Retell AI Alternative UK: Stay or Switch Guide 2026

If your team built its phone agent on Retell AI, a deadline has already been set for you. Retell is retiring its multi-prompt agent type: new multi-prompt agents cannot be created after 15 December 2026, existing ones cannot be edited after 1 February 2027, and Retell then converts whatever is left to its conversation flow format. Single prompt and conversation flow agents are not affected. Any UK operations team running a multi-prompt build therefore has a rebuild, a re-test and a cutover coming, whether it stays or not.

That makes this winter a natural decision point. In McKinsey's State of AI survey (November 2025), 88% of organisations reported using AI in at least one business function, but only 33% had it in production, which is where a live phone agent sits. A production agent that has to be restructured anyway is the cheapest moment to ask whether the platform underneath it is still the right one, which is the question an AI placement diagnostic is built to answer. This guide is written for the Head of Customer Operations who owns that agent and has to sign off the plan.

Scope, stated plainly: this post covers the migration decision and the UK questions a Retell move raises. It cedes per-minute pricing and certification detail to our published head-to-head comparison, the five-vendor matrix to the best AI voice agent 2026 buyer guide, the general teaching on who carries PECR and Ofcom duties to our earlier platform-switching guide, and the generic exit checklist to the post on voice AI vendor exit and offboarding.

This guide is shipped by the team behind Dilr Voice, a multi-agent voice AI platform that chains specialised agents into a single phone call. Or see DATS, the five-stage AI consulting system we use to place AI inside enterprise operations.

What does Retell's multi-prompt retirement change for a UK team?

Retell AI's multi-prompt retirement means every multi-prompt agent must become a conversation flow agent or a single prompt agent. New multi-prompt creation ends on 15 December 2026, editing ends on 1 February 2027, and Retell then auto-converts the remainder. Calls keep flowing during migration, but a UK team that does nothing loses the ability to change a live agent and inherits a conversion it never tested.

Some of the change is already live. According to Retell's deprecation notice, the dashboard no longer offers new multi-prompt agents or lets you assign them to new phone numbers, while existing phone assignments keep working. So a team that wants to add a second number for a new campaign on its existing multi-prompt design is already blocked in the dashboard today, not in December.

The conversion itself is not a toggle. Retell's Conductor tool builds a separate conversation flow agent with a new agent ID. Your phone numbers stay pointed at the original agent until you switch them, and anything in your own code that references the old ID, including batch calls and transfers, has to be updated by hand. Retell is direct about this: "Automatic migration does not update your integration code." The notice also warns that conversion can change how the agent handles calls, which is why testing is listed as a separate step.

For an operations lead, that adds up to three pieces of work with a hard date on them: a rebuild in a new format, a regression test of every call path, and an integration update across whatever CRM, booking and reporting systems touch the agent ID. It is the same work you would do to move platforms. The only question is which destination the work points at, and the honest answer depends on what your agent does, which the next sections take in turn. The DATS methodology we use with clients starts from exactly that question: what the agent is for, before which tool runs it.

Should you convert to conversation flow, drop to a single prompt, or leave Retell?

A Retell AI multi-prompt team has three real options: convert to conversation flow inside Retell, simplify into a single prompt agent inside Retell, or move to another platform such as Dilr Voice. The right choice follows the call design. Strict ordered steps favour conversation flow, a describable task favours a single prompt, and a need Retell does not cover, such as multi-agent handover with built-in per-country calling rules, favours a move.

Retell's own guidance is a useful starting point, because it argues against over-engineering. Its page on choosing between single prompt and conversation flow recommends starting with a single prompt agent and switching to conversation flow only when a single prompt can no longer meet your requirements. It adds that having multiple steps does not by itself require conversation flow, and that conversation flow is not automatically more reliable, because node boundaries and transition conditions need careful design of their own.

