Zoom Virtual Agent Alternative UK: Test the Claims First

Zoom now sells an AI agent that answers the phone. For a UK team already running Zoom Phone or Zoom Contact Center, that changes the question from "which voice AI vendor?" to "do we need anything outside Zoom at all?" Zoom's own product page answers with confidence: the Virtual Agent understands intent, completes tasks, works across phone and messaging, and is pitched as running in half an hour. The page also carries customer stories and links to a white paper with benchmark figures.

This guide is for the Head of Customer Operations who has to decide, on evidence, whether Zoom Virtual Agent should take the phone work or whether a separate voice AI platform should. The single test it applies is Zoom's own evidence: what each published claim and figure actually measured, which product it describes, and how to re-run it on your own UK calls before you commit. It is deliberately narrow. CRM write-back after a qualification call is covered in the Aircall alternative guide, vendor data location and merger risk in the Dialpad alternative guide, accent handling in depth in our accent and dialect recognition guide, and response time in the voice agent latency benchmarks. This post links to those rather than repeating them.

This guide is shipped by the team behind Dilr Voice, a multi-agent voice AI platform that handles inbound and outbound calls on one platform. Or see DATS, our five-stage AI consulting system, if the harder question is where AI should sit in your operation at all.

What is Zoom Virtual Agent, and is the AI receptionist the same product?

Zoom Virtual Agent is Zoom's conversational AI agent for customer service, and it reaches callers in two shapes. ZVA Receptionist answers calls on a Zoom Phone number, routes callers, takes messages and books appointments. The fuller Virtual Agent runs across voice and digital channels and hands off into Zoom Contact Center. Both sit under one name, so a UK buyer has to establish which shape a claim describes.

Zoom's Virtual Agent product page describes an agent that "communicates naturally across phone and digital messaging channels" and lists three use cases: self-service support, an AI receptionist and an internal IT helpdesk. The receptionist is the shape a UK office on Zoom Phone would meet first, because it attaches to an existing Zoom Phone number rather than requiring a contact centre.

Zoom's own ZVA Receptionist Start Guide is precise about what that shape needs. The prerequisites are an Account Owner or Admin role, a Zoom Phone or Zoom Contact Center licence, a Zoom Virtual Agent licence, and a Zoom Scheduler licence only if the agent will book appointments. The receptionist is built in AI Studio inside the Zoom web portal, was previously called AI Concierge, and reaches callers through Auto Receptionist, the Zoom Phone component that connects an inbound number to the published agent.

The guide also defines what the receptionist can do through Skills: routing to Zoom Phone destinations, Dial-by-Name and appointment scheduling, on top of answering questions from a Knowledge Base built from uploaded files or linked web pages. That is a clear and honest scope. It is also the scope you should hold every later claim against, because a figure measured on the fuller Virtual Agent, or on chat, does not automatically describe a receptionist answering your main number.

This matters before any comparison starts. If your requirement is front-desk answering and routing on Zoom Phone, you are evaluating ZVA Receptionist. If it is resolving customer requests end to end with a hand-off to live agents, you are evaluating the Virtual Agent with Zoom Contact Center, a different licence footprint. Our enterprise AI voice agents guide sets out how those two jobs differ on any platform.

What does Zoom's own evidence actually show?

Zoom publishes two kinds of evidence for its Virtual Agent: customer stories on the product page and a November 2025 white paper reporting testing Zoom ran with Coval. Read closely, the customer figures describe chat or the wider Zoom CX estate rather than phone calls, and the benchmark figures are vendor-published lower bounds. Neither is a measurement of a UK phone line.

Start with the customer stories, because they carry the largest numbers. On the same product page, Vensure's chief information officer says that under 30 percent of its chats were self-service before moving to Zoom and that it was trending towards 75 percent within two months. Cricut's quoted figure is a 50 percent self-service containment rate "across Zoom CX". Zoom's own "Zoom on Zoom" resource reports a 97 percent self-service rate from Zoom's own AI chatbot, with a 28 percent increase in CSAT. Every one of those is either chat or a blended figure. None tells you how the agent performs when a caller rings your number.

That is not a criticism of Zoom. It is a reason to keep the figures in their lane. A phone call adds things a typed chat does not have, such as background noise, speech recognition errors and account numbers read aloud, so a chat containment rate is not evidence for a phone line. If a business case for the phone line borrows one, the ROI model built on it rests on the wrong channel.

