00 · Overview
The Framework at a Glance
The Product Intelligence Loop is a multi-agent system designed around the five most universal failures in product organizations. Each layer addresses a distinct, named problem. The loop closes when insights from Layer 5 feed back into Layer 1 — customer signals, retro learnings, and decision logs become inputs to the next planning cycle.
Layer 1
Capture
Discovery is a black box. Ideas get lost, duplicated, or never properly defined.
Intake AgentDuplicate Detector
Layer 2
Connect
Strategy lives in a deck nobody reads while tickets get built for the wrong reasons.
Alignment ScorerRouting Agent
Layer 3 · Core Intelligence
Plan
Quarterly planning is driven by politics, not data. The loudest voice sets the roadmap. Nobody asks which initiatives are better together.
Portfolio Scoring AgentClustering AgentScenario ModelerPlanning Prep AgentScoring Integrity LayerPlanning LockException Process
Layer 4
Monitor
PMs hand off to engineering and go dark. Surprises surface at deadlines, not before.
Delivery Pulse AgentDecision Capture Agent
Layer 5
Learn
Every sprint ends and the same mistakes recur. Institutional knowledge walks out with every PM departure.
VoC Synthesis AgentRetro Intelligence Agent
The Closing of the Loop
"Insights from Layer 5 feed directly back into Layer 1. Customer themes inform the next intake conversation. Retro patterns shape the scoring model. Decision logs become knowledge base inputs for future duplication checks. The system compounds over time."
Layer 1
Capture
"Good ideas get lost, duplicated, or never properly defined before they consume PM time."
Agent 1.1
Product Intake Conversational Agent
Problem It Solves
Stakeholders submit raw, unstructured ideas with no consistency. PMs spend hours in back-and-forth just to understand what's actually being asked.
What It Does
Runs an adaptive multi-turn conversation to extract problem statement, personas, goals, constraints, and success metrics. Searches the KB and active projects in real time. Flags gaps and asks targeted follow-ups dynamically.
Value In One Line
Turns a raw idea into a structured intake record in under 5 minutes — without a PM in the room.
Agent 1.2
Duplicate & Overlap Detector
Problem It Solves
The same idea gets pitched repeatedly across teams, quarters, and stakeholders. PMs waste weeks scoping work that already exists or was killed for a reason nobody remembers.
What It Does
Runs semantic search — not keyword match — across active epics, backlog items, shipped work, and archived PRDs. Returns similarity scores, relationship type (duplicate / adjacent / dependency), and a recommended action.
Value In One Line
Catches redundant work before a PRD ever gets started.
Layer 2
Connect
"Strategy lives in a deck nobody reads while tickets get built for the wrong reasons."
Agent 2.1
Strategic Alignment Scorer
Problem It Solves
Work gets prioritized based on who asked loudest, not whether it moves the business. OKRs are written in January and irrelevant by March.
What It Does
Maps each intake record to active OKRs, product principles, and stated company bets. Produces an alignment score with rationale. Flags submissions that conflict with or have no connection to strategic priorities.
Value In One Line
Every piece of work enters the system with a documented reason it matters — or a flag that it doesn't.
Agent 2.2
Stakeholder Routing Agent
Problem It Solves
Requests land with the wrong team, the wrong PM, or nobody at all. Work sits unowned while the submitter assumes someone is on it.
What It Does
Uses product taxonomy, org structure, and historical ownership patterns to recommend the right PM, product area, and stakeholder reviewers. Notifies automatically and logs the routing decision.
Value In One Line
Every submission gets to the right person without a coordinator in the middle.
Layer 3
Plan
"Quarterly planning is driven by advocacy, not evidence. Nobody asks which initiatives compound each other."
Agent 3.1
Portfolio Scoring Agent
Problem It Solves
Every initiative gets scored in isolation, inconsistently, by the person who wants it built. There's no shared rubric, no cross-initiative comparison, and no audit trail.
What It Does
Applies an 8-dimension scoring matrix to all in-scope submissions simultaneously. Weights are configured per quarter based on planning mode. Runs in preview mode during open window; full final run triggers at Planning Lock.
Value In One Line
The ranking reflects the strategy and the evidence — not the org chart.
Agent 3.2 · Differentiator
Initiative Clustering & Sequencing Agent
Problem It Solves
Teams pick initiatives independently without asking which ones are better together. Shared infrastructure gets built twice. Quick wins that unlock bigger bets get deprioritized because they look small in isolation.
What It Does
Reads scored profiles — not just ranks — to find dependency chains, natural bundles, foundation bets, and stranded investments. Recommends which initiatives should be done together and in what sequence to maximize portfolio value, not just individual initiative value.
Value In One Line
Turns a prioritized list into an intelligent sequence — the roadmap tells a coherent story instead of being a stack rank.
