Zendesk's Real-Time Voice Translation Is a Bet on Phone Support, Not a Fix for It
Zendesk has introduced Real-Time Voice Translation for its Contact Center product, a feature that translates live phone conversations between agents and customers as they speak. The company disclosed the launch on September 17, 2026, in a report by Small Business Trends, citing its own research that voice interactions make up 40% of contact center volume. Zendesk President of Product, Engineering and AI Shashi Upadhyay framed the case for the tool by arguing customers "should not have to choose between speaking the language that feels most natural to them and getting the support they need." The feature is set to roll out in October through a closed Early Access Program, meaning general availability, pricing and supported languages are not yet public.
What does the feature actually do?
Agents can toggle live translation on or off mid-call, and admins configure which queues and languages get access, without hiring multilingual staff.
That's the pitch, at least. It's aimed squarely at the routing problem every contact center with international customers already knows: language-specific queues that create wait times, or fallback to email and SMS when a caller needs something resolved now. Voice translation, if it works reliably, removes a staffing constraint rather than a technology one — you stop needing an agent who happens to speak Portuguese, and start needing any available agent plus the tool.
How solid is Zendesk's 40% number?
It's Zendesk's own research, measured on Zendesk's own product mix, with no methodology published in the coverage so far.
A vendor-reported share-of-volume figure is a reasonable planning input, but it's not independently audited, and "voice" as a category will vary wildly by industry — a SaaS help desk and an airline are not comparable here. Buyers should ask Zendesk what counts as a voice interaction in that 40%, over what customer base it was measured, and whether it reflects call volume or call duration. Those are different problems with different translation-latency tolerances.
What should support leaders ask before signing up for Early Access?
Ask about translation latency, accuracy on accented speech, language coverage at launch, and what happens when translation fails mid-call.
None of this is answered yet because the program hasn't opened. Real-time voice translation is a harder engineering problem than chat or email translation — it has to handle interruptions, accents, background noise and industry jargon while an anxious customer is on the line. Small Business Trends' own coverage flags this as an open question, not a solved one, noting that translation quality "will need to be monitored" once agents are actually using it. Any team evaluating this should ask for a pilot with their own call recordings, not a vendor demo script.
Where does this sit next to the rest of the AI-support wave?
It extends the same automation logic already applied to chat and ticket routing — reduce headcount dependency — but voice is less forgiving of errors.
Text-based AI support tools can be reviewed and corrected before a customer sees them; a live voice mistranslation happens in real time, in front of the customer, with no edit pass. That raises the bar for what "good enough" means here compared with the chatbot and ticket-triage tools this site has covered.
Frequently asked questions
When can customers try it?
Zendesk says the feature enters a closed Early Access Program in October 2026; broader availability hasn't been announced.
Do agents need to speak the customer's language?
No — the tool is designed to translate in real time during the call, though independent accuracy data isn't yet available.
Is this free for existing Contact Center customers?
Pricing hasn't been disclosed in current reporting; ask Zendesk directly before assuming inclusion in your current plan.
Source: Small Business Trends.