Explainer Randy's PM Frameworks

Friction Stage Tagging — Customer Journey Stages

Your customer-side friction taxonomy (Learn → Plan → Build → Execute → Measure, with Support as a modifier): how records are tagged, how the Friction Stage Score is measured, its role in AI Agent for PM, and how it's positioned as published.

Generated
November 18, 2026, 2:48 PM MT
Built from
  • wiki/concepts/customer-journey-stages.md
  • wiki/sources/friction-stage-tagging-spec.md
  • wiki/sources/friction-stage-tagging-project-notes.md
  • wiki/sources/ai-pm-agent-architecture-v1-2.md
  • wiki/entities/ai-agent-for-pm.md

A customer-side diagnostic taxonomy. Customer conversations are tagged by the stage the customer was in when they hit friction. Aggregating tag frequency then shows where customers most often get stuck.

The stages describe how a person learns and does tasks in their role, not how they navigate an app. It's a customer-side tool, not an internal engineering taxonomy.

Naming

Filed in the wiki as Customer Journey Stages, published as Friction Stage Tagging (FST). The spec renamed it to avoid colliding with the marketing lifecycle: it tags where in their own workflow a customer hit friction.

5stages
1cross-cutting modifier: Support
2friction types
Multitags allowed per record
Quarterlyreview cadence
Mediumpublished as FST

The five stages and the Support modifier

The five stages in order, with Support as a cross-cutting modifier rather than a sixth stage. A record that fits no stage goes to the "doesn't fit" pile, and a growing pile signals a missing category.

Support is a modifier, not a stage

Support is a yes/no field on every record, tagged as, for example, "Execute — Support". It's part of FST itself, not a pipeline addition. The "Gas Town for PMs" draft reuses the same modifier pattern for the Backlog Tagging Agent's Security flag.

Learn vs. Measure

The spec includes a test question for telling Learn from Measure apart. Access blocks are a Measure sub-pattern.

The per-stage definitions and the test question itself live in the project spec. The wiki summary records only that they exist.

Not the Lean Startup loop

Ries: Build-Measure-LearnA team-level validation cycle
FSTA per-person diagnostic schema applied at the point of friction

The shared stage names are superficial.

How a record is tagged

Record fields

FieldWhat it holds
Journey stageLearn, Plan, Build, Execute or Measure. Multi-tagging allowed.
Friction typeFunction (something is broken) or opportunity (a capability gap)
SupportYes/no modifier
Customer / accountOn every record. In the pipeline, Account ID is joined from CRM (v1.2).
Doesn't fitFlagged when no stage fits. A growing pile signals a missing category.

Tagging rules

  • Multi-tagging is allowed.
  • Loop or iteration count is a metric, not a stage.
  • Access blocks are a Measure sub-pattern.
  • AI prompts need an explicit "doesn't fit" escape hatch, because overconfident tagging is a real risk.

Friction types

Function friction

Something is broken.

Opportunity friction

A capability gap. The taxonomy maps most naturally to this one.

Which records get a stage (v2 scope decision)

Record sourceJourney stageUses instead
Customer-facing (tickets, meetings, sales calls, feedback forms)Tagged—
Usage / behavioralNullSignal type + strategic dimension
CompetitiveNullSignal type + strategic dimension

Friction type is also null for usage and competitive records (v1.2 record schema).

✓ Contradiction resolved

The article draft's ingestion section now limits journey stage to customer-facing records, matching v1.2.

Measuring it: the Friction Stage Score

account breadth = accounts with a friction tag in the stage ÷ active customer accounts

The primary metric. The denominator comes from CRM/billing. Secondary views are value-weighted breadth and conversation share. The score lags, and must be read against what shipped.

Known limitations (spec) and where they stand

LimitationStatus
No denominator by defaultAnswered in the pipeline: account breadth uses active customer accounts
Internal vs. customer signalTrack them separately
Small samplesKnown limitation
Voiced frustration onlyKnown limitation: friction nobody mentions isn't tagged
AI-tagging disagreement rateReal, and needs spot-checks

Its role in AI Agent for PM

  • The backbone of normalization: "tagging is the normalization mechanism." The per-source classifiers at Ingest apply it to customer sources, and the Journey Stage Labeling Agent labels new or changed themes.
  • Main tag-level signal (v1.4): features carry their journey-stage tag, and ongoing product health is reported per stage rather than per feature.
  • Outcome signal for the light check: a feature's measurement plan can use its stage's score as the Outcome row, e.g. the stage's breadth for the affected accounts.
  • Allowed to loop back: unlike per-feature verdicts, the Friction Stage Score informs the next cycle's Problem Definition Agent, because it describes customer pain rather than how well individual features did.
  • Backlog Tagging Agent (supply side): Customer-Journey-Driven vs. Engineering-Driven work, diffed against the Friction Stage Score.
The tag-level loop: the only measurement signal allowed back into the next cycle.

Published as Friction Stage Tagging

Positioning

  • A channel-agnostic normalization layer for AI classifier pipelines. That's its main differentiator.
  • Channels: support tickets, meetings, sales calls, feedback forms.
  • Built for AI agents and classifier pipelines.

Publishing

? Open
  • The article prose still skews toward support. Was the channel-agnostic body-copy edit made?

The spec asks for a quarterly review of the taxonomy.