Chatbots run in almost every second contact centre, and voicebots with open speech recognition reach around 40 per cent for the first time — according to a current trend study of the German-speaking market, based on 399 practitioners. You would expect phone volumes to drop accordingly. They do not. The share of voice in total contact volume has barely moved in two years.
We see this pattern in many projects, and it has less to do with poor AI than with a thinking error during rollout.
The paradox: more bots, the same number of calls
Here is the resolution. A voicebot takes the easy cases off the team: opening hours, delivery status, moving an appointment. What remains and reaches a person is the difficult rest. The number of conversations therefore falls only moderately, while average handling time per conversation rises.
Anyone looking only at call statistics sees no success — even though the bot is doing its job. The work has not disappeared, it has changed shape.
Where relief turns into extra work
There is one case in which a voicebot genuinely creates more work: when it fails and does not hand the customer over cleanly. The typical pattern:
- The customer spends two minutes in a bot dialogue without getting anywhere.
- They are placed in a queue and eventually reach an agent.
- The agent cannot see the bot history and asks the same questions again.
- The customer is annoyed before the actual conversation has even started.
That single conversation now takes longer than it would have without the bot — and leaves a dissatisfied customer behind.
A recent study on voice technology adds a useful detail: only around 17 per cent of respondents have any experience with modern voice agents, but 71 per cent show general interest in concrete use cases. The openness is there. It is simply destroyed quickly when the handover does not work.
The handover is the real technology
Discussion about voicebots almost always centres on speech recognition, model quality and dialogue design. In our experience a less conspicuous point decides between success and failure: what happens in the second the bot gives up?
In myContactCenter the handover is not a break but a transition. The agent receives the complete bot history, the recognised intent and any data already captured — customer number, case reference — directly in their interface. The customer does not repeat themselves, and the conversation starts where the bot stopped.
Which figures actually tell you something
If call volume is the wrong measure, what is the right one? Three figures give a more honest picture:
- Containment rate with follow-up check. How many cases does the bot settle — and how many of those customers call again within 48 hours? A case only counts as settled if it does not come back.
- Handling time after handover. Does it fall because context is carried across, or rise because the agent starts from scratch?
- Distribution of case types. If simple enquiries drop and complex ones stay the same, the bot is working — regardless of what the total says.
Measured this way, the paradox dissolves. The bot has not failed; the wrong yardstick was applied.
