Retail demand now moves at the speed of a search query. A product can trend on social media on Monday, show up in AI shopping assistants by Wednesday, and sell out in a handful of stores by Friday.

McKinsey and ICSC make this point in their 2026 report, Shopping in the age of AI: Redefining stores for a new era. The authors say in-store merchandising cycles need to speed up, because planning assortments once or twice a year no longer works when search trends, social signals, and AI-assisted discovery can shift demand in a matter of weeks.

AI demand forecasting is how retailers keep pace. It reads digital search trends, local sell-through, and first-party customer data, then turns them into daily decisions on staffing, replenishment, local allocation, and assortment.

BSPK, the customer intelligence platform powered by AI clienteling and first-party data, adds the signal most forecasting tools miss: what individual customers tell your teams in stores, in messages, and in appointments. This guide explains how AI-powered demand forecasting works, where it pays off first, and how to put it into production.

Five takeaways retail leaders can act on this quarter

• Merchandising cycles have to move faster. McKinsey's 2026 retail research calls for real-time demand signals, such as search trends and local sell-through, to update replenishment and test new products continuously instead of waiting for the next seasonal reset.

• Demand sensing closes the gap between search and shelf. Short-term signals refresh demand forecasts every day, so a store can react to a trend within days instead of weeks.

• AI forecasting improves staffing as much as stock. Predictive demand models tell managers when traffic will peak, so the right associates are on the floor when shoppers arrive.

• Inventory distortion is still a $1.7 trillion problem. IHL Group's 2026 Inventory Distortion Study puts the global cost of out-of-stocks and overstocks at 6.2% of retail sales, and retailers using AI in these functions grow faster.

• First-party customer data makes every forecast sharper. Wishlists, waitlists, appointments, and associate notes show intent before a transaction happens, and BSPK turns that intent into data your planning tools can use.

AI demand forecasting turns live signals into retail decisions

Demand forecasting is the practice of predicting how much of each product customers will buy, where they will buy it, and when. Retailers use demand forecasts to set purchase orders, plan labor, allocate stock to stores, and price promotions.

AI demand forecasting applies machine learning to the same question with far more inputs. Instead of projecting last year's historical sales forward, AI-based demand models weigh dozens of signals at once and retrain as new data arrives.

Demand forecasting is different from demand planning. The forecast estimates what customers will buy, and demand planning turns that estimate into purchase orders, allocation plans, and labor schedules across the supply chain process.

Predictive analytics sits underneath both. Classic predictive analytics fits a statistical model to past patterns, while AI forecasting adds machine learning that finds relationships planners would never code by hand, such as how a local weather change shifts demand for one category but not another.

For multi-location businesses, the scale is the reason to automate. A retailer with 200 stores and 5,000 active SKUs has a million store and SKU combinations to forecast, and no planning team can review each one by hand every week.

The difference shows up in decision speed. A planner who once rebuilt a forecast every month can now review a forecast that updates every night and flags only the SKUs and stores that need attention.

Traditional demand forecasting methods lag behind shopper behavior

Traditional demand forecasting methods lean on past sales, a seasonal curve, and a planner's manual adjustments in a spreadsheet. That approach works for stable products in stable periods.

It breaks when demand shifts faster than the planning calendar. A viral product, a heat wave, or a new store opening nearby can change demand in a single region within days, and a monthly forecast will not see it until the sales are already lost.

McKinsey's operations research found that applying AI-driven forecasting to supply chain management can cut forecast errors by 20 to 50 percent and reduce lost sales and product unavailability by up to 65 percent. Those gains come from reading more signals, more often, at a finer level of detail.

AI-driven forecasting reads more signals at a finer grain

AI-driven forecasting works at the SKU, store, and day level. It can combine point-of-sale data, ecommerce traffic, search trends, weather, local events, promotional calendars, and customer data in one model.

The output is a forecast that changes as conditions change. Planners stop rebuilding spreadsheets and start reviewing exceptions, such as a store that is selling a SKU at twice the forecast rate.

Common inputs for retail demand models include:

• Historical sales and returns by SKU, store, and channel

• Ecommerce sessions, product page views, and on-site search terms

• External search and social trends for categories and specific products

• Promotional calendars, price changes, and markdown history

• Weather, local events, and store traffic counts

• First-party customer data such as wishlists, waitlists, appointments, and clienteling notes

The last input is the one most retailers underuse. It is also the one no outside data provider can sell you, because it comes from your own customer relationships.