The second source is more relevant to voice. Zoom's white paper, Elevating AI quality with Zoom Virtual Agent, dated November 2025, says Zoom "partnered with Coval", which it describes as an independent benchmarking firm, to test the Virtual Agent against 12 key metrics using simulated conditions that include multi-intent conversations, background noise, rapid-fire questions and adversarial inputs. Zoom also says it validated performance with "thousands of real and simulated calls". The paper reports the Coval results as thresholds rather than point values.

Zoom reports above 95 percent intent recognition accuracy across all test variations, above 68 percent success on multi-intent processing under "extreme adversarial scenarios", above 70 percent appropriate escalation decisions, and a response failure rate below 1 percent. The same paper reports a 6.48 percent word error rate for Zoom's speech recognition on the Hugging Face Open ASR Leaderboard, which it describes as top five at the time; leaderboards move, so treat that ranking as a November 2025 snapshot.

Three things about those numbers matter to a buyer. They are published by Zoom in its own marketing material, which does not make them false but does mean you are reading the vendor's selection of results. They are thresholds: a figure reported as above 70 percent could be 71 or 99, so ask Zoom for the measured values, especially for escalation, the decision a caller who wanted a person notices first. And they were measured on Zoom's test conversations, not on your callers. The useful move is to treat them as the claims you will re-test, which is the next section.

How do you turn Zoom's benchmark into a test on your own UK calls?

Turning Zoom's benchmark into your own test means keeping its metric names and replacing its conditions with yours. Take the four reported measures, intent recognition, multi-intent handling, escalation decisions and response failures, and score them on a sample of your real UK call types, then run the same sample through any alternative platform. The comparison only counts when both agents face identical calls.

Zoom gives you a head start here. Its product page promises "programmatic phone-based testing that validates voice agents across languages, accents, and real-world noise variations", and the Receptionist guide tells admins to start with two to three core routing paths and to review at least 10 to 20 conversations when tuning. Those are sensible instructions. Use them to build a test you own rather than one the vendor runs.

We suggest a sample in four parts. First, your top call reasons by volume, drawn from the last quarter's call logs, so the test reflects demand rather than a demo script. Second, a set of multi-intent calls, where the caller changes topic halfway through, because Zoom's own multi-intent figure is its lowest. Third, calls where escalation is the right answer: a complaint, a vulnerable caller, a request outside the knowledge base. Score whether each agent escalates when it should and stays when it should. Fourth, calls recorded in the conditions your callers actually ring from: mobiles on the move, speakerphones, regional accents.

Score each call on outcome, not on how natural the voice sounded. Did the caller's request end resolved, correctly escalated, wrongly escalated, or failed? Our QA and testing framework for AI voice agents sets out a scoring sheet for exactly this.

Two practical cautions. Keep the knowledge base identical across candidates, or you are testing content rather than platforms. And test the hand-off end to end, including what the human agent sees when the call arrives; Zoom says its agent works natively with Zoom Contact Center and Zoom Phone for hand-offs, which is a real advantage inside its own estate, and our warm transfer and context hand-off guide covers what to check on any platform. The same diagnostic logic underpins our AI placement diagnostic, which ranks where automation will actually move cost before anyone signs a contract.

What does Zoom's UK hosting change for a team already on Zoom?

Zoom's UK hosting puts Zoom Virtual Agent, Zoom Phone and Zoom Contact Center on UK data centres, and Zoom announced it live on 8 June 2026. The qualifier matters: Zoom said it was available to new paid UK customers at no extra cost and "coming soon" for existing paid customers. A team already on Zoom should confirm its own migration date in writing before relying on UK residency.

Zoom's UK infrastructure announcement is specific. It says the regional instance covers twenty services, that Zoom Virtual Agent is "deployed on UK infrastructure", and that UK-hosted Virtual Agent runs on UK data centres alongside Zoom Contact Center. It points to a UK Infrastructure Fact Sheet on the Zoom Privacy Trust Center for the technical detail, and it names public sector as the lead audience while also addressing financial services, legal and utilities.

For the reader of this guide, the existing-customer line is the one to act on. If your organisation bought Zoom Phone before June 2026, the announcement does not by itself move your tenant. Ask Zoom three things: when your account moves, whether Virtual Agent processing for your account moves with it, and which processing, if any, still happens outside the UK. Get the answers from the Fact Sheet or your contract, not from a sales call.