That matters for multi-prompt teams, because many multi-prompt builds were multi-state simply because the format encouraged it. Before converting, ask whether the states were doing real work, and who in your AI operating model owns that call. If the agent is an appointment line that greets, qualifies, books and confirms, a well-written single prompt may handle it with less to maintain. If it is a regulated flow where the order of disclosures matters, conversation flow gives tighter control over execution order, although Retell notes that some decisions still depend on the model even then.

Every row carries a re-test in its Main work column, and two of the three also mean re-pointing numbers or integrations. That is the point worth taking to a steering meeting. Staying is not the zero-effort option, it is a different rebuild. If it helps, book a scoping call and we will walk the three paths against your own agent.

When is staying on Retell the right call?

Staying on Retell AI is the right call when inbound handling is the whole job, when procurement requires SOC 2 or a signed HIPAA business associate agreement, or when your engineers value choosing each model component and paying for each line item. Retell's documentation states SOC 2 Type 1 and Type 2 certification and HIPAA and GDPR compliance, and Dilr Voice does not show a SOC 2 report or offer HIPAA.

That concession deserves to be stated without hedging. Retell's security and compliance page says the business associate agreement and the data processing addendum, which includes EU Standard Contractual Clauses, can be self-signed at no additional fee. If a security questionnaire is the gate between your agent and go-live, Retell answers questions that Dilr Voice cannot, and our published comparison says the same in more detail, including where Retell is cheaper on a lean model mix.

Staying also makes sense when the agent is already a conversation flow or single prompt build. The retirement does not touch those agent types, so there is no forced rebuild and no reason to move on the deadline alone. And if your team has invested in Retell's testing tools, such as simulation testing of call paths, that tooling is part of the asset you would be giving up.

The cases where staying is weaker are specific. A team that needs one agent to greet, a second to qualify and a third to take an action, each with its own prompt, model and tools, is rebuilding that structure either way. A team running outbound calling into UK numbers needs consent capture, opt-out recognition and permitted calling hours somewhere in the stack, and has to decide whether that lives in the platform or in its own code. Those are the cases the rest of this guide is written for, and our guide to orchestration versus platform choices covers the wider split between the two models.

What does conversion actually change in a live agent?

Converting a Retell AI multi-prompt agent changes how decisions are made inside the call. Multi-prompt states move between each other through model tool calls described in the state prompt, while conversation flow moves between nodes on transition conditions, which can be prompt conditions judged by the model or deterministic rules. The same caller can therefore take a different path after conversion, so every route needs re-testing.

Retell's multi-prompt documentation describes the old model plainly: each edge between states is implemented as a tool call, so the state prompt has to tell the model when to take it. Conversation flow replaces that with nodes and edges, plus logic split nodes, global nodes and reusable subflows. Those are useful, but they are a different mental model, and a converted agent is a machine translation of your design into it. If you are weighing a rebuild against a platform change, our build, orchestrate or buy guide sets out the wider trade-offs. That is why Retell's own notice recommends reviewing the generated flow and tools and running simulation or playground tests before switching numbers.

There is also a cost behaviour worth knowing before you choose a mode. Retell's Flex Mode documentation explains that Flex Mode compiles a whole flow into one structured prompt at runtime, so the agent can handle callers who jump between tasks. The trade is token volume: when the combined prompt exceeds 4,000 tokens, Retell's token scaling billing rule applies, which the page says can multiply costs several times over. A team that converts a large multi-prompt agent and then enables Flex Mode to recover the old flexibility should model that before it reaches production.

None of this is a reason to leave on its own. It is a reason to treat conversion as a rebuild with a test plan, not an upgrade. The same diagnostic logic underpins our AI execution office, the embedded delivery team that runs changes like this inside client operations.

Where does UK caller data sit on a Retell deployment?

For a UK buyer, the data question on Retell AI starts with location. Retell's compliance page says it complies with GDPR through Amazon Web Services and does not currently operate services within the European Union, while its homepage also offers an enterprise on-premises deployment for residency requirements. A UK team should ask where its call data is processed and which transfer mechanism covers it.

