Start a Conversation

The Decision Economy: Retail 101 for the AI-Enabled Enterprise

14 October, 2026 | 6 min read

Saumil Trivedi

By Saumil Trivedi

Retail Subject Matter Expert

The Decision Economy: Retail 101 for the AI-Enabled Enterprise

The retail operating model is being rewritten

Enterprise retail still runs on an operating model built for a simpler era: plan the season, buy the inventory, set the prices, and react when reality diverges. That model is now failing at industrial scale.
IHL Group’s 2026 study states that inventory distortion, the combined cost of out-of-stocks and overstocks, costs global retail about $1.7 trillion a year, roughly 6.2% of global retail sales, despite two decades of digital transformation spending.

What’s causing this is not a shortage of data or software. Most enterprise retailers have more of both than ever. The problem is that demand, inventory, pricing, and customer decisions are still made separately, by different teams, on different numbers, at different speeds. Each decision can be locally sound and still be wrong for the business.

This guide is a Retail 101 for that new reality: the AI reality. It explains how enterprise retail creates and loses value, why the decision chain breaks, the five shifts redefining the operating model, and what separates the retailers capturing value from AI from the majority that are not.

What retail is, and how enterprise retailers make money

Retail is the last link in a value chain running from manufacturers and suppliers, through distribution, to the consumer. Profit is the spread between landed cost and the price finally realized after discounts, returns, and fulfillment. Five measures carry that story: gross margin, contribution margin, inventory investment (working capital tied up in stock), sell-through, and customer value across the relationship.

The economics differ sharply by sub-industry:

Segment The decision that matters most
Fashion & Apparel Buy depth and markdown timing in short, trend-driven seasons
Grocery & Hypermarket Replenishment speed and freshness pricing at the highest SKU velocity
Specialty Retail Assortment and discount discipline on curated ranges
Consumer Electronics Competitive pricing and end-of-life clearance
Health, Beauty & Pharmacy Availability and personalization across many variants
Home & Furniture Inventory placement for bulky, long-lead-time goods
Luxury Full-price sell-through and allocation

The retail decision chain, and why it runs in silos

The retail decision chain, and why it runs in silos

Each link inherits the quality of the one before it. A forecast that never sees the promotion calendar under-orders; a promotion that never sees inventory sells out in the wrong stores. Yet each link usually has its own owner, system, and KPI. Disconnected POS, ERP, and warehouse data tell different versions of the truth: Target found that half of its out-of-stocks were invisible to its systems. Spreadsheets hold the real plan, KPIs conflict (full shelves versus lean inventory), and decisions run on last week’s data, adjusted by manual overrides.

How the economics have changed

Retail has always lived on the spread between what a product costs and what it finally sells for. What has changed is how many ways that spread now leaks.

Pressure What the data shows Why it matters
Inventory distortion $1.73 trillion lost each year to out-of-stocks and overstocks, 6.2% of global retail sales (IHL Group, 2025) Lost sales and forced markdowns stem from the same misplaced inventory
The AI divide Fewer than one in four retailers have rolled out AI or machine learning where distortion hits hardest; those that have report 2.3x higher sales growth and 2.5x higher profit growth (IHL Group, 2025) The gap between AI adopters and everyone else is widening
Returns US retailers expect 15.8% of 2025 sales, or $849.9 billion, to be returned, rising to 19.3% of online sales (NRF, 2025) Every return reverses a sale and re-enters inventory at a cost
Promotions 30 to 40% of retail promotions are inefficient or unprofitable, and 40 to 60% of those expected to deliver high ROI end up delivering low ROI (BCG, 2023) Discount spend often buys sales that would have happened anyway
Customer expectations 71% of consumers expect personalized interactions, and 76% are frustrated when they don’t get them (McKinsey, 2021) Generic experiences now cost conversion and loyalty

Look closely and these are not separate problems. Stockouts and overstocks are two outcomes of the same inaccurate plan. Promotions that lose money are often run on inventory that should never have been bought, and many returns begin with recommendations that never matched the customer.

