All work
PrototypeAnonymized

Trade Policy Intelligence / Independent product case

Tariff Impact Navigator

The useful answer is not the tariff headline. It is the trace from policy evidence to exposure, scenario, and accountable response.

How a fragile no-code experiment became a tested multi-workflow prototype designed for repeatable procurement analysis.

My mandate
AI Product & Portfolio Lead · decision trace, evaluation gates, and scale-up direction
Verified product evidence
Delivered a working prototype combining policy research, impact analysis, scenario planning, and reporting.
  • Policy intelligence
  • Scenario product
  • Deterministic + generative AI
  • Trust before automation
Public-safe reconstruction of the bounded prototype workflowIllustrates product logic, not employer data
Primary user
A procurement, category, finance, or trade-risk leader responding to a policy development.
Decision enabled
Whether the exposure is real or potential, which scenario requires response, and who should validate, mitigate, or monitor it next.
My mandate
Connect a volatile external-policy problem to a bounded, testable business workflow and define the trust gates required before automation and enterprise handoff.
Two evidence layers
External policy plus exposure data
Multi-turn
Research, calculation, and scenarios
Quality loop
Regression and human review

Evidence first

What is proven, what I changed, and what remains open.

A 90-second evidence map for hiring review. Synthetic product views explain mechanics; they never count as outcome proof.

Verified now
A working prototype connecting current-policy research, deterministic exposure calculation, scenario comparison, and source-backed communication.
Next evidence gate
SME-led UAT, governed refresh, enterprise data connection, and acceptance evidence before automated monitoring.
01

User / workflow evidence

Verified record

Requirements and prototype review showed that policy news alone could not answer which material or business exposure mattered, while the earlier workflow was inconsistent and difficult to scale.

02

Product decision

Verified record

I separated policy truth from exposure math, treated scenarios as explicit assumptions rather than forecasts, and held automated refresh behind an evidence gate.

03

AI / system boundary

Supported synthesis

AI retrieves and interprets policy evidence; deterministic logic resolves and calculates exposure; the accountable owner validates assumptions, exceptions, and the communication decision.

04

Evaluation → change

Verified record

Grounded-research checks, calculation regression, repeated variants, model-assisted review, and human product review converted relevance, status, exposure, and communication failures into separate release controls.

05

Outcome / boundary

Verified record

The bounded policy-to-impact workflow and trust-before-automation roadmap are evidenced. Real-time refresh, production adoption, and realized savings are not claimed.

AI product decision record

The choices behind the product, architecture, and release gate.

Use AI for changing policy evidence; use deterministic computation for exposure and scenario math.

01

AI fit

Supported synthesis

Use an LLM to find, interpret, and explain policy evidence while code calculates business exposure.

Policy language is unstructured and changes over time, but financial exposure must remain reproducible.

Alternative consideredPure rules cannot interpret every policy variation; model-generated arithmetic is too fragile for the decision.

02

Knowledge & context

Supported synthesis

Use dated RAG over approved primary policy sources, then resolve entities and rates into a governed exposure model.

Freshness, jurisdiction, effective date, and source authority determine whether the answer is valid.

Alternative consideredModel memory or undated web search cannot support a current-policy claim.

03

System architecture

Supported synthesis

Use a sequential research → resolve → calculate → compare → communicate workflow with explicit checkpoints.

Each step has a different error class and needs a visible handoff before the next one compounds it.

Alternative consideredAutonomous monitoring, multi-agent routing, or MCP-connected actions remain behind freshness, permission, and SME acceptance gates.

04

Evaluation & release

Verified record

Separate policy groundedness, calculation regression, scenario assumptions, repeated variants, and human communication review.

A current policy answer can still yield wrong exposure, and correct math can still communicate the wrong decision.

Alternative consideredOne combined score would obscure which control failed and whether the product is safe to refresh automatically.

Evidence I still need from the private project record3 open items
  1. 01

    SME UAT findings and accepted thresholds.

  2. 02

    A current-policy freshness and fallback receipt.

  3. 03

    Measured decision-time, exposure-quality, or operational-use evidence.

Verified = direct approved recordSupported = defensible synthesisReported = source receipt pendingDesigned = future test or control

The product contract

One user moment, carried all the way to a decision.

A tested policy-to-impact workflow that connects current trade developments with material exposure, scenario analysis, and an accountable business brief.

