MFP Agent is a working prototype of an agentic AI planning assistant designed for retail merchandise planners. It combines deep domain expertise in Merchandise Financial Planning (MFP) with modern AI reasoning to help planners identify risk, manage inventory profitability, and make faster, more confident decisions.
The agent was designed, specified, and built as a portfolio project to demonstrate how AI can augment — not replace — the way retail planners work.
Merchandise planners spend significant time each week analyzing the same questions:
Am I on track to hit my sales plan?
Where am I overstocked, and what's the margin at risk?
Should I cancel open POs, push out receipts, or take a markdown — and which option costs me less?
Do I have a Basic item problem or a Fashion item problem? (The answer determines the entire action path.)
These questions require pulling data from multiple sources, doing mental math across sell-through rates, OTB positions, and EOS projections — and then making a judgment call. MFP Agent does that analysis in seconds and walks the planner through a structured recommendation.
The agent is loaded with a department's full planning data — class-level sales actuals, LY comparisons, on-hand inventory (cost and retail), on-order flow (cost and retail), plan targets, and gross margin goals.
When a planner asks a question, the agent follows a structured reasoning chain built from real MFP domain expertise:
Triage at Department — Is the department on pace? What's the projected season-end vs. plan?
Drill to Class — Which classes are driving risk? What's the sell-through variance?
Classify: Basic vs. Fashion — This determines the entire action path. Basic items have different levers than Fashion items.
Basic Item Path — Can we cancel or push out POs? Are there understocked stores that can absorb inventory before we consider a markdown?
Fashion Item Path — Is inventory allocated to the right stores, in the right size curve? Reallocation before markdown. If markdown is needed, what timing and depth minimizes margin damage?
Margin Quantification — Every recommendation includes a $ impact so the planner can make a business decision, not just a gut call.
Allocation Handoff — The agent explicitly flags when an action requires the Allocation team, respecting the real organizational boundary between MFP and store planning.
This isn't a chatbot with retail vocabulary. The reasoning logic reflects how experienced planners actually think — the same logic used in Oracle RPAS-based MFP implementations at retailers ranging from mid-market to Fortune 500.
The design principle: AI should reason like a senior analyst, not retrieve like a search engine.
For retail technology companies: A clear point of view on how AI should be embedded in planning workflows — domain-grounded, action-oriented, and aware of organizational boundaries (MFP vs. Allocation vs. Markdown Planning).
For product management: End-to-end product thinking — from data model design to reasoning logic to user experience — built on 15+ years of MFP implementation expertise across Oracle, JustEnough, and Logic Information Systems engagements.
For AI literacy: Practical understanding of how large language model APIs work, how system prompts encode domain reasoning, and how agentic AI differs from traditional planning software.
Vercel (user interface)
Claude API by Anthropic (AI reasoning engine)
Sample data: Women's Sportswear, Spring 2025 — Department + 4 classes (Basic and Fashion mix) with realistic cost/retail values, on-order flow, and margin targets
Designed and specified by Elizangela Martins. Built as a portfolio project exploring the intersection of retail planning domain expertise and agentic AI.