The most telling detail is what separates the winners. IHL’s 2025 research found that the retailers outperforming on inventory share systematic data management and integrated technology platforms rather than point solutions. BCG reached the same conclusion on promotions: many retailers silo their pricing, promotion, markdown, and personalization teams with separate tools, so each decides on different facts. Those are operating model problems, not market ones.

Retail runs on a chain of decisions, and that chain needs is broken

Every enterprise retailer runs the same decision chain, whether or not it is ever drawn.

  • Demand informs the financial plan, the plan shapes the assortment, and the assortment sets inventory.
  • Inventory drives replenishment and fulfillment, while price, promotion, and markdown decide how that inventory converts into margin.
  • Experience and retention then determine whether the customer returns to start the cycle again.

In most organizations, each link has its own owner, its own system, and its own KPI. Merchandising is rewarded for availability, finance for lean inventory, marketing for promotional volume, and ecommerce for conversion. The failures are predictable. A forecast that never sees the promotion calendar under-orders. A promotion that never sees inventory sells out in the wrong stores. A markdown engine that never sees the promotion layers a second discount on the same item.

Target’s own data shows how deep the gap runs. When the retailer audited physical inventory, it found that half of its out-of-stocks were unknown to its systems, so replenishment never triggered for products that were already gone.

No downstream optimization can fix a chain that starts from the wrong number. Here’s where the shifts start to feel real.

Shift 1: Demand becomes probabilistic

The traditional forecast is a single number per item per week, built largely on history and adjusted by planners. It breaks exactly where retail is hardest: new products, promotions, weather swings, and trend-driven categories.

Leading retailers are replacing it with forecasts that are granular, signal-rich, and expressed as ranges.

In action: Walmart built a multi-horizon neural network in-house that draws on past demand, planned events, and current global and local trends to predict demand at several points in the future. The result is inventory planned further in advance, with less excess safety stock sitting in warehouses.

Two technical changes are bringing this capability beyond the largest players.

  1. Foundation models trained on broad demand patterns can now forecast zero-shot on a new catalog, which removes the multi-month training period that once delayed value.
  2. Probabilistic outputs (conservative, median, and upside) let buyers size risk instead of guessing it.

Growdhi’s Plan IQ is built on this same pattern, forecasting at SKU-location level across 1 to 90 day horizons and fine-tuning weekly on each retailer’s data.

The strategic point is easy to miss. Forecast accuracy is not the goal. A forecast creates value only when it changes a buy, an allocation, or a price.

Shift 2: Inventory truth replaces inventory visibility

For a decade, the industry’s goal was visibility: knowing where stock sits across stores, distribution centers, and channels. Visibility proved insufficient, because what systems report is often not what is on the shelf.

The new standard is inventory truth: a continuously reconciled position built from POS, ERP, and warehouse data, with AI inferring and correcting the errors that annual audits used to find months later.

In action: Target’s Inventory Ledger is the clearest public example. An ensemble of category-specific models infers hidden out-of-stocks, corrects inventory records, and triggers replenishment automatically. Target reports substantial sales lift on products that would otherwise have been unavailable, and it now runs the approach across almost every category.

Once inventory is accurate, replenishment and logistics can become dynamic. Reorders can be calculated from demand and supplier lead time together, within real supplier terms such as MOQs and order multiples. Routing can then choose which location fulfills an order at the lowest total cost across store, warehouse, and marketplace channels. 

This is the operating logic behind Growdhi’s Supply IQ, which connects replenishment, allocation, and routing to the same forecast.

The headroom is large. IHL found that fewer than 1 in 4 retailers have rolled out AI or machine learning in the areas most affected by inventory distortion.

Shift 3: Price becomes a precision instrument

Price is the fastest lever in retail and the one most often pulled by reflex. 

Competitor matching, calendar promotions, and end-of-season clearance are still largely rules-based, and the results show it. 

BCG estimates that 30 to 40% of retail promotions are inefficient or unprofitable, and that 40 to 60% of the promotions retailers expect to deliver high ROI end up delivering low ROI.