01 / User
A procurement, category, finance, or trade-risk leader responding to a policy development.
02 / Trigger
A new or changing trade measure may affect material exposure and leadership needs a defensible scenario before acting.
03 / Inputs
Policy question · origin and destination · material or category · timing · exposure lens · scenario rate
04 / Product action
Classify policy status and timing, keep external evidence separate from business exposure, resolve the calculation inputs, run deterministic scenarios, and explain the implications with assumptions visible.
05 / Output
A source-backed policy record, exposure calculation, scenario comparison, audit trail, management brief, owner, and proposed next action.
06 / Human decision
Whether the exposure is real or potential, which scenario requires response, and who should validate, mitigate, or monitor it next.

Customer-grade product walkthrough

See one decision move through the product.

Public-safe representative scenario

A new trade measure is announced. A category leader needs to know whether it is effective or potential, what exposure may be affected, and how a changed rate would alter the scenario.

Tariff Impact NavigatorPolicy-to-impact workspace
ScenarioSYN-TIN-017
Simulated product scenario
01 / Policy signal

Establish what changed before calculating what it might mean.

The workflow distinguishes an effective measure from a proposal, threat, agreement, or de-escalation and retains the source date.

01 / 06
External signal
A fictional notice proposes an additional duty on composite material moving from Region A to Region B.

Illustrative event · status and effective timing not yet confirmed

How the product handles this state

  1. 01

    Prioritize official evidence over repeated reporting.

  2. 02

    Classify status, trade direction, category, and timing.

  3. 03

    Separate confirmed facts from emerging interpretation.

Decision signals

Policy status
ProposedNot yet an effective rate.
Trade direction
Region A → BSynthetic origin and destination.
Timing
Review window openEffective date remains unresolved.
Product outputPolicy record created

The event is tracked as a proposed measure with unresolved timing, so downstream calculations must be labelled as scenarios.

  • Proposal is not represented as current policy.
  • Direction and category are explicit.
  • Source date travels with every downstream output.

Select a state or use arrow keys in the workflow. Content is synthetic and public-safe.

All policy events, materials, rates, spend values, sources, and outputs shown here are synthetic. This public-safe walkthrough demonstrates product logic, not live legal advice, employer exposure, or production coverage.

Showing Classify the policy, step 1 of 6.

The complete product

More than a generated answer.

The product earns trust by combining a defined decision contract, inspectable evidence, deterministic controls, AI assistance, and an accountable handoff.

01Implemented in bounded prototype

Policy-state intelligence

Classifies a development as effective, proposed, threatened, agreed, or de-escalating with timing and trade direction attached.

Why it matters

The product prevents a headline from being treated as an implemented rate or a settled forecast.

02Implemented in bounded prototype

Evidence separation

Keeps external policy truth, internal exposure data, and scenario assumptions as distinct reviewable layers.

Why it matters

Each layer has different ownership, freshness, and failure modes.

03Implemented in bounded prototype

Exposure resolver

Clarifies origin, destination, material, tier, timing, exemption, business unit, and aggregation lens before calculation.

Why it matters

A policy rate alone does not reveal which spend or flow is affected.

04Implemented in bounded prototype

Deterministic scenario engine

Applies an explicit rate and scope assumption, calculates exposure, and compares scenario deltas reproducibly.

Why it matters

Generative AI does not own the math or hide the assumptions behind a fluent answer.

05Implemented in bounded prototype

Source-backed management brief

Explains the policy state, calculation basis, scenario implication, uncertainty, and recommended review path.

Why it matters

Leadership receives a decision record that can be challenged and updated as evidence changes.

06Public-safe product design

Governed refresh and action

Defines how policy and exposure updates should be reviewed, versioned, assigned, and handed into the enterprise workflow.

Why it matters

Automation follows a trust and ownership model rather than turning changing news directly into action.

AI operating model

The model assists. The product controls. The human decides.

Separating responsibilities is both a system design and a product leadership choice. It defines what can be tested, what must be reviewed, and where accountability stays.

01
Deterministic

Protect exposure math, scenario reproducibility, and auditability.

  • Policy status and effective-date fields
  • Origin, destination, category, tier, and exemption rules
  • Exposure aggregation and rate application
  • Scenario delta and version record
02
AI-assisted

Find, classify, and communicate changing policy evidence.

  • Policy retrieval and relevance filtering
  • Event and uncertainty classification
  • Multi-turn intent routing
  • Source-backed management synthesis
03
Human-owned

Own the assumptions, exceptions, and business response.