Agent 3.3
Planning Scenario Modeler
Problem It Solves
Planning produces one roadmap and everyone defends it even when conditions change. There's no lightweight way to model tradeoffs before committing.
What It Does
Generates 2–3 roadmap scenarios from the scored, clustered portfolio using different strategic weight configurations. Shows what moves in and out under each scenario. Can be re-run mid-quarter when a major input changes.
Value In One Line
Leadership chooses between informed options instead of building a roadmap from scratch in a conference room.
Agent 3.4
Planning Prep Agent
Problem It Solves
Planning meetings start with everyone staring at a Jira board trying to remember context. The first 45 minutes are recap, not decision-making.
What It Does
Generates the planning pre-read from gate outputs: current scores, recommended sequence, scenario options, capacity summary, and the specific decisions that require human judgment. Everything else has already been handled.
Value In One Line
The planning meeting starts at the decision, not the recap.
Layer 4
Monitor
"PMs hand off to engineering and go dark — then get surprised when something slips, changes, or quietly breaks."
Agent 4.1
Delivery Pulse Agent
Problem It Solves
PMs are flying blind between kickoff and launch. Scope changes, blocked tickets, and missed dependencies surface too late for anyone to course-correct without a crisis.
What It Does
Watches sprint progress, ticket status changes, PR activity, and blocker flags across active initiatives. Sends PMs a lightweight digest — what's on track, what's drifting, what needs a decision. Escalates anomalies proactively before they become misses.
Value In One Line
PMs stay informed without attending every standup — and can intervene before a slip becomes a miss.
Agent 4.2
Decision Capture Agent
Problem It Solves
Critical product decisions get made in Slack threads, standups, and sidebar conversations — and vanish. Six months later nobody remembers why something was built the way it was.
What It Does
Monitors Slack channels, Jira comments, and Confluence pages for decision signals. Extracts the decision, rationale, and decision-maker, and writes it to a searchable decision log tied to the relevant initiative.
Value In One Line
Nothing important gets decided and forgotten again.
Layer 5
Learn
"Every sprint ends, every quarter closes, and the same mistakes get made again because nobody captured what actually happened."
Agent 5.1
Voice of Customer Synthesis Agent
Problem It Solves
Customer signal lives in support tickets, NPS responses, sales calls, CSM notes, and user research. Nobody synthesizes it. The same pain gets reported 40 times before anyone notices a pattern.
What It Does
Ingests signals from all connected customer-facing channels. Clusters by theme, tags to product area, tracks volume and urgency trends. Surfaces emerging patterns when a threshold is crossed and links them to relevant backlog items.
Value In One Line
The customer's voice enters every prioritization conversation, automatically.
Agent 5.2
Retro Intelligence Agent
Problem It Solves
Retrospectives produce sticky notes that get archived and never referenced again. The same delivery problems recur quarter after quarter with no systemic response.
What It Does
After each sprint or quarter close, pulls ticket outcomes, incident notes, delivery metrics, and retro inputs. Produces a structured summary — what shipped, what slipped, root causes, cross-retro patterns, and specific recommendations for the next planning cycle.
Value In One Line
Retros stop being a ritual and start being a real input to how the team works next.
Appendix A · Scoring
The PIL Scoring Matrix
Eight dimensions across three categories. All scored 1–5. Weights are configured per planning cycle based on the organization's current mode — not fixed across quarters. Strategic Alignment anchors at 20% in all modes; it is the non-negotiable baseline.
| Dimension | Category | What a 5 Looks Like | Growth Mode | Consolidation | Competitive Response |
|---|---|---|---|---|---|
| Strategic Alignment | Strategic Fit | Directly named in current company strategy or OKR | |||
| Market Timing | Strategic Fit | Clear time-sensitive window — competitor, regulation, or contract | |||
| Customer Impact | Value Creation | Transformative improvement for a core persona — validated by research | |||
| Revenue & Retention | Value Creation | Directly tied to a named revenue lever — new ARR, churn reduction, expansion | |||
| Competitive Differentiation | Value Creation | Creates a capability competitors can't easily replicate in 2 quarters | |||
| Effort & Complexity | Execution Reality | High value, low complexity — well-scoped, familiar tech, small team | |||
| Dependency Risk | Execution Reality | Fully self-contained — no external team, system, or decision dependencies | |||
| Confidence Level | Execution Reality | Validated problem, clear solution direction, well-scoped and ready to build |
Scoring Integrity Note
"The matrix only works if scores reflect reality, not advocacy. Evidence gating, agent-assisted scoring, consistency auditing, and historical calibration are what give these numbers teeth."
Scoring Integrity Layer — Five Mechanisms
1. Evidence-Gated Scoring
No score is accepted without an attached evidence input. The agent validates that evidence is real, retrievable, and actually supports the score claimed. Challenges with a specific rationale if it doesn't.
