How do you run continuous discovery?

THE SHORT ANSWER

Continuous discovery used to mean two customer interviews a week and a monthly synthesis session. Now agents ingest every sales call, support ticket, NPS response, and app review your company generates, extract the signal automatically, and hand you a ranked opportunity brief every Monday morning. When a signal is clear, you prototype in hours and put a working thing in front of the customer the same day. The cycle that took four to eight weeks runs same-day when the signal is obvious and inside a week when the problem needs a human in the room.

Talking to customers still matters. Empathy still matters. Teresa Torres was right about the habit of continuous discovery. What changed is everything around the conversation: how you find signals, how you process them, and how fast you go from "I think this is a problem" to "here, try this."

Why the old model is too slow

Before AI, good PMs ran two customer interviews a week and synthesized monthly. That cycle took four to eight weeks from signal to decision. Meanwhile your company was generating thousands of customer signals every day: support tickets, sales call recordings, emails, NPS responses, app reviews, product analytics. 99 percent of it sat in different systems, unread by anyone on the product team. Two interviews a week was the best we could do with human bandwidth. It is not the best we can do anymore.

The new discovery loop

Six steps, running in days.

  1. Ingest everything. Every customer touchpoint flows into an AI processing layer. Agents transcribe and analyze, you do not read raw.
  2. Extract the signal. From a 45-minute sales call, an agent pulls the problems described, what they use, what made them hesitate, which competitor they mentioned. From 200 support tickets, it surfaces the top five emerging pains and what is trending up versus last week. The extraction that matters most is the outcome, not the feature.
  3. Surface opportunities. Every Monday, before you open Slack, you have a synthesized view: thousands of data points compressed into the five to ten things that actually matter.
  4. Prototype immediately. When you spot an opportunity, you do not write a spec. You build a working prototype today, in hours.
  5. Give it to the customer. Not a mockup. A working thing they can use. Their reaction to a real thing is worth more than ten interviews about a hypothetical.
  6. Get feedback, iterate, scale. Eyes light up, scale it. They shrug, iterate. They do not use it, kill it. You spent hours, not months.

Setting up the engine

Do not build all of it at once. Week one, turn on transcription with Gong, Fireflies, or Otter so every call becomes text. Week two, build your first signal agent on your highest-volume channel, probably support tickets, that reads last week's tickets, extracts the top 10 problems, and sends a Monday summary. Week three, add a second channel so patterns that span both become gold. Week four, when Monday's brief surfaces an opportunity, build a prototype instead of scheduling more research.

Pick one this week. Turn on transcription, or build one signal agent, or run the full loop once: signal Monday, prototype Tuesday, customer Wednesday, feedback Thursday, decision Friday. Do it once and you will not go back.

SOURCES

THE LONG VERSION

RELATED ANSWERS

Last reviewed 2026-07-31 · 3 min read