Merchandising cycles have to speed up in the AI shopping era

McKinsey and ICSC surveyed 3,004 US consumers in January 2026 for the Shopping in the age of AI report. Their guidance for retailers is direct: AI is moving research and comparison upstream, so stores need faster ways to adjust what they feature, how they allocate local inventory, and which promotions they run.

The consumer data supports that view. According to the report, 68 percent of consumers used at least one AI-enabled tool in the past three months, and 62 percent used AI to compare brands, models, prices, or reviews.

Shoppers also check stock before they travel. The same research found that 85 percent of shoppers do some online research before visiting a store, and 37 percent ranked reliably finding what they came for among their top three reasons for choosing a retailer, the most cited factor in the survey.

For merchandising teams, those numbers change the job. A forecast that is right on average across the chain but wrong in a given store on a given week now costs a visit, because the shopper already knows the item is not there.

AI shopping agents check your inventory before the shopper arrives

Agentic commerce raises the stakes on forecast accuracy. McKinsey estimates the US B2C retail market could see up to $1 trillion in revenue from agentic commerce by 2030.

When an AI agent builds a basket, it favors retailers that publish accurate local availability and reliable pickup windows. McKinsey's report warns that a store unable to provide structured product data and dependable fulfillment windows will be bypassed in agent-led journeys.

That turns inventory accuracy into a marketing issue. If AI demand forecasting keeps local stock aligned with demand, and inventory management systems publish that stock correctly, AI agents can recommend your stores with confidence.

Few retailers are ready. In a McKinsey survey of 50 executives, 45 said they had considered introducing an agentic commerce tool, but fewer than five had a board-aligned strategy for it.

Store missions change what the demand forecast has to answer

McKinsey sorts store trips into two missions: convenience and discovery. Convenience stores need guaranteed availability on high-demand items, so the forecast has to protect in-stock rates on a focused assortment.

Discovery stores need curated, rotating collections and associates who know their clients. There, the forecast has to predict interest in new and limited products, which past sales alone cannot predict.

One forecasting system can serve both missions if it runs at store level. The service targets, safety stock rules, and input weights should differ by store mission, and the planning team should be able to see which rules apply where.

Demand sensing connects search trends to sell-through

Demand sensing is the short-horizon layer of AI demand forecasting. It uses the most recent signals, often from the last few days, to correct the baseline forecast for the next one to four weeks.

McKinsey names two signals merchants should use: digital search trends and local sell-through data. Search shows where demand is heading, and sell-through shows where demand is landing.

Demand sensing started in consumer goods, where retail CPG teams use it to plan shipments to retailer warehouses. In the retail sector, the same method works closer to the shopper, at the level of a single store and a single SKU.

Neither signal is enough alone. Search interest without sales can mean shoppers are comparing and leaving, while strong sell-through without search interest can point to a local driver, such as an event or a store team that knows how to sell the item.

AI-driven demand sensing catches shifts in days

AI-driven demand sensing compares current sell-through with the forecast every day. When a SKU sells well above plan in a cluster of stores, the system flags it and recommends a response, such as a transfer, a reorder, or a change in display.

Search data adds a leading indicator. A rise in searches for a product or style, on your site and in the wider market, often comes before the sales spike and gives planners time to act before shelves empty.

The value comes from response speed. A signal that reaches a buyer in two days drives a decision, while the same signal in a monthly report only explains what went wrong.

First-party customer data adds signals search cannot see

Search and sales data describe mostly anonymous behavior. They miss the demand customers express directly to your teams: the client who asks to hear when a size returns, the regular who books an appointment to see a new collection, or the shopper who saves items to a wishlist.

BSPK calls this the problem of thin signals for AI. Many AI models train on limited or generic fields instead of rich, contextual first-party interaction data, so their forecasts miss intent that never reached a transaction.

Capturing those interactions as structured data gives forecasting models a forward view of demand. A waitlist of named clients for a returning style is a stronger signal than a search trend, because each name is a person who already asked to buy. Our guide to what clienteling is and how it works explains how store teams collect this data during everyday selling.

AI-powered demand forecasting sets store staffing to match traffic

McKinsey's report notes that AI-driven demand forecasting can help stores align staffing with peak traffic. In convenience stores, that means enough associates on the floor for fast product location and checkout during rushes.

