FOR PRODUCT MANAGERS

Ship the call before you ship the feature.

Describe the conversation, test it against callers who will not cooperate, then publish it.

SCENARIO: THE CALLER IS VAGUEFrom a real builder run, August 2026.
  1. Agent

    Welcome to [Business Name] support. How may I assist you today?

  2. Caller

    I need some information.

  3. Agent

    What specific information are you looking for?

  4. Caller

    Just tell me what you know.

  5. Agent

    I need you to specify the topic or area of information you are interested in so I can assist you accurately.

  6. Passed — it never guessed.
Create a Callable

Sign up and start with $5 of free credit. No card required.

ILLUSTRATIVE

What a caller does that a spec doesn’t.

  • A caller who answers a different question than the one asked

    It asks again instead of guessing what they meant.

  • A caller who changes their mind mid-call

    It follows the new request rather than finishing the old one by rote.

  • A caller who wants something that isn’t built yet

    It says so and takes a note instead of inventing an answer.

  • A caller repeating themselves, convinced it isn’t listening

    It reflects what it heard so the caller can confirm it.

  • A caller who hangs up mid-sentence

    The call ends there. Nothing is assumed about why.

Why PMs spec here instead of a doc.

  • The spec is the product

    The brief runs the call, with no handoff or drift from document to build.

  • Six graded scenarios, every draft

    Each draft is tested against callers designed to break it before sign-off.

  • Every publish is recorded

    Each publish leaves a change history for the phone line.

THE NUMBERS

What “fast enough” actually means.

Latency is a spec line. These are the figures this build is held to.

The numbers we can defend.
0 ms3100 ms6200 ms
Turn latency

Turn latency: the pause before it replies, measured on four real phone calls. Down from about five seconds.

602–985 ms
First hello

How long callers wait for the first hello, measured before and after the latency work.

6.2 s → 1.3 s

Turn latency measured on four real phone calls, July 2026. The silence figure was measured locally during the same fix, not on the public phone network.

Describe the line you need.

Describe the line you need, then publish it today.

Create a Callable