Voice AI for Insurance Broker Renewals: A 2026 Guide

Renewal season is the most predictable capacity crisis in UK broking. The book renews on a calendar the broker did not choose, mid-term adjustments arrive at random, and every conversation carries a regulatory payload a general contact centre script does not. The British Insurance Brokers' Association represents more than 1,700 regulated firms employing over 130,000 people, and much of that headcount spends the peaks confirming nothing has changed, capturing what has, and explaining an unexpected premium movement.

Voice AI is an obvious fit for the first, a conditional fit for the second, and a bad fit for the third. That distinction matters more in insurance than almost anywhere else, because the renewal conversation sits inside a rulebook rewritten specifically to stop firms quietly profiting from customers who do not shop around. Get the boundary wrong and you scale a compliance problem across the book.

This guide is shipped by the team behind Dilr Voice, enterprise voice AI built for regulated deployments. Or see DATS, our five-stage AI consulting system.

Why do broker renewal calls break down at scale?

Broker renewal calls break down because demand is spiky and the work is not uniform. A renewal book concentrates into peaks, so a team comfortable in a quiet fortnight is overwhelmed in the fortnight that matters, and what gets squeezed is the low-value confirmation rather than the complex case. Dilr Voice is generally deployed against that first category, where the volume sits and the judgement does not.

The problem is rarely headcount but a mismatch between when demand arrives and when capacity exists. An account handler spending forty minutes of a peak confirming that a fleet has not changed size is not doing the work only they can do. What broker teams lack is a safe way to separate the two streams before the call is answered, the same problem covered in our analysis of blended human and AI staffing at peak.

The second failure is data. Renewal quality depends on whether the broker management system reflects what the client has, and that record decays with every adjustment handled by voicemail, email or a note nobody rekeyed.

Which renewal rules bind a broker, and which do not?

This is where most voice AI projects in broking go wrong. The FCA's pricing rules in ICOBS 6B apply to home and motor insurance sold to a consumer. They do not apply to commercial lines. Since brokers place most UK commercial insurance, a broker's book is split across two regulatory worlds, and an automation design treating them as one will overreach on one side and underprotect on the other. The three rule sets differ in scope, so take them separately.

ICOBS 6B, the pricing rules. ICOBS 6B.2.1R states that "a firm must not set a renewal price that is higher than the equivalent new business price". A broker is caught not because it sets the insurer's price, but because ICOBS 6B.1.1R also bites where a firm is "determining the level of remuneration, including in particular any fees earned by the firm when distributing a product at renewal". Your commission and fees at renewal are in scope even when the underwriting is not yours.

One detail matters once you add a voice channel. ICOBS 6B.2.5R requires that, in working out the equivalent new business price, a firm "must assume that the existing customer has approached the firm through the same channel as they used when they first purchased their policy". An AI-handled renewal channel is not a cheaper-to-serve channel you can price against.

ICOBS 6.5, the renewal disclosure rules. These apply to consumer general insurance contracts that are not group policies and run for ten months or more. The customer must be told the renewal premium alongside last year's premium, presented so the two are easily compared.

ICOBS 6A.6, the auto-renewal cancellation rules. Broader than the pricing rules: they cover all consumer general insurance contracts with an automatic renewal feature, excluding only private health or medical insurance and pet insurance. If your book has consumer auto-renewing policies outside home and motor, ICOBS 6B may not touch them but ICOBS 6A.6 does.

For commercial customers none of the three applies as it does for consumers, and the Consumer Duty is directed at retail customers. That does not make commercial renewals a free-for-all. It makes them a different conversation, governed by the terms of business and the duty a broker owes its client.

The ground has moved. In July 2025 the FCA published Evaluation Paper 25/2, assessing whether the pricing reforms worked. Before them, an existing home insurance policyholder paid on average £95.38 more than a new customer. After them, that differential almost halved, to £49.17.

Motor moved differently. After the reforms new customer prices rose by £111.14 while existing customer prices rose by only £22.71, which is what the regulator expected given that renewing motor customers pose lower average risk. The FCA's causal analysis put the average fall in motor prices attributable to the reforms at £6.63 per policy, with a ten-year benefit range of roughly £163 million to £3.0 billion and a central estimate near £1.6 billion. In home it found no statistically significant causal link either way, and said plainly that this should not be read as evidence consumers were made worse off.

One rule illustrates the whole problem. From the fourth renewal onward, ICOBS 6.5.1R requires the firm to include this exact statement, in writing or another durable medium:

"You have been with us a number of years. You may be able to get the insurance cover you want at a better price if you shop around."