Labor is one of the largest costs a store manager controls, so the same forecast that drives stock should drive schedules. A store that predicts a Saturday spike for a new release can add floor coverage before lines form instead of after.

Predictive demand models schedule associates for peak hours

To forecast demand for labor, the model needs more than sales. Traffic counts, appointment bookings, pickup orders, and returns all create work for associates, and each follows its own daily pattern.

Predictive demand models for labor combine forecasted traffic, conversion rates, and transaction counts by hour. The output tells a manager how many associates each shift needs and which skills those associates should bring.

Shoppers want help on their own terms. In the McKinsey and ICSC survey, 78 percent of respondents said they prefer help from an associate only when they ask for it, yet they placed high value on the quality of that help when they did ask.

Workforce optimization based on demand forecasts also cuts idle hours. Shifts get shorter on slow mornings and longer before evening peaks, and managers spend less time rebuilding schedules by hand.

That makes timing the goal. A forecast that puts a knowledgeable associate within reach at the busiest hour serves shoppers better than a crowded floor at a quiet hour.

Discovery stores need the right specialists at the right moments

Discovery-led stores depend on associates with product expertise, styling skills, and client relationships. Forecasts that predict appointment demand and event attendance help managers schedule those specialists when high-value clients are likely to visit.

Tools for workforce management and clienteling connect here. When appointments and client activity sit in the same system as the forecast, schedules reflect booked demand as well as expected walk-in traffic.

Training matters as much as scheduling. McKinsey cites its State of Fashion research showing 47 percent of US consumer goods and retail employees say training is the most important factor for adopting gen AI at work, so new tools need simple workflows that associates can learn quickly.

Replenishment runs on daily predictive analytics

Replenishment is where forecast errors become visible to shoppers. A forecast that runs low leaves an empty shelf, and a forecast that runs high ties up cash in stock that later needs a markdown.

The cost is large. IHL Group's 2026 research puts global inventory distortion at $1.7 trillion, or 6.2 percent of retail sales, with out-of-stocks making up 65.6 percent of that total.

Inventory management improves when forecasts update daily

Daily forecasts let inventory management move from fixed reorder points to dynamic ones. Safety stock rises on SKUs with volatile demand and falls on steady sellers, which frees working capital without hurting availability.

McKinsey's report says retailers that invest in real-time inventory visibility and tighter replenishment improve on-shelf availability and reduce search friction. Both factors tie closely to conversion and customer satisfaction on convenience trips.

IHL's study shows the payoff of AI and predictive analytics in these functions. Retailers deploying AI and machine learning in distortion-related areas posted 2.3 times the sales growth and 2.5 times the profit growth of retailers that did not.

SKU-level forecasts by store prevent stockouts and markdowns

Inventory optimization uses the forecast to set order quantities, safety stock, and store targets together instead of one at a time. Markdown optimization then uses the same demand forecasts to time price cuts on items that are selling below plan, which protects margin better than a single end-of-season sale.

SKU-level forecasts by store reveal patterns that a category plan hides. One size may sell out in urban stores while the same item sits in suburban stockrooms.

When the forecast flags that imbalance early, the replenishment system can redirect inbound stock instead of ordering more. That cuts stockouts and markdowns at the same time, which is the rare inventory decision that helps both sales and margin.

Returns belong in the same model. A store with high return rates on a category needs a different replenishment rule than a store where the same items stay sold.

Local allocation puts the right stock in the right store

McKinsey's guidance asks retailers to change how they allocate local inventory as demand shifts. Allocation decides which stores get which units, and in fashion, specialty retail, and beauty it often decides how the season ends.

Chain-level forecasts average away local differences. AI-based demand forecasting at store level reflects the customers each location serves, from tourist-heavy flagships to suburban stores on daily commuter routes.

Demand forecasting in retail works best at the store and customer level

Store-level demand forecasting in retail starts with local sell-through and adds context about who shops there. McKinsey recommends a store file that includes customer behavior and local context, such as how far customers travel, how often they return, and how visits connect to online research.

McKinsey also describes geospatial analytics that leading retailers embed in decision-making, continuously reassessing store roles, fulfillment flows, and space allocation as demand patterns shift. Those analytics combine transaction data with mobility, traffic, and population data to show how each store creates value.

