. Costs work differently from web chat, set out in WhatsApp Business API pricing in the UK.
What types of provider are there?
There are five, and each has trade-offs.
Enterprise conversational AI vendors. Large platforms with their own models, tooling and enterprise sales teams. They handle scale and security reviews well. They are built for big operators, so contracts, timelines and minimum volumes can be heavy for a regional altnet.
Voice AI specialists. Companies focused on agents that answer the phone. If your volume is mostly calls, they go deep on speech, latency and telephony. Check how they cover chat and WhatsApp, and who does the integration work.
Contact-centre consultancies and integrators. Firms that design and run whole contact centres, often on top of the large cloud contact-centre platforms. A good fit if you are replacing your telephony or workforce tools at the same time. AI may be one strand of a larger programme, with the pace and price that implies.
Build-and-run agencies. Smaller teams that design the agent, connect it to your systems, then keep running and improving it for a monthly fee. Faster to start and closer to the work. The questions to press are continuity (what happens if a key person leaves) and whether they have done it in telecoms before.
In-house. Your own team builds on an off-the-shelf platform. You keep full control and the knowledge stays with you. It works if you have people who can own it after launch, because an agent is never finished: prices, products and network incidents change every week. The cost people miss is the second year, when the person who built it has moved on and nobody knows why the agent says what it says.
By size, roughly. A national provider with a large contact centre usually looks at enterprise vendors and integrators first. A regional ISP or altnet with tens of thousands of customers often gets more from a build-and-run agency or a focused in-house team, because the agent needs to fit its systems more than it needs a platform. A small ISP should start with one channel and one job, such as out-of-hours support or leavers.
Who builds conversational AI for telecoms in the UK?
These are the providers we found with genuine telecoms or contact-centre relevance. Each description comes from the company's own site or a public source. We have left out anything we could not check, including prices and company sizes.
Last checked: 9 November 2026.
Fiveleaf. We wrote this guide. Fiveleaf is a London-based team that designs, builds and runs AI chatbots and voice agents, mostly for UK ISPs, altnets and telecoms, on web chat, WhatsApp and voice. We stay on after go-live and run the agent monthly. For a UK fibre ISP, the cancellation save rate is 71% of customers coming through our channel. That work showed us where ISP agents break: the account lookup, the outage nobody told the bot about, and the customer who wanted a person three messages ago.
PolyAI. A voice AI company founded in London in 2017. It builds voice agents for enterprise customer service, and its telecoms page covers identification and verification, engineer appointments, troubleshooting and account questions.
Sabio. A customer experience and contact-centre specialist. Its conversational AI work handles routine enquiries over the phone before passing callers to a person, and its site carries customer stories from telecoms operators.
Content Guru. A customer experience and contact-centre-as-a-service provider headquartered in Bracknell, with conversational AI and self-service among its products. It lists tech, media and telecoms as an industry it serves and names Vodafone as a client.
Netcall. An AIM-listed company based in Bedford, according to public market data. Its Liberty platform includes Liberty Converse, which it calls an AI-powered contact centre, plus AI virtual assistants and chatbots. The customers on its site include NHS trusts, councils, Transport for London and Legal & General.
Worktual. An AI contact-centre platform aimed at UK businesses, covering voice, chat, email and social. Its virtual agent handles common questions and routes harder ones to a person with the context kept. It has a page for telecommunications.
EBI.AI. A UK conversational AI company that provides AI assistants for web chat and the phone, from self-service up to a fully managed service. Its public work is mostly with local councils, which makes it a useful comparison for high-volume public contact.
The Bot Forge. A UK-based AI consultancy that builds chatbots and AI agents that connect to existing systems. It lists a Google Cloud partner accreditation for contact centre AI.
What should you ask any provider?
How will it connect to our OSS, BSS and CRM? Ask what they have connected to before and how. An agent that cannot see the account, the order or the fault can only deflect. Ask who builds the connection and what they need from your developers, since that is usually what sets the timeline.
Who runs it after go-live? Someone has to update the agent when prices change, read the conversations it got wrong, and fix them. Find out whether that is you, them, or nobody, and what it costs.
How does it handle Ofcom complaints and vulnerability? Ask them to show you, in a test conversation, the agent recognising a complaint that was never called one, and handing over a customer who mentions a bereavement. Remember that compliance stays your responsibility whoever builds it.
Where is the data held and processed? Ask for the data location, the subprocessors and the retention period, and whether they will sign your data processing agreement.
Can we speak to another ISP? A reference from a provider with similar systems and customers tells you more than any demo.
Do you measure resolution or containment? Containment counts conversations that never reached a person, including the ones where the customer gave up. Resolution counts the ones where the customer got what they came for. Ask which they report and how. Our view on the numbers is in what is a good containment rate.
What does leaving look like? Ask who owns the build, the knowledge base and the conversation history if you part ways.
Where to start
Pick one job that is high volume and well understood: out-of-hours support, outage messages, or leavers. Write down the conversations it must handle and the ones it must hand over before anyone builds anything. Agree how you will measure it, then prove it on that one job before adding the next.
Whichever provider you choose, put the same test conversations through each shortlisted agent: a customer who cannot get online, one whose engineer did not turn up, one who is leaving, and one who mentions they are struggling to pay. You will see the differences within an hour. If you want to see how an agent handles an ISP's conversations, you can try our agent or talk to us about yours.
