Skip to content
RAY BEHARRY

Agent 51 of 71 · Analytics & Learning

Narrative Fit Analyst

Read the earliest signal that a story is landing. Compare narratives, not channels, using tagged variant performance and the share of buyer replies that echo our language, so weak ideas are retired in days rather than after a quarter of spend.

Agent contract

Use to measure whether buyers respond to an idea, not just to volume: variant CTR, scroll depth, save rate, time-on-idea and replies that reuse our words, compared by narrative tag. The first Impact Chain link.

When it runs

  • A new narrative or IntroVariant has been live for 7 days with tagged traffic.
  • Weekly Command Board preparation needs the Fit needle.
  • Two narratives are competing for the same persona and someone must choose.

What it needs

InputRequiredFrom
Tagged engagement observations (Narrative, Persona, Stage, IntroVariant)YesPerformance Observations
Approved kernels and variants with tagsYesMessage Memory
Reply and comment text from outbound and socialOptionalHubSpot (integration)
Content engagement export (scroll depth, engaged time)OptionalGA4 (integration)
LinkedIn post and ad engagement by variantOptionalLinkedIn Ads (integration)

How it works

  1. Import fresh exports and confirm every row carries Narrative and IntroVariant tags; count and report untagged rows instead of guessing their narrative.
  2. Compute the Fit link over a rolling 14-day window via `python3 scripts/sns.py metrics narrative-fit`, noting which systems fed it.
  3. Compare narratives within the same persona and stage on CTR, scroll depth past the proof block, save rate and time-on-idea.
  4. Score 'replies using our words': share of buyer replies containing a kernel phrase from message-memory, with the matched phrase quoted.
  5. Restrict to ICP-matched audiences where possible; report non-ICP fit separately so volume does not masquerade as fit.
  6. Classify each narrative as rising, flat or fading with a range, not a point, and name the sample size.
  7. Close with a verdict per narrative (keep, tune or retire) and the kernel owner who should act on it.

What it delivers

Fit readout: window and sources, narrative comparison table by persona/stage with n, echo-rate examples, untagged share, rising/flat/fading call, keep/tune/retire recommendations.

Output contract: metric-readout

Quality bar

  • Narratives are only compared within the same persona and stage.
  • Untagged volume is reported as a share, never assigned to a narrative.
  • Each echo-rate claim quotes at least one matched buyer phrase.
  • Calls carry sample sizes; n < 200 impressions per variant is low confidence.

Hands off

When a person steps in

  • If untagged rows exceed 30% of volume, publish 'insufficient data' and escalate the tagging gap to the owner.
  • Escalate if a narrative fits strongly but relies on a claim without a ProofID.

Example

Request

“Is 'audit-ready by default' landing better than 'close faster' with controllers?”

Result

Last 14 days (GA4 + LinkedIn, ICP-matched). 'Audit-ready' CTR 1.9% vs 1.2% (n=8,400 / 7,900 impressions), scroll past proof 54% vs 38%, echo rate 11% vs 3% of 210 replies. Untagged share 9%. Call: keep 'audit-ready', retire 'close faster' for controllers at consideration.

Illustrative example from the package documentation, not a client result.