The regulator requires you to tell your longest-standing customers to consider leaving. That is the posture the renewal conversation has to hold, and it should tell you immediately that a retention-optimised sales script is the wrong template for an automated renewal call.

What can a voice AI agent safely do on a renewal call?

A voice agent can safely confirm facts, capture changes, explain what a document says, book adviser time and complete the administrative half of a renewal. It cannot recommend cover, re-rate a risk, or decide that a policy still meets a client's demands and needs. The boundary is not about model quality. Some of these steps are regulated advice and the rest are administration, and only one can be delegated to software.

That splits the call into a safe envelope and a referral set. Inside: identity and record confirmation, capturing changes to a declared risk, confirming payment method, explaining the renewal notice already issued, and scheduling time with the named adviser. Outside: any judgement about whether cover is right, any sign of vulnerability, and any dispute over the premium.

The fourth stage is the one teams skip and the one carrying the risk. A customer who says "I am not sure I can afford this one" has not asked for advice, but they have signalled something a confirmation-optimised agent will talk past. Detection should be deliberately over-sensitive: a false referral costs an adviser three minutes, a missed one costs a complaint. The same logic drives our approach to vulnerable customer detection, and the exit design matters as much as the trigger, which is why escalation and human handover deserves as much attention as the happy path.

Outbound renewal calling adds constraints, since a reminder placed to a consumer sits inside the UK rules on electronic communications as well as the insurance rulebook. Read our guide to outbound calling under GDPR and PECR before scaling the dialler.

How should a broker handle mid-term adjustments with voice AI?

Mid-term adjustments are the strongest voice AI use case in a broker's book and the most underrated. An MTA is usually short, factual and well bounded: a new vehicle, a changed address, an added driver, a revised sum insured. The customer knows what they want, the required fields are known in advance, and the call is expensive to staff precisely because it is short and unpredictable.

The regulatory reason to care about MTA quality is not obvious until you read the disclosure rule closely. ICOBS 6.5.1R does not simply require you to show last year's premium. Where mid-term changes were made, it requires an amount calculated by annualising or otherwise adjusting the premium in effect following the most recent change, excluding all fees or charges associated with those changes. The comparison figure your customer sees at renewal is therefore derived from your MTA history. Sloppy records surface twelve months later as an incorrect disclosure.

The same diagnostic logic underpins our voice AI agents, scoped against a named call type before any deployment commitment.

This changes the design brief. An MTA voice agent is not a call-deflection tool, it is a data-capture tool that happens to run over the phone. Success is whether the adjustment lands in the broker management system as structured, complete, correctly dated data with fee treatment recorded separately. An agent that resolves the call but writes a free-text note has moved the problem, not solved it.

Two rules follow. Refuse to complete an adjustment the agent cannot fully structure. And confirm the change back in the same call, because an unconfirmed adjustment is the most common source of a disputed renewal premium.

What is the cancellation trap in an AI renewal journey?

The cancellation trap is a renewal journey where saying yes is instant and automated but saying no is slow and human. The FCA has already ruled on this. ICOBS 6A.6.4R requires that the methods a firm provides for cancelling automatic renewal "must include at least all the methods by which a consumer is able to purchase a new policy with the firm", and the accompanying guidance names long waits to cancel as an unnecessary barrier.

That guidance, ICOBS 6A.6.5G, is unusually specific for a voice deployment. It says an unnecessary barrier may include "unreasonably longer call waiting times to cancel the automatic renewal feature than to purchase a new policy", or unnecessary questions and steps before a customer can confirm a cancellation instruction. Read that against a typical deployment plan: the AI answers renewals in two seconds, while cancellations are marked sensitive, routed to a retention queue of nine minutes, and gated behind three qualifying questions. That is a close description of the barrier the guidance warns against.

This was no accident of drafting. The FCA's pricing work targeted what it called sludge practices, including requiring customers to telephone to cancel auto-renewal when purchase was available online. A firm that automates the profitable half of the journey and leaves friction on the exit is walking back toward the behaviour the rules were written to stop.

The fix is a design constraint. Whatever the AI can do to renew, it must do to stop a renewal, at comparable speed and with no extra interrogation. Retention conversations remain legitimate, but they belong after the instruction is captured, not as a gate in front of it.

How does voice AI fit a broker's existing platform?

Voice AI fits through the broker management system, not around it. In the UK that means Acturis or Open GI for most firms, with SSP serving part of the market, and the question is whether the agent can read the policy record and write a structured change back to it. If it cannot, the deployment generates work rather than removing it.