Customer data makes that store file more accurate. If a store's top clients favor a certain category or size range, the allocation model can weight inbound units toward that preference before the first sale.

Live inventory visibility handles the misses. When associates can check stock in nearby stores and reserve a unit for a client, a forecast error becomes a saved sale, and each request becomes a demand signal for the next allocation.

Continuous assortment decisions replace seasonal resets

McKinsey describes a move from seasonal resets to continuous testing. Retailers should use real-time demand signals to refine endcaps and continuously test new products throughout the season.

Continuous assortment planning needs a forecast that can predict demand for products with little or no history. AI models handle this by learning from similar items, from early sell-through in test stores, and from customer interest collected before launch.

Weekly sell-through data guides test-and-learn assortment

A test-and-learn approach places a new product in a small set of stores, reads sell-through for one to two weeks, and then scales or cuts it. The forecast updates as results arrive, so buyers commit to depth only after demand is proven.

This lowers the risk of large buys on unproven items. It also gives discovery stores a steady flow of new products, which McKinsey identifies as a reason for shoppers to make the trip.

AI demand forecasting helps size the first buy, too. By matching a new item to similar products by price, category, color, and store cluster, the model gives buyers a starting forecast that improves with every week of real sales.

Pre-launch interest makes the test sharper. When associates share a new piece with clients who like that category and record who responds, buyers see demand by store before the product even lands.

Promotional planning follows measured demand

Promotions distort demand, and older planning tools struggle to separate promotional lift from baseline demand. AI forecasting models use promotional analytics to estimate the lift of each promotional mechanic by store and customer group, so retailers can pick offers that drive profit instead of volume alone.

Personal outreach is often a better lever than a broad discount. A message to clients who have shown interest in a category can move slow inventory without training every shopper to wait for a sale.

Fragmented data is the biggest barrier to AI-based demand forecasting

AI models are only as good as the data they read. BSPK's positioning names fragmented data as the first structural problem retailers face, with customer information scattered across ecommerce, CRM, POS, marketing platforms, and store-level tools.

The good news is that most businesses already hold enough data to begin. McKinsey's forecasting research found that most organizations have enough data to get value from AI-driven forecasting, even when their historical records are limited, as long as they choose the right model and add useful external data.

Frontline turnover adds a second gap. When associates leave, the customer knowledge in their heads and on their personal phones leaves with them, and the forecast loses signals it never captured.

A retail CRM built for store teams fixes both problems at the source. It records interactions in a shared system, connects them to sales and inventory through integration with existing platforms, and keeps that knowledge with the brand.

A 90-day plan gets AI forecasting into production

Most retailers do not need to replace their planning tools to start. The fastest path builds on existing data and proves value on a small scope before scaling.

1. Weeks 1 to 3: Pick one category and a group of stores, and connect sales, inventory, and ecommerce data for them.

2. Weeks 2 to 5: Add first-party customer signals, such as wishlists, waitlists, and appointments, through integration with your clienteling or CRM platform.

3. Weeks 4 to 8: Run the AI forecast beside your current forecast and compare accuracy by SKU and store.

4. Weeks 6 to 10: Use the new forecast for one decision first, such as replenishment or staffing, and track in-stock rates and labor hours.

5. Weeks 10 to 13: Review results with merchandising, store operations, and finance, then extend to more categories and decisions.

When you evaluate AI-powered demand forecasting software, ask how it handles integration with your POS, ecommerce, ERP, and customer data. Ask, too, whether planners can see why the model changed a forecast, because demand planning teams will not act on numbers they cannot explain.

Bring your demand planning lead, a store operations leader, and a merchant into the pilot from week one. Forecast applications fail more often from low adoption than from weak models, and shared ownership fixes that early.

Track a short list of metrics for the pilot stores against a control group:

• Forecast error and bias by SKU and store

• In-stock rate on top-selling items

• Sell-through and markdown rate by category

• Labor hours per transaction at peak times

• Revenue from client outreach tied to waitlists and restocks

Those numbers make the business case for scaling, and they show merchandising and store teams that the forecast is earning trust.

How BSPK, the customer intelligence platform powered by AI clienteling and first-party data, can help

BSPK turns first-party customer data into repeatable customer growth. It is the system that turns in-store and remote customer interactions into AI-ready data, so a brand can treat every customer as an individual, at scale.