Hosting location is also only one part of the obligation. Under UK GDPR, the organisation that decides why and how caller data is processed is the controller, and a platform answering calls on its behalf is often a processor acting on its instructions, although the roles turn on the facts; the ICO's guidance on controllers and processors sets out the test. Moving the data to a UK data centre does not change who is responsible for it. Retention of recordings and transcripts is a separate decision again, and the wider rules for UK and EU callers are in our UK and EU voice AI compliance guide.

Which languages will a UK caller hear from Zoom Virtual Agent?

Zoom Virtual Agent's language coverage depends on which product and which date you read. Zoom's July 2026 blog lists six languages for ZVA Receptionist, while the November 2025 white paper says the Virtual Agent supports 13 voice languages and 25 chat languages. The figures come from different documents and dates, one about the receptionist and one about the Virtual Agent as a whole, so confirm your exact languages, including the English voice, with Zoom.

Zoom's multilingual AI receptionist post, published on 30 July 2026, says ZVA Receptionist can be configured to respond in English, Spanish, French, German, Portuguese and Japanese, and that each is supported for both speech recognition and text-to-speech. That is a clean list for the receptionist shape. The white paper does not name its 13 voice languages, so you cannot read your own language requirement off it.

The white paper also reports that, in testing within one week of multi-language launch, inquiry resolution stayed within 12 to 17 percent of the English baseline across French, Italian, German and Spanish. That is a useful disclosure, because it measures other languages against English rather than on their own. It also tells you which languages were tested: Italian is not on the receptionist's list of six, and Welsh, Polish, Punjabi and Urdu were not tested at all.

For a UK line the more common gap is not language but accent and register inside English: Glaswegian, Scouse, Welsh English, callers whose first language is not English, and older callers on poor lines. The white paper's figures do not break results down that way. Add those callers to the test sample described above, and use the accent guide linked in the introduction to build that sample without guessing.

When is staying on Zoom Virtual Agent the right choice, and when is it not?

Staying on Zoom Virtual Agent is the right choice when the job is inbound answering, routing and resolution on an existing Zoom Phone or Zoom Contact Center estate, and the systems it must act in are ones Zoom documents. A separate voice AI platform makes more sense when the phone work includes outbound campaigns, specialised agents chained in one call, or telephony that does not run on Zoom.

The case for staying is real and worth stating first. Zoom says its Virtual Agent integrates natively with Zoom Contact Center and Zoom Phone for hand-offs that keep context, and its Contact Center virtual agent page names connections to Zoom Quality Management, to CRM and ticketing systems including Salesforce, Zendesk, Microsoft Dynamics, Intercom and Kustomer, and to third-party contact centre platforms such as Genesys Cloud and Amazon Connect. The same page says the agent can create and update tickets and perform actions within your existing tools. If your agents, numbers, recordings and quality reviews already live in Zoom, keeping the voice agent with the same vendor is a genuine operating simplicity. For a receptionist job on Zoom Phone, that may be all you need.

The case for a separate platform depends on your own call mix, so list the actions your calls need: a booking written to a calendar, a lead scored and logged in a CRM, a payment status checked, a follow-up text sent. Check each against the systems Zoom names, and ask Zoom to show any that are not named working in a demo on your data. Look at outbound too. If the requirement includes campaigns to a contact list with calling-hour rules, retries and DNC checks, read Zoom's documentation for that requirement specifically rather than assuming the Virtual Agent covers it; the Receptionist guide describes an inbound agent.

Then look at Zoom's "zero lock-in" claim with care. On the product page it sits under the heading "Future-proof AI, zero lock-in", and the sentence beneath says Zoom continuously upgrades the virtual agent's AI backbone. The page describes that as "without vendor dependency", but the sentence is about Zoom upgrading the AI behind the agent, not about moving your build elsewhere. Your knowledge base, Agent Guidance and Skills are configured in Zoom's AI Studio, and the page does not say how they would be exported. Ask Zoom directly, and ask every other vendor on your list the same question, which is a good reason to decide the job before the tool.

Our AI operating model work helps a team decide who owns that job, and if the answer is "nobody yet", that is a better first project than a vendor switch. For teams weighing the same native-versus-separate decision on other estates, the Genesys Cloud and Five9 alternative guides take the contact-centre view.

How does Dilr Voice differ from Zoom Virtual Agent?

Dilr Voice is a multi-agent voice AI platform whose telephony runs on Twilio or Vobiz rather than on a business phone system such as Zoom Phone. It chains specialised agents, such as a greeter, qualifier and action agent, into a single call, and handles inbound and outbound work on one platform. Zoom Virtual Agent's advantage is native hand-off inside the Zoom estate.