The shift is from discounting as a habit to pricing as a measured decision, and it rests on three capabilities:

  • Elasticity modeling estimates how demand for each SKU responds to price, competitor moves, and inventory position.
  • Incrementality measurement judges a promotion by the sales it added, not the sales that happened during it, and tests for cannibalization of full-price items before launch.
  • Markdown optimization replaces one deep end-of-season cut with a multi-step ladder timed to the sell-through curve.

The harder problem is coordination. In many retailers, pricing, promotion, and clearance run in separate tools, so one SKU can be discounted from two directions at once. Growdhi’s Price IQ is designed around that failure: when an item enters clearance, promotions hold back competing offers and the shelf price holds steady, so every SKU carries one coordinated pricing decision.

The payoff is not lower prices – It is fewer unnecessary ones.

Shift 4: Customer value replaces traffic

For most of the ecommerce era, growth meant traffic: buy more visits and convert more of them. That model strains as acquisition gets more expensive and customers expect more. 

McKinsey found that personalization most often drives a 10 to 15% revenue lift, with company-specific results ranging from 5 to 25%, and that 78% of consumers say personalized content makes them more likely to repurchase.

The shift has two halves:

  1. The first is the session. Search that fails on a typo, recommendations that ignore intent, and chat that treats a first-time visitor like a loyal customer all leak the conversion that merchandising and supply teams worked to earn. AI can now read live signals within a session and adapt search, recommendations, and assistance in real time, which is the role of Growdhi’s Experience IQ.
  2. The second half is the relationship. Market basket analysis reveals what customers buy together, lifetime value modeling shows who is worth investing in, and churn models flag a valuable customer drifting away early enough to win them back.

Growdhi’s Growth IQ brings these together, so retention becomes a managed decision rather than a quarterly report.

The link between the two halves is where most retailers fall short. A churn signal is worth something only if the storefront, the offer, and the price respond to it.

Shift 5: Decisions connect across the chain

The first four shifts improve individual decisions. The fifth changes the operating model itself, from functions that optimize their own metrics to a system that optimizes the business.

Retail analytics has moved from reporting what happened, to explaining why, to predicting what comes next, to optimizing single functions. Decision intelligence is the next stage: combining prediction, optimization, and business rules to decide what the business should do, then acting on it.

Gartner has predicted that by 2026, 75% of Global 500 companies will apply decision intelligence practices, including logging decisions for later analysis.

In retail, connection means a change in one domain reaches the others in time to matter. 

A demand spike should reach replenishment before stock runs out. A clearance decision should reach the promotion engine before a competing offer goes live. A churn-risk flag should reach pricing, recommendations, and the shopping assistant within the same session.

Growdhi was built on this premise, coordinating 15 agents across planning, supply, pricing, experience, and growth on one shared foundation, governed by each retailer’s own rules.

Connected decisions are also safer decisions. When every agent works from the same truth, conflicts surface where people can see them instead of quietly cancelling each other out.

What retail leaders are doing wrong

BCG’s 2024 survey of 1,000 executives found that 74% of companies had yet to show tangible value from AI, while leaders drew 62% of their AI value from core business processes. In retail, four mistakes explain most of the gap:

  • Buying a better tool for every silo, which improves each function and leaves the gaps between them untouched.
  • Wiring AI to dashboards instead of decisions, so insight never reaches a system that acts.
  • Switching automation on all at once, so teams that cannot see why a model decided something work around it.
  • Measuring activity instead of impact, with no do-nothing baseline to prove what AI added.

The retail KPI framework

Most KPIs are owned by one function but moved by several, so each should be tied to every decision that drives it.

Area KPIs Decisions that move them
Growth Revenue, conversion, AOV Search, recommendations, price, availability
Margin Gross margin, markdown depth Pricing, promotion incrementality, markdown timing, buy depth
Inventory Stockout rate, sell-through, inventory accuracy Forecasting, replenishment, allocation
Customer Repeat rate, CLTV, churn Personalization, bundles, winback timing
Operations Forecast accuracy, cost per shipment, on-time delivery Forecasting, routing, fulfillment

Why most retail AI stalls, and what leaders do differently

Adoption is no longer the problem – Value is. 