Retell's homepage describes that on-premises option as deploying Retell within your own infrastructure, which changes the residency answer entirely, so establish which deployment you are actually buying. Under UK GDPR, sending personal data outside the UK is a restricted transfer unless the destination is covered by UK adequacy regulations or the transfer carries an appropriate safeguard. The ICO's guidance on adequacy regulations, last updated on 30 July 2026, explains how the UK Extension to the EU-US Data Privacy Framework works, and its guidance on appropriate safeguards names the UK IDTA and the Addendum as the standard data protection clauses. Retell's addendum includes EU Standard Contractual Clauses. Whether it also carries the UK Addendum, or whether Retell relies on the UK Extension, is a question to put to Retell in writing, not something to assume. Our post on UK international transfers for voice AI covers the risk assessment itself.

Retention is the second question, and the defaults matter. Retell's data retention policy is set per agent, defaults to keeping call data forever, and offers fixed periods from one day up to 730 days. Its data storage settings offer three modes, everything, everything except PII, or basic attributes only, and when you opt out of storage the raw recording, transcript and log files are deleted about ten minutes after post-call processing. That is a strong control, but it is per agent. A migration that creates a new agent ID is exactly the moment a retention setting gets left on its default, so check it on the converted agent, not just the original.

Retell also states that enterprise customers with a master services agreement do not have their data used to train models, and that its model and speech subprocessors operate under zero data retention agreements. Those are useful commitments to record in your own processing records. Whichever platform you choose, the controller duties, such as telling callers how their data is used and honouring objections, stay with you as the organisation deploying the agent, a point our UK and EU voice AI compliance guide sets out in full.

How does Dilr Voice differ from a Retell build?

Dilr Voice is built around chaining specialised agents into one phone call: a greeter, a qualifier, a knowledge agent, a customer care agent, an action agent and custom types, each with its own prompt, model and tools, passing context on handover. A Retell build structures one agent as a single prompt or a flow of nodes. Dilr Voice also ships per-country calling rules by default, so consent, opt-out and calling hours are configured rather than coded.

The multi-agent design matters for a migrating team because it is the closest analogue to what multi-prompt states were trying to be. On the Dilr Voice platform, each specialised agent is a separate unit with its own instructions and tools, and the handover carries what the previous agent learnt. A team whose multi-prompt build had a qualifying state and a booking state can map those to a qualifier agent and an action agent rather than redrawing them as nodes and transition conditions.

Compliance is the second difference. Dilr Voice ships per-country compliance rules by default: recording consent, DNC registry checks, opt-out recognition, permitted calling hours and a full audit trail on every call. For an outbound campaign into UK numbers, that puts registry checks and calling windows in the platform configuration rather than in custom functions your team maintains. Ofcom's refresher on silent and abandoned calls is a reminder of why that matters: any person or organisation can be liable for persistent misuse, including call centres, with penalties of up to 2 million pounds, so the organisation running a campaign cannot treat it as the software's problem.

The third difference is the operational surface. Dilr Voice runs outbound campaigns from an uploaded contact list, with scheduling windows, retry logic and auto-pause at the configured daily end time, alongside an inbound AI front desk with business hours, after-hours routing and warm transfer to a human with full context. Post-call automation updates CRM records and sends email, SMS and Slack notifications without code, with live integrations for HubSpot, Salesforce and Google Calendar. Retell covers much of the same ground through its own tools, batch calling and custom functions; the difference is how much of it your engineers configure rather than build.

Where Retell keeps an edge is just as clear. It publishes component-level pricing, it holds certifications Dilr Voice does not, and it lets a builder pick model and voice providers with each priced as its own line item. If those points decide your purchase, they should. Those trade-offs are the substance of the decision, not a footnote to it.

How do you cut over from Retell without dropping calls?

Cutting over from Retell AI without dropping calls means running old and new side by side and moving numbers last. Build and test the replacement, whether a converted Retell agent or a Dilr Voice deployment, on test numbers first. Then move live traffic in stages, keep the original agent untouched as a rollback, and update every integration that references the old agent ID.

