Lab note

Voice AI and traditional IVR

An IVR is a decision tree the caller walks. A voice agent is a loop that walks itself. Both properties have consequences.

NEOB Research Published 5 min read
IVRVoice agent
InputKeypad or fixed phrasesNatural speech
PathFixed tree, authoredChosen at runtime
Unanticipated requestDead end or "0 for operator"Attempts an answer
AuditabilityComplete - every path is writtenPartial - behaviour is sampled
Change costRe-record, re-author the treeEdit instructions or knowledge
Worst caseCaller trapped in a menuConfidently wrong answer

The last row is the trade that matters. An IVR fails visibly and annoyingly; a voice agent can fail invisibly and plausibly. Moving from one to the other exchanges a frustrating failure mode for a subtler and occasionally more expensive one.

What IVR got right

  • Predictability. The same input produces the same path, every time, forever.
  • Auditability. You can enumerate everything the system can do.
  • Cost. Effectively zero marginal cost per call.
  • Compliance. A required disclosure is played because it is in the tree, not because a model decided to include it.

The last of those is why hybrids are common and sensible. Regulatory announcements, consent capture and recording notices sit better in deterministic code than in a generated turn. Nothing is gained by making a mandatory sentence probabilistic.

The honest comparison

A voice agent is better for open-ended interaction and for anything with a long tail of requests. An IVR is better where the process is fixed, the volume is high and every path must be provable. Most real deployments want both, with the deterministic parts wrapping the conversational ones.