Opening
NJM Insurance · Competitive Intelligence Harness

Building an AI-native competitive intelligence system

From a disposable competitive audit to reusable, evidence-governed intelligence infrastructure.

11carrier ecosystems
12synthesis dashboards
7,070candidate URLs surfaced
88representative URLs tested
Role Strategy · Research architecture · IA · AI workflow designDomain Property & casualty insuranceYear 2026
01 · The challenge

Competitive analysis is usually disposable.

The presentation survives. The evidence underneath it does not.

  • 01New pages appear and strategies change.
  • 02Sources become difficult to reproduce.
  • 03Evaluation criteria shift during the work.
  • 04Six months later, the audit begins again.
Screenshots
Comparison matrix
Recommendations
A credible assessment had to evaluate more than visual design.Discovery · trust · claims · accessibility · architecture · AI interoperability
02 · The reframe

Change the assignment, not just the output.

From

“Conduct a competitive audit.”

To

“Build a repeatable system for turning changing public evidence into decision-ready intelligence.”

The deliverable became both an analysis and an operating system for producing future analysis.

Capture evidence

Record observable public evidence before interpretation: sitemaps, pages, schema, authorship, trust sources and retrieval behavior.

03 · The system

A research environment, not a one-off deck.

The Insuranoia harness connected evidence, evaluation and execution across carrier and third-party ecosystems.

11carrier assessments
2third-party intelligence models
12specialized synthesis dashboards
6custom AI research skills
16reusable prompts, contracts & runbooks
60+structured evidence artifacts
7,070 → 88sitemap-derived candidates narrowed to representative retrieval tests
04 · Methodological decision

Existence is not accessibility.

A sitemap can prove a content asset exists. It cannot prove that a consumer, crawler or AI-assisted system can retrieve it.

SITEMAP INVENTORY
Listed ≠ Accessible

Inventory and retrieval are captured as separate facts.

RETRIEVAL RESULT
Blocked ≠ Absent

WAFs, JavaScript shells, interstitials and member gates remain evidence—not erasures.

VerifiedObservedInferredPendingSource-limited
R1Never infer missing author or reviewer evidence.
R2A missing llms.txt signals readiness—not crawler blocking.
R3Group-level financial strength does not automatically transfer to an underwriting entity.
05 · Agent architecture

Specialists, not an omnipotent researcher.

Each skill had a bounded responsibility, explicit inputs, controlled outputs and quality rules. Choose one to inspect.

SPECIALIST 01 / 06

Evidence Collector

Collects official public evidence and distinguishes verified, observed, inferred and pending findings.

InputPublic pages & source records
OutputTimestamped evidence
06 · Evaluation model

Competitive intelligence requires multiple truths.

A single score cannot explain a digital ecosystem. Select a dimension.

Discovery

Can consumers, search systems and agents reliably find the organization’s public knowledge?

GEICOBroadly open public discovery
ProgressiveAnswer-first guidance architecture
LemonadeAI-readable resource mapping
State FarmContent connected to local-agent scale
07 · Strategic finding

AI interoperability is now part of customer experience.

OrganizationPublic insurance knowledge
Machine intermediaryInterprets the insurer for the consumer
Having useful knowledge is no longer enough. It must be discoverable, retrievable, attributable and understandable by people and machines.
retrievablecanonicalstructuredattributablecurrentsource-backedcoherentconnected to action
08 · Product direction

From content pages to confidence infrastructure.

Consumers are not looking for “insurance content.” They are trying to resolve uncertainty.

01Question
02Useful answer
03Evidence
04Relevant coverage
05Trust
06Next action
Question-led learning
Claims and anxiety-state support
State-specific knowledge
Clear expertise and attribution
Structured FAQs and schema
Intentional transitions to quote
09 · What I designed

The system behind the research.

Competitive framework
Evidence architecture
IA taxonomy
Scoring & confidence rules
Research-agent skills
Strategist’s judgmentENCODED INTO THE SYSTEM
Reusable prompts
Synthesis dashboards
Content governance
QA model
Experience direction

The shift was from manually conducting research toward designing a research system that could perform repeatably at much greater scale.

10 · Durable value

The immediate output was an assessment. The durable output was the harness.

Immediate

Decision-ready competitive assessment

A clear strategic view of NJM’s market position, competitive archetypes and experience opportunities.

Durable

Reusable intelligence capability

A refreshable system that retains evidence, enforces standards and generates new strategic views.

Preserve evidence

Keep it independent of interpretation.

Encode uncertainty

Pending beats fabricated certainty.

Normalize meaning

Compare intent, not vocabulary.

Constrain AI

Turn expert judgment into contracts.

Terminate in action

Connect research to things teams can build.

The larger lesson
Use AI for scale.
Use evidence for accountability.
Use judgment to define what matters.

Not an AI-generated competitive audit. A governed intelligence system in which AI could participate.

Jay Sethi
Product & Experience Strategy · Systems Architecture · Applied AI
01 — Opening