What is AI-based assortment optimization?

AI-based assortment optimization is the ability to decide which products to keep, push or delist starting from the real margin of each category, channel and SKU, rather than from sales alone or habit. It means answering questions like "does this item create or destroy value, given the space it takes and the costs it generates?" with numbers, not intuition.

It is one of the main uses of a decision intelligence platform: software that integrates internal data and external variables, structures them into economic models and uses AI to make every decision systematic, scalable and measurable. Unlike a report or a spreadsheet, it doesn't just describe what sold: it shows where margin is generated and where it is eroded, and makes every assortment choice reliable, governed and justifiable with a number.

Why do retailers struggle to see where value is created and destroyed?

Because real margin stays hidden. Over 60% of retailers lack granular visibility of margin by SKU, category or store; assortment decisions often ignore the real saturation of shelf space; and more than 50% of allocations fail to correctly account for commercial cost drivers (source: McKinsey; BCG Pricing & Profitability). The result is a portfolio where items that look strong by revenue erode margin, and vice versa.

Behind this lies a broader problem: AI on its own amplifies how a company already decides. 72% of companies say they have adopted AI, but only 12% get measurable value (PwC Global AI Jobs Barometer 2025; McKinsey State of AI 2025). The recurring causes are four:

• Data fragmented in silos. Over 70% of company data isn't shared in an integrated way: sales, costs and stock stay in different departments, channels or people.
• Growing complexity. The relevant variables have tripled over the past decade and not all are monitored.
• Limited economic-impact analysis. Margin and profitability are estimated after the fact, not before changing the assortment.
• Department-level optimization. Purchasing, category and sales each maximize their own KPI: local optimization creates global sub-optimality.

In retail this weighs even more, because channels, logistics and sales are interdependent, decisions are made at finite shelf capacity (space and stock are real constraints) and margin is under pressure, with small changes in mix or price shifting EBITDA. Without an integrated decision system, AI risks amplifying inefficiencies instead of reducing them.

How does a decision intelligence platform make margin visible by category, channel and SKU?

By making margin a calculated figure, not an estimate. A decision intelligence platform links sales, costs and stock in a single model and returns profitability at different levels of detail, explaining its determinants:

• Margin by customer, category and SKU. Margin is reconstructed item by item, including direct and indirect costs (sourcing, logistics, cost-to-serve), not just the first commercial markup.
• Value and loss drivers. For each category it isolates what generates margin and what erodes it (price, purchase cost, mix, returns, promotional pressure), so actions hit the cause and not the symptom.
• Advanced segmentation and benchmarks. Products, categories and stores are compared on a like-for-like basis, to tell what is truly profitable from what only appears to be.

This turns portfolio analysis from an aggregate snapshot into an operational tool: you see where to concentrate space, items and investment and where to rationalize.

How do you decide assortment given finite shelf space?

By balancing margin and real constraints, not maximizing a single metric. Shelf space is a limited resource: adding one item means taking space from another, with effects on turnover, availability and category margin. A decision intelligence platform holds these trade-offs together:

• it links the assortment choice to the real saturation of shelf space and to stock capacity;
• it evaluates the impact of adding, replacing or delisting an item on the overall category margin, not just on its sales;
• it lets you compare alternatives with what-if simulations before changing the planogram, so SKU rationalization starts from economic consequences and not from revenue alone.

The principle is the one in the WhAI deck: avoid local optimization (one category, one channel) in favor of a choice that is coherent at banner level.

What is KPI root cause analysis applied to margin?

KPI root cause analysis is the ability to trace from causes to effects along the KPI tree: understanding why a category's margin moves, not just that it moved. If a department's profitability drops, the platform breaks the result down into its drivers (price, sourcing cost, mix, promotional pressure, returns) and isolates the real cause.

It is the heart of commercial control in a decision-oriented sense: KPIs are not isolated numbers on a dashboard but nodes of a model where every change is linked to its determinants. This is what turns management control from ex-post reporting into continuous decision support, with alerts on variances before commercial damage is done.

What should a retailer evaluate when choosing a decision intelligence platform?

Seven criteria, with specific attention to assortment and margin decisions:

• Real margin, not just sales. It must reconstruct margin at SKU, category and store level, including direct and indirect costs.
• ERP, WMS and sales-data integration. It must connect to systems and existing files without long IT projects, with manual upload as a fallback.
• Real space and stock constraints. It must model shelf saturation and stock capacity, not just aggregate averages.
• Cross-functional, cross-channel view. It must avoid department-level optimization and connect purchasing, logistics, sales and finance in a single model.
• Explainable drivers. It must show why a category creates or destroys value, not just how much.
• Governed, traceable decisions. Certified data, reliable calculations and verifiable, quantitatively justifiable output.
• Time-to-value and scalability. First insights as early as the data-connection phase, every new use case on the same infrastructure, with no lock-in.

How does Vedrai's WhAI support assortment and margin decisions?

WhAI is Vedrai's operating system for business decisions: it lets AI generate real, measurable value in retail. Its framework rests on four elements: Data (integrating internal data with relevant external variables and defining a shared semantic KPI layer), Technology (using AI for speed and accessibility and to orchestrate custom services and workflows), Analytics and Models (advanced economic models to handle every decision systematically) and Decisions and governance (reliable, governed and traceable decisions, quantitatively justifiable).

On assortment, WhAI reconstructs margin by customer, category and SKU, integrates direct and indirect costs into the models, identifies value and loss drivers and provides segmentation and performance benchmarks. In practice the path starts by mapping the portfolio (categories, channels, SKUs), builds the margin and cost-allocation models, identifies the key economic drivers and activates insights and operational decisions on assortment. It is modular and starts from where the company is today, and it is available in three modes: directly on the platform, through your own LLM, or as a managed service.

In keeping with this approach, the first insights emerge as early as the data-connection phase, use requires no data-science team, every new use case builds on the same infrastructure, and the models remain compatible with any AI solution, with no lock-in.

Key takeaways

• AI-based assortment optimization moves the decision from gross sales to real margin by category, channel and SKU.
• The underlying problem is poor visibility: over 60% of retailers don't know granular margin and more than 50% of allocations ignore cost drivers.
• A decision intelligence platform makes margin a calculated figure, explains value and loss drivers, and accounts for finite shelf space.
• KPI root cause analysis explains why a category's margin moves, not just that it moved.
• Choices should be made in an integrated way across channels and stores, avoiding single-department optimization.
• Vedrai's WhAI supports assortment and margin decisions in a governed, modular way, integrable with ERP and WMS.

Frequently Asked Questions (FAQ)

What is assortment optimization in retail? It is the choice of which products to keep, push or delist based on the real margin of each category and SKU, accounting for shelf space, stock and costs, not just sales.

Why aren't sales enough to decide assortment? Because an item can sell a lot and erode margin (through sourcing, logistics, returns or promotional pressure). You need real margin, with direct and indirect costs, to see where value is truly created.

How does a decision intelligence platform help with assortment? It links sales, costs and stock in a single model, reconstructs margin by category, channel and SKU, explains value and loss drivers, and evaluates space trade-offs before changing the planogram.

What is KPI root cause analysis applied to margin? It is cause-and-effect analysis that breaks a category's margin down into its drivers (price, cost, mix, promotions, returns) to isolate the real cause of a change, instead of merely flagging it.

Do you need a team of data scientists to use it in retail? No. WhAI translates technical complexity into tools accessible to decision-makers, with guided setup and expert support, and can start even with manual data upload.