01Observed
Conversations
An AI workspace that can research, cite sources and suggest downstream work items.
Retrospective PRDOutside-in21 August 2026
From scattered signal
to a decision teams
can defend.
A user-friendly reconstruction of Picnic's current product intent, grounded in the role brief and the live beta—not a claim to know the founders' internal specification.
01 / The product decision
Problem
AI made building faster. It did not make judgement easier.
Product evidence remains scattered across customer conversations, commercial systems, research, analytics and delivery tools. Teams can now produce convincing artefacts—and working software—before the decision underneath them has earned confidence.
The smallest useful proofA Product Manager starts with one real signal, makes a human-owned decision whose evidence can be challenged, and produces a living PRD and safe delivery plan without losing the reasoning.
02 / Current state
The beta already expresses the shape of a decision system.
These capabilities were visible in the live workspace after sign-in. They are observations, not assumptions about Picnic's roadmap.
02Observed
Opportunities
A surface generated from customer signals, with structured opportunity thinking.
03Observed
Create
Templates for PRDs, Opportunity Briefs, User Stories and a PM Co-Pilot.
04Observed
Context
Objectives, products, projects and 49 potential integrations across the product stack.
Product tension
A new workspace presents a broad setup path before the core promise is experienced. The initial wedge should prove one evidence-linked decision first; organisational context and integrations can then make that loop stronger.
03 / The core loop
One chain of understanding.
Each stage should preserve the intent, evidence and ownership established before it.
Signal
A customer conversation, commercial pattern or operational observation.
Evidence
Sources are attributable; inference, contradiction and missing context are visible.
Opportunity
The problem, impact, strategic fit, effort and risk become challengeable.
Decision
A named human weighs alternatives and owns the call.
Delivery
The decision becomes a living PRD, safe plan and clear system design.
Outcome
Reality is compared with the original expectation and becomes new evidence.
04 / Requirements
Build trust through the whole product loop.
Open each section for the prioritised requirements. P0 proves the wedge; P1 strengthens repeatability; P2 compounds learning after the core loop is trusted.
01 / StartReach one useful decision before broad setup+
A new user should be able to begin with one real signal and understand what additional context would improve the decision.
ACT-01P0Start from pasted text, an uploaded artefact or an existing conversation without connecting an integration.
ACT-02P0Ask for the decision, intended outcome, affected user and decision deadline before generating a solution.
ACT-03P0Return a useful first synthesis in one session and state what material context is missing.
CTX-01P1Retain objectives, products, product areas and projects as durable organisational context.
02 / GroundMake the evidence inspectable+
Picnic should help the team understand why an opportunity exists without turning model output into apparent fact.
EVD-01P0Preserve source, author where available, timestamp and source-system permission context.
EVD-02P0Label direct evidence, model inference, assumption, contradiction and unanswered question.
EVD-03P0Let users include, exclude or challenge evidence and record the reason.
INT-01P1Suggest integrations progressively, based on the live decision rather than catalogue breadth.
03 / DecideTurn opportunity into an accountable choice+
The decision record is the product's trust-bearing object: alternatives, uncertainty and ownership must survive the AI synthesis.
OPP-01P0Show supporting, opposing and missing evidence against the opportunity.
DEC-01P0Record owner, participants, deadline, expected outcome and affected product area.
DEC-02P0Present at least one credible alternative, including the option not to build.
DEC-03P0Require explicit human approval before a recommendation becomes a decision.
04 / DeliverCarry intent into buildable work+
The approved decision should produce maintained artefacts, not a one-off document that immediately drifts from delivery.
ART-01P0Create a PRD covering problem, outcome, scope, flow, requirements, edge cases, metrics and open questions.
ART-02P0Link material requirements back to the decision and its evidence.
ART-03P0Confirm shared understanding between the product owner and engineer—human or agentic.
DEL-01P0Derive a maintained plan of testable slices that leaves the product usable and protects against regression.
DEL-02P1Link a design document covering behaviour, interfaces, data flows, constraints and failure modes.
05 / LearnDetermine whether the decision was right+
Delivery is not the end of the loop. Picnic should reconnect actual behaviour and business outcomes to the original expectation.
OUT-01P0Define leading, lagging and guardrail measures when the decision is approved.
OUT-02P1Bring outcome evidence from analytics, customer channels and operations into the decision record.
OUT-03P1Prompt an outcome review and compare actual results with the expected outcome.
OUT-04P2Reuse validated learning when related opportunities arise.
05 / Non-functional trust
Data
Permission-aware retrieval, tenant isolation, controllable retention and minimum necessary context.
AI
Model output never masquerades as source evidence. Corrections preserve the underlying record.
Usability
Empty states identify the smallest useful action. Setup distinguishes essential steps from enrichment.
Performance
First synthesis feels conversational; longer work exposes progress, partial results and recoverable failures.
Observability
Tenant-safe logs explain ingestion, retrieval, model, permission, generation and synchronisation failures.
Change
Versioned decisions, PRDs and plans show exactly when evidence, scope or implementation diverges.
06 / Measures
North star
Evidence-linked decisions completed and reviewed per active product team.
Count only decisions with a named owner, inspectable evidence, an expected outcome and a later review.
Time to first decision
How quickly a new user reaches a useful decision; evidence coverage; opportunity-to-decision conversion.
Decision quality
Less avoidable rework; more outcome reviews; team expansion; design-partner retention and willingness to pay.
False confidence
Unsupported claims accepted, user corrections, permission incidents and speed gained while outcome quality falls.
07 / Delivery sequence
Each slice leaves a usable product.
Source/inference separation and an opportunity brief.
Alternatives, evidence, owner, rationale and approval.
Traceable requirements, versioning and shared understanding.
Usable slices, acceptance criteria and regression protection.
Outcome review and evidence reused in the next decision.
08 / Questions that could change this PRD
Where is the outside-in view wrong?
- 01Which single decision creates the most avoidable rework for current design partners?
- 02Who must experience the first value: the PM, Head of Product, engineering lead or commercial leader?
- 03Is the commercial promise better decisions, faster alignment, less rework—or faster delivery?
- 04What must a user achieve before Picnic considers a workspace activated?
- 05Which part of the loop is trusted today, and which remains an intended product direction?
- 06What evidence would cause the team to narrow—or reject—the proposed wedge?
The next validation is to watch design partners carry one consequential decision from raw signal to human approval, delivery and outcome review—and identify where trust, understanding or traceability breaks.