  • Confirm exposure scope and data
  • Validate policy interpretation
  • Approve the scenario and communication
  • Assign mitigation or monitoring action

The broken workflow

From policy change to material exposure and scenario response

Rapid policy changes forced procurement teams to repeatedly interpret external news, update tariff assumptions, calculate exposure, and rebuild communications across materials and geographies.

01Volatility

The external facts changed faster than a manual tracker could support.

Teams needed current, reputable policy evidence while distinguishing measures already in effect from proposals, threats, agreements, or de-escalation.

02Exposure

A policy announcement did not reveal which business exposure mattered.

The analysis still had to resolve trade direction, material context, tier, spend, exemptions, and the user’s aggregation lens.

03Trust

The earlier no-code experience was inconsistent and difficult to scale.

A custom workflow was needed to route research and calculations separately, preserve assumptions, and test the output against reviewable evidence.

Product decisions and rationale

Product judgment lives in the choices and boundaries.

These records reconstruct the evidence-backed product logic: direction, competing alternative, rationale, and consequence. They are not presented as verbatim decision minutes.

01Policy news assistant vs. policy-to-impact product
Product direction

One decision trace from policy evidence through exposure, scenario, communication, and owner action.

Rejected alternative
Optimize a standalone news summarizer that stops before business consequence.
Why
The user value begins when a verified event can be connected to a reviewable business decision.
Product consequence
Research, calculation, scenario, and reporting became one accountable product workflow.
02Generative calculation vs. deterministic exposure math
Product direction

Use AI for research, classification, and explanation; use reproducible logic for calculation and aggregation.

Rejected alternative
Ask the model to infer and calculate the final impact inside a narrative response.
Why
Calculation correctness, repeatability, and auditability are non-negotiable in a high-consequence workflow.
Product consequence
The evaluation model could test math independently from language quality.
03Scenario as prediction vs. scenario as explicit assumption
Product direction

Label the policy state and let the user compare named rate and scope assumptions.

Rejected alternative
Present one probable future as the expected outcome.
Why
The product must support decisions under uncertainty without fabricating policy certainty.
Product consequence
Assumptions, deltas, and evidence changes remain visible and updatable.
04Automate refresh immediately vs. hold automation behind trust
Product direction

Prove the bounded routing, calculation, fallback, and review model before automating rates and exposure refresh.

Rejected alternative
Scale ingestion and access while core failure modes remained difficult to observe.
Why
Automation multiplies both useful throughput and unverified error.
Product consequence
UAT, freshness, auditability, ownership, and fallback became explicit investment gates.

Evaluation → product change

A failure is useful only when it changes the product.

The bounded prototype completed functional, regression, model-assisted, and human-supervised evaluation. The public case uses synthetic exposure and omits source names, formulas tied to employer data, scores, and test volumes.

Representative failure casePublic-safe representative case

A public announcement contains a proposed rate, an uncertain effective date, and possible exemptions while the business exposure record has incomplete origin and tier data.

Initial failure
A system can combine proposal and exposure into a precise-looking impact number that implies more policy certainty and data completeness than exists.
Designed control
Separate status, evidence, exposure, and scenario layers; require explicit assumptions; block or qualify calculations when policy or exposure fields are unresolved.
Acceptance test
The product must label the measure as potential, show which exposure assumptions are incomplete, and calculate only an explicitly named scenario with a traceable formula.
Acceptance gate
No output may present scenario impact as actual exposure unless policy status, effective scope, exposure fields, and calculation version are all reviewable.
Representative evaluation contract

What changed in the selected trade measure, which exposure could be affected, and how does the impact change under two explicit rate scenarios?

Expected evidence
Relevant policy evidence with status and timing, the correct exposure slice, the applicable assumptions, and reproducible scenario calculations.
Expected behavior
Exclude unrelated policy content, distinguish actual from potential, resolve or expose missing inputs, calculate outside the language model, and communicate the result with sources and uncertainty intact.
Release gate
Functional and regression checks must reproduce the calculation; model-assisted and human review must confirm grounded policy interpretation, uncertainty, completeness, and professional usefulness.
Quality dimensions
  1. 01Policy retrieval relevance
  2. 02Status and timing classification
  3. 03Exposure-scope correctness
  4. 04Formula and aggregation fidelity
  5. 05Scenario reproducibility
  6. 06Assumption visibility
  7. 07Multi-turn routing
  8. 08Business communication quality

My product leadership

From an ambiguous ask to an investable next gate.

Connect a volatile external-policy problem to a bounded, testable business workflow and define the trust gates required before automation and enterprise handoff.

