3-Week Decision-Intelligence POC

This project is password protected

← Back to Portfolio
Work About Resume
← Back

3-week UX strategy sprint / AI-assisted POC build

3-Week Decision-Intelligence POC

I turned a broad AI opportunity into a working product direction for commercial negotiation strategy.

View live build

Role

UX strategy + POC lead

Timeline

3 weeks

Users

Commercial negotiators

Output

Working POC + roadmap

Demo walkthrough of the ATLAS direction: priority signals, scenario controls, buyer context, and recommended negotiation moves.

POC demo / 1.5x

The Story

ATLAS started as a 3-week challenge: how far can UX push an ambiguous AI opportunity when the team builds quickly, listens closely, and keeps the product grounded in real decision-making?

01

The prompt was exciting, but too open-ended to build from.

PepsiCo had fragmented pricing, margin, market, buyer, and negotiation data. Without a strong product point of view, a fast AI sprint could have produced novelty instead of utility.

02

I pushed the sprint toward user decisions.

Instead of starting with what AI could generate, I focused the team on what users needed to decide: what changed, why it mattered, what move to make, and what evidence could defend it.

03

I created a research cadence to de-risk the product.

I framed weekly user sessions around discovery, co-design, concept trust, and final validation. Research was not a checkpoint. It was the engine that kept the POC grounded while we moved fast.

04

The product became proof of a faster UX operating model.

In 3 weeks, the team moved from a broad AI idea to a functional decision-intelligence POC, a sharper product thesis, and a set of roadmap choices grounded in user feedback.

The Problem

Context

Negotiation work did not follow a clean software workflow.

Work moved across Excel, PowerPoint, SharePoint, Outlook, customer systems, commodity data, meetings, calls, and human memory.

User need

Users had data. They needed confidence.

The real question was not "what does the data say?" It was "so what do we do now, and can we defend it under pressure?"

Product challenge

AI recommendations had to be explainable.

In a high-stakes commercial setting, a confident answer without provenance creates more risk than value.

Before vs. After

Legacy workflow

Manual synthesis before strategy.

  1. 01 / Fragmented data sources
  2. 02 / Manual synthesis
  3. 03 / Stakeholder debate
  4. 04 / Scenario guesswork
  5. 05 / Deck and email prep

ATLAS workflow

Evidence-backed decisions before the room.

  1. 01 / Signal detection
  2. 02 / Risk diagnosis
  3. 03 / Scenario comparison
  4. 04 / Buyer prediction
  5. 05 / Evidence-backed argument

POC architecture / decision loop

The loop that anchored the product direction.

I narrowed the POC around one repeatable decision flow: notice the change, understand the risk, compare possible moves, prepare the argument, then capture what happened so the system gets smarter.

01

Monitor change

What shifted in pricing, margin, market signals, customer ask, or source freshness?

02

Diagnose risk

Which negotiations are exposed, and where are target, red line, or relationship risks emerging?

03

Model scenarios

Compare recommended, conservative, aggressive, and user-authored moves with tradeoffs visible.

04

Predict buyer

Use buyer memory to anticipate likely objections, timing pressure, and response patterns.

05

Prepare argument

Package evidence, pushback responses, and audience-safe outputs for the next conversation.

06

Capture debrief

Turn negotiation outcomes into memory that informs future buyer guidance and scenario quality.

Debrief learning feeds back into buyer memory, source confidence, and the next scenario recommendation.

User Feedback → Product Capabilities

User feedback showed the problem was not access to more information. It was confidence in what to do next. Each major feature translated one feedback theme into a product capability.

ATLAS intelligent alerting screen showing prioritized buying group alerts and recommended scenario actions.

Feature 01

Intelligent Alerting

User insight

Users did not need another report; they needed "now what?" guidance from data.

How it shaped the build

The hub prioritized what changed, why it mattered, and what action should happen next.

ATLAS manual scenario controls screen showing adjustable levers and predicted buyer response.

Feature 02

Manual Scenario Controls

User insight

Scenarios need custom variables because buyer, market, and region context varies.

How it shaped the build

Editable assumptions, custom levers, and market/buyer-specific inputs let users adjust AI recommendations.