The difference is architectural, and it cuts both ways. On Dilr Voice, each agent in a call has its own prompt, model and tools and passes context to the next on handover, so a front-desk greeter can hand to a qualifier and then to an action agent within the same call. Post-call automation needs no code: CRM records, email, SMS and Slack notifications, and lead status and score updates. Named integrations are HubSpot, Salesforce and Google Calendar, and DILR publishes support for 30+ languages.

Outbound is where the two diverge most. Dilr Voice runs outbound campaigns from an uploaded contact list, with scheduling windows, retry logic, auto-pause at the configured daily end time, and real-time outcome and sentiment analytics. Per-country compliance rules ship by default, including recording consent, DNC registry checks, opt-out recognition, permitted calling hours and full audit trails on every call. Outbound campaigns require the customer's own Twilio or Vobiz account. DILR states its own measured response time as under 500 milliseconds, on its own on-platform measurement, which you should re-test on your calls exactly as you would Zoom's figures.

The concessions are just as concrete. Dilr Voice's telephony runs on Twilio or Vobiz, so a team that wants the agent and its human colleagues all on Zoom gets that only from Zoom. If you need the agent to hand into a Zoom Contact Center queue with Zoom's own context passing, Zoom does that natively. Dilr Voice is hosted on Google Cloud Platform with regional data residency options for enterprise; confirm the region for your account with us, as you would with Zoom. Dilr Voice offers a free trial with no credit card required, and the Dilr Voice product page shows the real screens behind each job. If you would rather talk through a call sample first, book a scoping call.

Zoom Virtual Agent and Dilr Voice side by side

The table below compares the two on what each vendor states publicly. It is a starting list for your own test, not a verdict.

What is the best Zoom Virtual Agent alternative for UK teams in 2026?

The best Zoom Virtual Agent alternative for a UK team in 2026 depends on whether the phone work outgrows a receptionist. For inbound work inside a Zoom estate, staying on Zoom is often right. For outbound campaigns, specialised agents chained in one call, or telephony outside Zoom, a separate platform such as Dilr Voice can fit better. Decide on your own re-tested calls, not on either vendor's figures.

Use four criteria. First, scope: is the job reception and routing, or resolution and action across systems? Second, estate: do your agents, numbers and quality reviews live in Zoom, and will they stay there? Third, direction: is any of the work outbound, and does the candidate document outbound for your use case? Fourth, evidence: did the candidate perform on your own sample of UK calls, scored by outcome?

Zoom wins clearly on the second criterion for a Zoom-native team, and that should not be argued away. Other names on a UK shortlist include PolyAI, Parloa and Sierra, and orchestration tools such as Vapi and Retell AI for teams that want to assemble their own stack, a trade-off our orchestration versus platform guide works through. Our buyer's guide to the best AI voice agent in 2026 compares the field on the same criteria.

One wider number keeps the decision in proportion. McKinsey's State of AI 2025 reported that about a third of organisations have AI in production, which means many are still at the pilot stage. A short test on real calls is the most direct way to find out whether a voice agent is ready for production on your line.

Does Zoom Virtual Agent need Zoom Contact Center to answer calls?

Zoom Virtual Agent does not always need Zoom Contact Center. Zoom's Receptionist Start Guide lists Zoom Phone or Zoom Contact Center as the prerequisite, alongside a Zoom Virtual Agent licence, and the receptionist reaches callers through Auto Receptionist on a Zoom Phone number. A Zoom Scheduler licence is needed only if the agent books appointments, so licence scope should be confirmed against your use case.

Does "zero lock-in" mean a Zoom Virtual Agent build is portable?

Zoom's "zero lock-in" line sits beside its statement that Zoom continuously upgrades the virtual agent's AI backbone without vendor dependency. The product page does not say whether a knowledge base, Agent Guidance and Skills built in Zoom AI Studio can be exported to another platform. Buyers should ask Zoom, and every other vendor, how a build would be moved before they commit.

Where to go next

Zoom's quoted promise is that "Plug-and-play setup gets your virtual agents running in just 30 minutes". Running is the easy part. The work that decides the outcome is choosing the job, writing the knowledge, testing the hand-offs and measuring results on real calls. Our approach explains how we sequence that work.

Want to see this in production? Try Dilr Voice live, read our best AI voice agent comparison, see our DATS methodology, or browse more voice AI guides for UK teams.

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.