Level 1 · Start Here2. Agent-Assisted Scoring
Agents score first. Humans review and can challenge — but must submit a written justification that enters the permanent record. Shifts the burden from catching inflation to explaining disagreement.
Level 1 · Start Here3. Cross-Initiative Consistency Audit
Flags teams whose average scores are statistical outliers vs. the cohort. Identifies dimensions where distributions are implausibly skewed. Surfaces patterns as calibration flags — not accusations.
Level 24. Historical Accuracy Scoring
Post-launch reconciliation compares projected scores to actual outcomes. PMs and teams build a calibration track record over time. Consistent over-scorers receive automatic adjustments on future submissions.
Level 35. Blind Comparative Review
For high-stakes cycles: initiatives presented anonymously to a small review group. Agent aggregates independent scores, surfaces high-variance items (genuine disagreement), and flags outliers vs. group median. A 30-minute async exercise that surfaces where the real debate lives.
Level 3Appendix B · Planning Lock
The Planning Cycle Gate Model
The Planning Lock is the moment the scoring agent runs its full final pass. Everything before it is designed to give teams the best possible chance to make their case. Everything after it is designed to hold the line. Five phases govern when each component activates.
Phase 1
Open Submission Window
Weeks 1–3
- Intake agent fully active
- Duplicate detector runs continuously
- Scoring runs in preview mode — directional only
- No rankings visible across teams
Phase 2
Submission Freeze
3–5 Days Pre-Lock
- No new submissions accepted
- Minor evidence edits only
- Pre-gate consistency audit runs
- Teams notified of integrity flags
Phase 3
Planning Lock
Gate Day
- Full final scoring run
- Clustering agent runs immediately after
- Scenario modeler generates options
- Record locked with full audit trail
Phase 4
Planning Discussion
Human Layer
- Planning Prep Agent generates pre-read
- Humans decide between scenarios
- Resolve high-variance debates only
- Exception process handles post-lock requests
Phase 5
Post-Planning Lock
After Roadmap Confirmed
- Final scores logged as permanent record
- Historical accuracy tracking begins
- Deferred items tagged for next cycle
- Baseline set for post-launch reconciliation
Appendix C · Exceptions
The Exception Process
The Planning Lock holds. And legitimate urgency still gets a path forward. The exception process isn't a loophole — it's a pressure valve with a paper trail. Only four recognized triggers qualify. Everything else waits for the next cycle.
Type 1
Strategic Shift
Material change in company direction from executive leadership that makes a roadmap item less relevant and a new one urgent. Requires CPO or CEO sign-off.
Ex: Company pivots to enterprise mid-quarter. A new integration becomes critical path.
Type 2
Competitive Emergency
Direct competitor ships a capability creating demonstrable risk to pipeline, retention, or market position. Evidence required — not a feeling. Sales data or specific customer escalation.
Ex: Competitor releases a feature your top 10 accounts have been requesting. Two flag it in the same week.
Type 3
Customer or Revenue at Risk
A specific named customer or revenue commitment is at risk without this initiative. Dollar amount and timeline must be specified. CSM or Sales leadership co-signs.
Ex: $400K renewal contingent on a specific capability being delivered this quarter.
Type 4
Compliance or Legal Obligation
Regulatory deadline, legal requirement, or security obligation not known at planning time. Legal or compliance team co-signs.
Ex: New data residency regulation announced with a 90-day compliance window.
Five-Step Exception Flow
1
Formal Submission
Full intake record required — same standard as pre-lock. Urgency rationale, exception type, and executive sponsor all mandatory. No Slack messages to the CPO. No verbal asks in meetings.
2
Immediate Impact Assessment
Scoring agent runs immediately — no preview mode. Produces three outputs: the exception score vs. the locked portfolio; the displacement analysis (what comes out if this goes in); and the dependency check.
3
48-Hour Exception Review
Time-boxed to 48 hours. Attendees: CPO or VP Product, executive sponsor, planning owner, EM or CTO rep. Three outcomes: Approved, Approved with Deferral, or Denied with rationale.
4
Permanent Logging
Every exception — approved or denied — logged with submission date, submitter, type, agent scores, displacement analysis, decision-makers, rationale, and outcome. Visible to all planning stakeholders.
5
Retro Intelligence Flag
At quarter end, the Retro Intelligence Agent includes an exceptions summary: volume by type, approval rate, whether urgency claims proved accurate. Teams submitting repeated Type 1 exceptions have a strategy problem, not a planning problem.
Want to build this for your org?
The PIL is a real framework designed for real product orgs. I build and implement these systems — the agents, the scoring infrastructure, the planning processes. Let's talk.