ATLAS buying group intelligence screen showing buyer response, negotiator read, and active risk signals.

Feature 03

Buying Group Intelligence

User insight

Buyer behavior is tracked today in spreadsheets and memory, but it is not truly intelligent.

How it shaped the build

Buying group workspaces turned history into guidance: behavior patterns, objections, handoff context, and relationship cues.

ATLAS predictive buyer response screen showing likely counter behavior, response likelihood, and recommended negotiation move.

Feature 04

Predictive Buyer Response

User insight

Negotiation prep is about building a defensible argument, not just collecting data.

How it shaped the build

ATLAS connected recommendations to proof, objection forecasts, pushback maps, and suggested responses users could bring into the room.

Prioritization under constraint

What we chose to build first.

We collected more feedback than the sprint could responsibly absorb. My role was to turn that feedback into product decisions: build the capabilities that unlocked trust, decision quality, and user confidence first, then move lower-fit or higher-risk ideas into the backlog.

Used to continue the build

01

Clarify the information architecture

Feedback showed the IA was hard to scan, so the MVP shifted toward a triage-first model that made priority signals easier to find.

02

Increase the trust layer

Every number and recommendation needed source, freshness, confidence, and known gaps, so trust became visible product behavior.

03

Prioritize predictive modeling

Users could only test a limited number of scenarios manually, so ATLAS focused on evaluating many paths and returning the strongest few.

04

Support brief generation

Commercial negotiators were already gathering documents and reports before key conversations, so generated outputs became a practical handoff moment.

05

Strengthen buying group profiles

Negotiators often work with the same buyers for years, making buyer memory a core input for closed-loop recommendations.

Saved for later

01

Live negotiation capture

High-value, but live listening introduced legal and implementation complexity that made it wrong for the MVP.

02

Account-manager feature depth

Account-manager needs were important, but the clearest first wedge was the commercial strategy workflow, with broader persona support kept in the roadmap.

03

Standalone market views

Market context mattered, but research pointed to alerting and filtering as the sharper MVP expression.

04

Deep scenario precision

Product-level and highly granular modeling were valuable, but the sprint needed to validate the core scenario engine first.

Product Direction

01 / Product thesis

AI as decision support, not content generation

Signal

Users did not need another report.

Move

Make ATLAS reason through risk, compare paths, and recommend next moves.

Result

A sharper product thesis the team could explain through user pressure and decision quality.

02 / Trust model

Trust as a product feature

Signal

Recommendations had to be defensible to leadership, account teams, and customers.

Move

Design source labels, confidence states, assumption editing, and evidence trails into the core flow.

Result

Trust moved from explanation copy into visible product behavior.

03 / Core capability

Buyer memory as active intelligence

Signal

Negotiators keep mental models of buyer behavior and relationship dynamics.

Move

Turn profiles into prediction inputs: likely objections, past-round patterns, and concession preferences.

Result

ATLAS became more than static context. It could guide strategy for the next round.

Working model

How we collaborated to move faster.

The POC also became a test of how UX designers can use AI tools and GitHub to work more like a product team: make decisions visible, build in shared code, and give engineering, product, and leadership one artifact to inspect together.

01 / Codex as build partner

Design moved from static intent to working behavior.

Codex helped turn product decisions into functional UI quickly, making interactions, edge cases, and system logic easier to evaluate than a flat mockup.

02 / GitHub as source of truth

The work became reviewable and collaborative.

Using GitHub created a shared record of decisions, iterations, and implementation details, so the team could discuss the actual build instead of translating between separate design and engineering artifacts.

03 / UX closer to delivery

Designers could contribute earlier to technical alignment.

This model let UX explore IA, flows, states, and product logic in a format the larger team could validate, refine, and eventually hand off with less ambiguity.

Outcome + Reflection

Outcome

In 3 weeks, ATLAS moved from an ambiguous AI opportunity to a validated product direction, working POC, and prioritized roadmap.

Product impact

The sprint turned design exploration into a concrete artifact the team could critique, refine, and use to make roadmap decisions.

What this shows

I can lead a fast, evidence-backed POC: define the problem, prioritize scope, translate strategy into working behavior, and make the future tangible.

Let's Connect!

goudreaugrace@gmail.com LinkedIn