Voice Agents Beat Chat on Long Conversations by 31.7%
In short
- ElevenLabs compared voice and chat agents across hundreds of live deployments. On conversations longer than two minutes, voice succeeded 31.7% more often.
- Successful long conversations closed in 4.6 minutes on voice versus 7 minutes on chat.
- For a small shop with 500 long conversations a month, that is roughly 20 hours of handling time saved, about €240.
Everybody builds a chatbot for the website. ElevenLabs just published numbers saying that for hard problems, a voice agent does the job better. Here is what you can take from it.
What happened
ElevenLabs looked at hundreds of production agents that run on both voice (phone and WhatsApp) and chat. Each agent had at least 50 voice and 50 chat conversations between July and September 2026. They compared every agent against its own chat results, so the difference between businesses does not distort the picture.
The result depends on length. Under one minute, chat won by about 9 points. Between one and two minutes it was a tie. Over two minutes, voice success was 10.9 points higher, a 31.7% relative gain.
Speed is the second number. In successful conversations over two minutes, voice averaged 4.6 minutes and chat 7. In a separate test with 98 people rebooking a cancelled flight, 85% preferred voice. People speak about three times faster than they type.
One more detail: a 16% cut in latency went together with a 26% lift in success. Slow agents lose people.
How to use it
Keep chat for quick questions: opening hours, order status, a price. It wins there.
Put voice where the customer is stressed or the problem has several steps: a booking that went wrong, a return, a technical fault, a price quote with conditions. Add a phone number or a WhatsApp call to the agent for exactly those cases.
Do the math: say you get 500 long conversations a month. Voice saves 2.4 minutes on each, so 1,200 minutes, which is 20 hours. At €12 per hour for a support person (my assumption, use your own number), that is €240 a month and €2,880 a year. And the customer who is stressed gets solved faster.
What it means
The default “add a chat bubble” is not always the best choice. Companies that route complex cases to voice will resolve more of them without a human.
Be careful with the source. ElevenLabs sells voice agents, so it has a reason to like this result. Conversation length is also a rough stand-in for difficulty, and the preference test was a simulation, not real angry customers. The latency link is a correlation. Test it on your own traffic before you rebuild anything.
Ivan’s Take
Worth testing, not worth believing blindly. The vendor wrote the study, but the method is sound: every agent is compared with itself. I would keep chat for fast answers and try voice only on the long, annoying cases, then measure for a month.
Source: ElevenLabs
If it doesn’t save you time or money, you won’t see it here
- Facts, not hypeEvery price and number is checked against the official source
- Always the real priceYou see what it costs before you click
- I say when it’s weakEven when it’s OpenAI or Google
