01 — Project Overview
Sales Revenue
Management
Sales Revenue Management (SRM) is one of the most data-intensive functions inside a consumer goods organization. Pricing, promotional effectiveness, sales performance, pack formats, product mix, and market dynamics constantly change, making timely decision-making critical.
Traditional analytics workflows require analysts to identify potential causes, gather data from multiple sources, write SQL queries, create charts, validate statistical significance, prepare reports, and communicate recommendations. This manual workflow often delays decisions and creates heavy dependency on technical teams.
"Ask SRM reimagines this process as an AI-powered Commercial Decision Intelligence Platform. Instead of asking users to navigate dashboards, it allows them to ask natural-language business questions while an orchestration of AI agents investigates, validates, visualizes, and explains the underlying business drivers with complete transparency."
02 — Business Context
Commercial Analytics
Complexity
The commercial organization at MARS relies on numerous business metrics, including Net Sales Value (NSV), Gross Sales Value (GSV), promotional effectiveness, pricing, sales volume, brand performance, and product segmentation. When these metrics fluctuate unexpectedly, business leaders need immediate answers:
Answering these questions traditionally required multiple analysts working across BI tools, spreadsheets, SQL environments, and presentation software. The challenge was not data availability—it was making that data understandable, explainable, and actionable for decision-makers.
03 — Problem Statement
The Commercial
Data Bottleneck
Commercial decision-making suffered from four key challenges:
Analytics Silos
Business insights were spread across dashboards, SQL queries, Excel reports, and internal documents.
Technical Reliance
Commercial users relied heavily on analytics teams to write queries and validate hypotheses.
Delayed Investigations
Straightforward questions often required hours of manual exploration across multiple tools.
Black-box AI
Without visibility into reasoning, underlying queries, or evidence, recommendations were difficult to trust.
04 — Strategic Objectives
Primary Goals
05 — Interactive Prototype
Ask SRM AI
Agent Simulator
Select a commercial query from the buttons below to trigger the 4-step AI agent orchestration, demonstrating progressive disclosure from narrative to technical validation.
Select a preset commercial question above to launch Ask SRM.
06 — Responsibilities
My Design Role
As the Product Designer, I was responsible for designing a conversational analytics experience that balanced AI automation with enterprise transparency:
07 — Success Metrics
Business Outcomes
08 — Primary Users
User Personas
Revenue Managers
Monitor pricing, revenue, promotions, and overall portfolio profitability margins.
Sales Leaders
Track commercial performance across brands, countries, and major retailers.
Category Managers
Understand brand movement, product mix variations, and pack format performance.
Pricing Analysts
Investigate pricing impact and recommend price realization strategies.
09 — Research & Discovery
Behavioral Patterns
We identified several recurring behavioral patterns from research sessions with MARS commercial planners:
- Users think in questions, not reports: "Why did NSV decline?" over "Open dry food sales report."
- Trust requires explainability: Users want to validate AI outputs rather than blindly accept black-box advice.
- Layered outputs are mandatory: Executives value concise narratives, while analysts require technical SQL, Python code, and statistical significance validation.
10 — UX Design Principles
Core Principles
11 — User Journey
End-to-End Journey
The platform translates analysis from a sequence of dashboard clicks into a structured progression:
12 — Information Architecture
SRM Ecosystem IA
Ask SRM is structured into a persistent workspace layer connected to existing commercial modules:
360 Cockpit
Global commercial health dashboards and brand performance overviews.
Promo Effectiveness
Analysis of trade spend, discount depths, and incremental volume ROI.
Pricing Simulator
Scenario testing for list-price updates and pack format mix margin changes.
13 — Multi-Agent Orchestration
Intelligence Architecture
Instead of relying on a single LLM, Ask SRM distributes analysis across specialized sub-agents:
14 — Progressive Complexity
Three Levels of
Disclosure
To support diverse user roles (executives, analysts, developers), AI responses reveal detail progressively:
Level 1: Executive Summary
One-paragraph summary explaining the business outcome, drivers, and recommendations.
Level 2: Supporting Insights
Key business drivers, interactive charts, and contribution waterfall diagrams.
Level 3: Technical Validation
Direct access to generated SQL code, Python transformations, and statistical p-value validation tests.
15 — Root Cause Analysis
AI-Powered Root
Cause Factors
When commercial teams investigate declining sales value, the platform decomposes the variance into six contributing factors:
16-17 — Enterprise Trust & Design System
Trust UX & System
Commercial decisions influence millions of dollars in revenue. Ask SRM was designed to encourage trust through continuous human validation loops: rating AI responses, regenerating logs, refining parameters, and tracing evidence back to source tables.
Maintaining consistency across SRM modules required reusable UI assets: KPI cards, filter bars, AI conversation panels, recommendation checklists, and investigation timeline logs.
18 — Challenges Mapped
Design Challenges
Balancing Simplicity with Depth
Exposing highly technical statistical validations (ANOVA, regression correlation) without overwhelming category managers.
Reducing Dashboard Fatigue
Integrating conversational shortcuts directly into existing dashboards, avoiding the need for another standalone tool.
19 — Reflections & Summary
Key Learnings
This project demonstrated that successful enterprise AI experiences are not defined by conversational interfaces alone—they are defined by how effectively they support decisions. Exposing explainability, code correctness, and source attribution proved far more impactful than simply increasing model parameters.