For demand forecasting, that matters because a customer conversation is the earliest demand signal a retailer owns. BSPK follows a capture, unify, and activate model that feeds those signals into AI, merchandising, and store operations.

BSPK captures demand signals at every customer interaction

Associates log client preferences, wishlists, waitlists, appointments, and notes in the BSPK mobile app, which is designed to feel as familiar as a consumer messaging app. Each interaction becomes structured first-party data attached to a persistent customer identity.

Because that data lives in BSPK's cloud, it stays with the brand when associates change roles or leave. Every interaction your associates have becomes high-value data your brand can reuse in AI and marketing, even as teams change.

Webchat and abandoned-checkout task lists add online intent to the same record. A shopper who browses a style online and then asks an associate about it in store becomes one connected signal instead of two disconnected ones.

BSPK unifies sales, inventory, and customer data into one profile

BSPK's single customer view brings customer, transaction, product, inventory, and engagement data into one living profile. Real-time, two-way integrations with POS, ecommerce, CRM, and ERP systems mean a purchase or cart update shows up right away instead of in an overnight batch.

Associates see live inventory across stores and regions, can search the collection by name or SKU, and can message a stock runner to pull an item for a specific client. Those actions keep sales moving when allocation misses, and they record where demand appeared.

BSPK layers on top of the systems you already run. Through open APIs and integration with your existing stack, its first-party customer data can flow into the demand planning and inventory management tools your teams use today, with no rip and replace.

BSPK agents act on demand signals one customer at a time

BSPK's customer intelligence lets teams describe the clients they want in plain language and get a segment back, such as clients who bought a category last spring or clients waiting on a restock. Any segment becomes a task or campaign in one step.

BSPK agents then act on those signals with associates in control. The Engage Agent flags reorder cycles and restocks, the Convert Agent curates products from live inventory for in-store appointments, and the associate approves every message before it sends.

This is how a retailer moves slow stock or fills a waitlist without a blanket discount. Personal outreach to the clients most likely to buy turns forecast insight into sell-through.

BSPK analytics show HQ what drives sell-through

BSPK's analytics and reporting tie outreach to revenue by store, region, and associate. HQ teams see which client actions drove sales, which stores convert waitlists, and where demand is building before it shows up in POS data.

Predictive signals from BSPK, such as growing waitlists or rising interest in a collection, give planners an early read on demand by store. That read arrives weeks before the POS data would show the same trend.

That closes the loop between forecasting and action. Merchandising, planning, and store teams work from the same customer data, and each cycle sharpens the next forecast.

Frequently asked questions about AI demand forecasting

How is AI used in demand forecasting?

AI is used in demand forecasting to predict sales at the SKU, store, and day level from many signals at once. Models combine historical sales, promotions, weather, search trends, and customer data, then retrain as new data arrives.

Which AI is best for demand forecasting?

The best AI for demand forecasting is the one that fits your data, your categories, and the decisions you need to make. A good fit connects to your POS, ecommerce, and customer data, works at store level, and explains its forecasts so planners trust them.

How is AI being used in retail stores?

Retail stores use AI applications to forecast traffic for staffing, trigger replenishment, allocate stock, and give associates customer context on the floor. McKinsey's 2026 research describes AI-supported clienteling tools that show associates purchase history, preferences, and next-best actions.

Can ChatGPT do forecasting?

General-purpose chat assistants can explain forecasting methods and help analyze a small dataset you paste in. Production demand forecasting in retail needs models connected to live sales, inventory, and customer data that retrain daily, which a chat assistant alone does not provide.

What data does AI demand forecasting need?

AI demand forecasting needs clean historical sales, inventory positions, prices, and promotional history at the SKU and store level. Forecasts get sharper when you add search trends, local events, and first-party customer data such as wishlists, waitlists, and appointments.

Retailers that act on demand signals every week will win the next season

McKinsey's message is that merchandising can no longer run on a twice-a-year calendar for retail businesses of any size. AI demand forecasting gives retailers the speed to adjust staffing, replenishment, local allocation, and assortment as demand shifts.

The retailers that pull ahead will add first-party customer data to that forecast, because what clients tell your teams today predicts what they will buy next. In the age of AI, your real advantage is the customer data you own and how intelligently you use it.

Request a demo of BSPK to see how your own customer data can sharpen demand forecasting and sell-through. BSPK offers a two-week go-live with no rip and replace, and you see your own data in the demo.