miatzSign in

Letting callers say what they want instead of pressing a number

The caller states the reason for the call in their own words; the system classifies it, finishes the handful of requests it can finish end to end, and routes the rest to the right queue with what it has already gathered attached. Every path keeps a working exit to a person on the first ask.

Effort
Weeks of work
Skill level
Some technical skill
Organisation size
Mid-market
Value
Time saved, Quality

Tools named for this

  • Telephony-grade speech recognition tuned to the accents on your actual line
  • An intent classifier limited to the small set of requests you can complete end to end
  • A routing table that carries the collected context into the human queue
  • An always-available escape phrase that reaches a person without repeating the request

What to check before you ship it in India

  • TCCCPR 2018 defines commercial communication by a call's primary purpose being to inform about, advertise or solicit business, and its Explanation says that purpose may be inferred from the content, from the way the content is presented, and from what a caller reaches through a call-back number or a web link inside it. A line the customer dialled to ask about their own order is not that. The moment the same system calls a customer back to close a sale, that leg is commercial communication, and regulation 32 bars an entity not registered with its access provider from making one or causing one to be made.
  • The recording and its transcript are personal data, and they usually outlive the reason they were made. Section 8(7) requires erasure once the specified purpose is no longer served or consent is withdrawn, UNLESS retention is necessary for compliance with any law in force — and in financial services and telecom it often is, so check the sectoral retention rule before setting a deletion schedule, and a transcript store that no team owns is exactly where that duty is quietly missed.
  • An agent that will not yield is worse than a menu. Benchmarking of full-duplex voice agents finds no current system that is both reliably responsive and appropriately restrained, so the exit to a person has to work on the first attempt, every time, including mid-sentence.

Sources

Every claim on this page traces to one of these, on the date it was read.