This is a concentrated market, and concentration is a risk a broker should price. In June 2025 Applied Systems announced its withdrawal of Applied Epic from the UK broker management system market, citing complex product requirements and steep competition. Brokers built around that platform inherited a migration they did not plan. Any voice layer should therefore couple to your data model rather than one vendor's proprietary screens, a principle covered in our guide to CRM and telephony integration architecture.

Telephony is the easier half. Carriers such as Twilio handle the call path, and the constraint is rarely connectivity but what happens to the recording and transcript afterwards, which is a retention and lawful-basis question rather than an engineering one. Our guide to call recording retention under GDPR sets out the schedule most brokers land on. For firms sequencing across a book, the AI operating model engagement is the right container.

What is the best voice AI setup for an insurance broker in 2026?

The best voice AI setup for a UK insurance broker in 2026 is a platform-integrated agent scoped to renewals and mid-term adjustments on the personal lines book, with a hard referral rule into advisers and full cancellation parity. Dilr Voice is built for this regulated-deployment pattern, but the honest answer depends on what a broker is optimising for, and there are cases where a competitor wins.

With in-house engineering capacity and a need for control over call logic, developer-first platforms such as Vapi, Retell AI or Bland AI will get you further faster than a managed deployment, and you should probably use one. A very large personal lines operation running a contact centre measured in hundreds of seats will find PolyAI has more directly comparable scale references. Where the requirement is genuinely voice quality for a brand-sensitive outbound programme, ElevenLabs leads on that axis. Synthflow suits smaller books where speed to launch outranks integration depth.

A managed regulated deployment wins the middle case describing most brokers: a book split across personal and commercial lines, a broker management system that must be written to correctly, an FCA-facing audit requirement, and no appetite to staff a voice engineering team. The deciding criteria are integration depth into Acturis or Open GI, whether referral logic is a hard rule rather than a prompt instruction, whether cancellation parity is demonstrable, and whether call artefacts are retained on terms your data protection officer will sign. The DATS methodology exists because placement usually matters more than platform.

How do you measure whether renewal automation is working?

Measure renewal automation on four things: containment on the calls it was scoped to handle, renewal conversion against a like-for-like human baseline, mid-term adjustment data completeness, and cancellation parity. The trap is reporting a single containment figure across a mixed call population, which flatters the number by counting simple confirmations against a denominator including calls the agent was never meant to complete.

Containment is the metric most often quoted and least often defined consistently. An agent scoped to confirmations should be measured against confirmations, with referrals counted as successes when the referral rule fired correctly. Our containment rate benchmark sets out how to construct that denominator honestly.

The commercial metric that matters most is renewal conversion against a comparable human-handled cohort, run concurrently rather than against last year's book, because premium movements between years will otherwise swamp the signal. Alongside it, track complaint rate per thousand renewal calls and median time from cancellation instruction to confirmation, which is your evidence of parity if the FCA asks.

Set expectations against what the market achieves. McKinsey's State of AI research, published in November 2025, found that while 88% of organisations report using AI in at least one function, only 33% have taken it into production at scale and just 14% report material EBIT impact. Stanford's AI Index 2026 puts fewer than 10% of enterprises at full scale in any single function. A broker who gets renewal and MTA calls into production is ahead of the market, not catching up to it.

Does a voice AI call replace the written renewal notice?

No. ICOBS 6.5.1R requires the renewal information to be communicated clearly and accurately, in writing or another durable medium, and in a way that draws the customer's attention to it as key information. A phone call is not a durable medium. The voice agent can explain the renewal notice, answer questions about it and confirm the customer received it, but the written notice remains the compliance vehicle and must still be issued.

Want to see this in production? Try Dilr Voice live, book an AI placement diagnostic, see our DATS methodology, or read about our approach to placing AI inside enterprise systems.

Renewals are the highest-frequency regulated conversation most brokers have, which makes them both the best automation candidate in the book and the least forgiving place to get the boundary wrong. The firms that succeed treat the voice agent as an administrative layer with a hard edge, keep the advice with advisers, and prove cancellation parity before they scale. The rest is sequencing, which is what our AI execution office is built to run. For other regulated call types, the enterprise voice AI guide is the pillar, our insurance claims intake analysis covers the FNOL side of the same book, and the conveyancing status update guide shows the pattern elsewhere. More sector work sits in the industries category, or talk to us about your own book.

Written by the Dilr.ai engineering team, practitioners who ship enterprise AI in production. Follow us on LinkedIn for shipping notes, or subscribe via the RSS feed.