Inside Retell, the mechanism for a staged move is environment tags. Retell's agent versions documentation explains that a published version is read-only, that each agent starts with prod and staging tags (up to 10 tags in total), and that a phone number assigned to a tag follows whichever version the tag points at. That gives a clean rollback for an in-Retell conversion: point staging at the converted agent, test, then move prod. One caution for multi-prompt teams is that conversion creates a new agent with a new ID, so check how your numbers and tags are assigned before assuming a tag move will carry them. Our guide to release management for voice agents covers rollback triggers and approvals in general terms.

Moving to another platform follows the same shape with one extra layer, the telephony. Retell's custom telephony guide shows that numbers held with your own carrier, such as Twilio, Telnyx or Vonage, are connected to Retell through elastic SIP trunking and imported. Because the guide has you point a number at a SIP trunk and import it, the number itself stays in your carrier account, which makes a staged cutover simpler as long as the new platform supports that carrier. Dilr Voice telephony runs on Twilio or Vobiz, and outbound campaigns require the customer's own Twilio or Vobiz account.

A practical order for an operations team is: freeze changes to the current agent, export a sample of recent call outcomes as the baseline, build the replacement, run it against recorded scenarios, move one low-risk number or campaign, compare outcomes for a week, then move the rest. Keep the original agent and its numbers available until the new one has handled a full business cycle. Book a 30-minute scoping call if you want a second pair of eyes on that plan before the December date.

What is the best Retell AI alternative for UK teams in 2026?

The best Retell AI alternative for a UK team in 2026 depends on what the agent does. For a single inbound line needing SOC 2 or HIPAA paperwork, staying on Retell is usually right. For multi-step calls that hand between specialised agents, or outbound campaigns needing per-country calling rules built in, Dilr Voice fits. Teams wanting full control of their own model stack should compare Vapi.

Four criteria do most of the work in that verdict. First, call structure: one task, ordered steps, or several specialised roles. Second, direction: inbound only, or outbound campaigns into UK numbers where consent, opt-out and calling hours apply. Third, procurement evidence: whether a certification is a hard gate. Fourth, ownership: whether your engineers want to assemble components or configure a platform. Score your agent on those four before you look at a feature list.

Other names belong in the shortlist for specific jobs. Our buyer guide rates Bland AI for high-volume outbound, Synthflow for no-code builds and Vapi for engineering teams that want to bring their own model stack. That buyer guide compares five platforms, Dilr Voice among them, side by side with sourced figures. Where a rival leads on a specific axis, that guide says so, and the same honesty applies here: if certifications are your gate, Retell wins that axis today.

For a wider frame, our enterprise AI voice agents guide explains how to scope, measure and govern a voice agent regardless of vendor, including how to set the outcome measures a migration will be judged against.

Will a Retell multi-prompt agent stop taking calls on 1 February 2027?

A Retell AI multi-prompt agent does not stop taking calls on 1 February 2027. That date ends editing, not call handling. Retell's notice says existing multi-prompt agents continue handling calls during migration and that Retell will migrate remaining ones to conversation flow after editing ends. The risk is running a live agent you cannot change, followed by an untested automatic conversion.

Can you keep your phone numbers if you move off Retell?

You can often keep your phone numbers when moving off Retell AI if they sit in your own carrier account and are trunked into Retell over SIP, because the number stays with that carrier. The new platform must support the carrier: Dilr Voice runs on Twilio or Vobiz. Numbers bought through Retell itself need a separate check, so confirm the position in writing before setting a cutover date.

Does Flex Mode let a team avoid rebuilding a converted flow?

Retell AI's Flex Mode does not remove the rebuild, but it changes how a converted flow behaves. It compiles the nodes, transitions and tools into one structured prompt at runtime, so the agent can move between tasks more freely. Retell warns that this raises token use, and prompts above 4,000 tokens trigger its scaling billing rule, so test cost and behaviour together.

Want to see this in production? Try Dilr Voice live, read the best AI voice agent 2026 guide, see how an AI operating model assigns ownership of a live agent, or read about our approach to placing AI inside enterprise systems.

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.