Team model
Matrix collaboration across the product owner, procurement and trade SMEs, finance and exposure stakeholders, data, engineering, and AI delivery. No direct-report relationship is claimed.
My decision rights
I shaped the end-to-end decision trace, prototype scope, AI-versus-calculation boundary, evaluation model, product-owner checkpoints, and scale-up recommendation; business owners retained policy, financial, and investment accountability.
Prototype delivery record
The roadmap defined the transition from static bounded analysis to governed refresh, business validation, production-grade interface, fallback, auditability, and accountable ownership.
  1. 01

    Reframed a fragile no-code experiment as a multi-workflow product joining policy, exposure, scenario, and communication.

  2. 02

    Established separate evidence and test contracts for policy interpretation and deterministic impact calculation.

  3. 03

    Made uncertainty, missing exposure fields, and unsupported policy claims explicit fallback states.

  4. 04

    Used bounded workflow evidence to sequence SME UAT, governed data refresh, audit trail, ownership, and enterprise handoff.

Value without invented outcomes

Separate what worked from what scale must prove.

The prototype evidence supports product capability. Adoption, ROI, and realized business impact remain separate claims until measured and approved.

01 / Current workflow

Teams repeatedly monitor changing policy, interpret status and timing, locate the relevant exposure, rebuild calculations, compare scenarios, and rewrite leadership communications.

02 / Validated capability

The prototype joined policy research, impact calculation, scenario analysis, and reporting in one multi-turn workflow with repeated technical and human-supervised evaluation.

03 / Value hypothesis

A scaled product should shorten the path from policy signal to reviewable exposure, reduce inconsistent assumptions, and improve the speed and traceability of mitigation decisions.

04 / MVP success measures
  • Time from policy signal to reviewable scenario
  • Policy-classification correction rate
  • Calculation reproducibility
  • Exposure-coverage completeness
  • SME usefulness and revision rate
  • Time to assigned mitigation or monitoring action

Portfolio leverage

Specialist where it matters. Shared where it scales.

Policy-state semantics, exposure mapping, deterministic scenario logic, and high-consequence review require a specialist product contract beyond generic research or chat.

Shared through Procurement AI Hub
  • Identity and access
  • Retrieval and citations
  • Evaluation and observability
  • Review, action assignment, and enterprise handoff patterns
Kept specialist in this product
  • Policy status and timing model
  • Trade direction and exposure semantics
  • Scenario calculation and audit trail
  • Risk communication and mitigation workflow
Next investment gate

Automate refresh and broaden access only after policy usefulness, exposure interpretation, calculation fidelity, fallback, and ownership pass SME-led UAT.

Evidence boundary

Core workflow proven. Scale is a separate product decision.

01Bounded core workflow implemented and evaluated

What exists now

  • Working prototype joining current-policy research, exposure calculation, scenario planning, and reporting.
  • Multi-turn routing between research and impact workflows with a consolidated response.
  • Repeated functional, regression, automated, and human-supervised testing.
02MVP scale-up

What must become true next

  • SME UAT for policy relevance, impact interpretation, and ease of use.
  • Governed automation for policy rates and exposure-data refresh.
  • Production-grade interface, fallback, audit trail, ownership, and handoff.
03Not claimed

What this case does not imply

  • No claim of live production use, continuous real-time coverage, or realized cost avoidance.
  • No internal spend, material, supplier, policy-source, prompt, architecture, or quality-score details are published.

Evidence register

The work behind the visible product.

These records map each public claim to the approved requirements, demonstrated workflow, evaluation practice, or prototype-delivery record behind it. They are evidence references, not public copies of protected source files.

01Requirement

Policy-to-impact contract

Research, status, exposure, scenario, communication, uncertainty, and non-goals.

Verified project evidence
02Product workflow

Research and calculation split

Separate policy evidence and deterministic exposure paths joined in a reviewable brief.

Public-safe reconstruction
03Evaluation

Dual quality system

Policy relevance and classification plus calculation fidelity, regression, and human usefulness review.

Verified project evidence
04Handoff

Trust-before-automation roadmap

SME UAT, data refresh, audit trail, fallback, ownership, product UI, and enterprise integration.

Verified project evidence

Next product gate

The remaining work is explicit.

  1. 01

    Validate news usefulness and impact interpretation with SMEs

  2. 02

    Automate governed rate and exposure data refresh

  3. 03

    Harden fallback, auditability, and enterprise handoff

Part of the broader product directionProcurement AI Hub