BCG’s 2024 survey of 1,000 senior executives found that 74% of companies had yet to show tangible value from AI, and only 4% had built cutting-edge capabilities that consistently generate significant value. 

The leaders generate 62% of their AI value in core business processes, and they concentrate investment on a few high-priority opportunities rather than spreading it thin. Retail follows the same pattern. 

Four habits separate the retailers that scale from those stuck in pilots:

  1. They connect AI to the systems that act: Walmart and Target both wired their models into replenishment and inventory accounting, not just into dashboards.
  2. They earn autonomy in stages: Target began with human checks on every AI inference and moved to automated corrections only after independent measurement proved the models right.
  3. They measure against a baseline: Every automated decision is compared with what would have happened without it, so value is proven rather than assumed.
  4. They govern with guardrails: Business rules, approval thresholds, and escalation paths define what the system decides alone and what reaches a person.

None of these are technical choices. They are operating model choices, which is exactly why they are so often skipped.

The KPI shift: measuring decisions, not departments

Most retail KPIs are owned by one function but moved by several. Stockout rate belongs to supply chain, yet forecasting, promotions, and pricing all change it. A connected operating model measures each KPI against every decision that drives it.

KPI Traditional owner Decisions that actually move it
Stockout rate Supply chain Forecast, replenishment, allocation, and promotion timing
Gross margin Finance Price, promotion incrementality, markdown timing, and buy depth
Sell-through Merchandising Assortment, allocation, pricing, and discoverability
Conversion and AOV Ecommerce Search, recommendations, availability, and price
Repeat rate and CLTV Marketing Experience, personalized offers, availability, and winback timing
Working capital Finance Forecast accuracy, buy depth, and replenishment policy

When the decisions behind a KPI share one system, the KPI stops being a scorecard argued over in meetings and becomes something the business can actually steer.

The next 5 years of the retail operating model

The direction of travel is already visible in the retailers furthest ahead. Four transitions will define the operating model through the end of the decade:

From To
Seasonal planning cycles Continuous planning that updates as signals change
Dashboards that report Decision systems that recommend and act
Manual analysis and overrides AI-assisted decisions with human oversight on exceptions
Optimization by function Connected decisions across the whole chain

None of this removes people from retail. It moves them to where judgment matters most: setting strategy, defining guardrails, and resolving the exceptions a system escalates. The routine decisions, how much to reorder, what to charge today, and which offer to show, will increasingly run themselves, audited and measured against a baseline.

The retailers that lead the next decade will not be the ones with the most AI tools. They will be the ones whose decisions agree with each other.

The future of retail: continuously optimized decisions

The next operating model moves from periodic planning to continuous planning, from dashboards to decision systems, from manual analysis to AI-assisted decisions, and from isolated optimization to connected decisions. People move to where judgment matters: setting strategy, defining guardrails, and resolving the exceptions a system escalates. The retailers that lead the next decade will not be the ones with the most AI tools. They will be the ones whose decisions agree with each other.

Retail 101: frequently asked questions

A retail operating model is how a retailer makes and executes its core decisions, from demand planning and inventory to pricing, customer experience, and retention. An AI-enabled operating model connects those decisions so they share one set of signals.

Inventory distortion is the combined cost of out-of-stocks and overstocks. IHL Group estimates it costs global retail about $1.7 trillion a year, roughly 6.5% of global retail sales.

Decision intelligence combines prediction, optimization, and business rules to decide what a retailer should do next, then acts on it with human oversight. Analytics explains what happened; decision intelligence determines the next action.

Most promotions are judged on total sales during the offer rather than incremental lift. Without measuring incrementality and cannibalization, retailers pay discounts on sales they would have made anyway.

Start with one high-value decision connected to the system that executes it, run the AI alongside current practice, measure it against a baseline, and expand its autonomy as accuracy is proven.