Operations & Retail — AI for Fashion No. 07

The End of the Static Price Tag

For most of fashion retail’s history, pricing was a seasonal ritual. Buyers and merchandisers set prices months before garments hit the floor, factoring in cost-of-goods, competitive benchmarks, and a hoped-for margin. Markdowns came in two waves: a planned mid-season reduction, then a clearance blitz at the end. It was blunt, slow, and left enormous value on the table.

Airlines cracked this problem decades ago. Hotels followed. But fashion was different — too many SKUs, too much brand perception risk, too many physical constraints in-store. A blazer on a rail couldn’t be repriced like a seat on a flight.

E-commerce dissolved those constraints. And AI turned the residual complexity — thousands of variables, real-time signals, competitor monitoring — from a human impossibility into a machine’s specialty. The static price tag isn’t dead yet, but it is terminally ill.

What Dynamic Pricing Actually Means in Fashion

Dynamic pricing is not simply “changing prices often.” At its core, it is algorithmic price-setting that responds to real-time signals to maximise a defined objective — whether that is revenue, margin, sell-through rate, or some weighted combination of all three.

In fashion retail, the signals feeding these models are extraordinarily rich. Demand signals include page views, add-to-cart rates, wishlist additions, and purchase velocity. Supply signals include inventory depth by size and colour, days remaining in the season, and replenishment lead times. Competitive signals include scraped pricing from rival sites, updated every few minutes. Contextual signals include weather forecasts, influencer activity, and calendar events — a heatwave next week changes the value of a linen shirt today.

The AI’s job is to synthesise all of this into a single recommended price — or a price range with confidence bounds — and to do so continuously, at catalogue scale, faster than any human team could review a single SKU.

How the Models Work

At the heart of most fashion pricing engines is a demand estimation model — typically a gradient boosting algorithm or a neural network — trained on historical transaction data. It learns the relationship between price and sell-through: at £120, this jacket moved 40 units per week; at £135, it moved 28. That price-elasticity curve becomes the foundation for all downstream pricing decisions.

Layered on top is a reinforcement learning component in the most sophisticated implementations. Here, the model doesn’t just predict demand — it actively experiments with prices, observes outcomes, and updates its strategy. Over thousands of pricing cycles, it builds an increasingly refined understanding of where the demand curve actually sits for each product, in each market, at each point in the season.

Crucially, the model must also account for cannibalisation and complementarity. Discounting the trousers in a co-ordinate set affects the jacket’s demand. Raising the price of the premium denim might lift sales of the mid-range option. These portfolio effects are where simpler rules-based systems fail and where machine learning earns its keep.

Constraints are baked in from the start: minimum margins, maximum markdown depths, brand price floors, and promotional blackout periods. The AI operates within guardrails, but within them it moves freely — and fast.

Who Is Doing This — and How Well

Shein is the case study that keeps merchandising directors up at night. The ultra-fast fashion platform doesn’t just test new styles — it tests new prices. Every product launches at a micro-scale before being priced for volume, with algorithms adjusting in response to early demand signals that emerge within hours of listing. By the time a product reaches significant inventory commitment, Shein knows its optimal price with a precision that traditional buyers can only envy.

Zara’s approach is more restrained but equally sophisticated. Inditex has invested heavily in pricing intelligence that integrates store sell-through data from its RFID-tagged inventory with online browse and purchase signals. The result is a markdown cadence that is far more surgical than the industry average — deeper discounts on genuinely slow-moving pieces, minimal markdowns on anything showing demand strength.

ASOS operates one of the most aggressive dynamic pricing systems in European fashion e-commerce. Its algorithm is known to adjust prices on individual products multiple times per day based on competitor positioning and internal demand signals. Analysts tracking ASOS pricing have observed discounts appearing and disappearing within hours — a practice that would have been operationally impossible before machine learning.

At the luxury end, the calculus is different. Louis Vuitton, Hermès, and Chanel resist algorithmic markdowns on core lines with something approaching religious fervour. But even here, AI is reshaping pricing — through secondary market monitoring, regional price harmonisation, and the timing and depth of limited-edition pricing decisions.

The Markdown Problem — and How AI Solves It Differently

Fashion retail has a markdown addiction. An estimated 30–40% of all fashion inventory is ultimately sold at a discount, with markdown losses running into the billions annually across the industry. The traditional markdown playbook — wait until end of season, apply a standard 30% reduction, then 50% — is almost perfectly designed to destroy value.

AI attacks this problem from two directions simultaneously. First, it identifies slow-moving inventory earlier — sometimes within the first two weeks of a product’s life — and initiates modest, early price reductions that stimulate demand before stockrooms fill with units that will eventually require deeper discounting. A 10% reduction in week two is far less damaging than a 40% markdown in week twelve.

Second, it protects strong performers. Human buyers and merchandisers historically over-mark down products that are actually still selling well — either through caution, promotional momentum, or simply because they lack the time to monitor every SKU individually. An AI system with live sell-through data has no such blind spots. It holds price on items that don’t need discounting, which dramatically improves blended margin.

The net effect, reported by retailers deploying mature AI pricing systems, is a reduction in markdown depth of 30–40% — not by selling less at discount, but by discounting more precisely.

Personalised Pricing — The Next Frontier and Its Discontents

Catalogue-level dynamic pricing — the same price for all shoppers — is the industry standard today. But the technology is already moving toward something more granular: individualised pricing, where the price you see for a pair of jeans is different from the price your colleague sees, based on your respective browsing history, purchase patterns, and inferred price sensitivity.

Amazon has practised a version of this for years — surfacing different “personalised deals” to different customers, effectively offering price variation without explicitly showing different list prices. Several fashion retailers are exploring similar mechanics, using loyalty programme data and browsing behaviour to offer targeted promotional codes that function as individualised pricing.

The commercial logic is compelling. A customer who consistently pays full price doesn’t need a discount to convert. A customer who always waits for sale might convert at 15% off. Offering the former a discount you didn’t need to give is margin destruction. Offering the latter the right incentive at the right moment is revenue recovery.

But personalised pricing walks a razor’s edge. When customers discover they are paying different prices for identical items, the reputational fallout can be severe — as several airline and hotel operators have found. Fashion brands, which trade heavily on aspiration and trust, face heightened exposure. The regulatory environment is also tightening, with the EU’s Omnibus Directive and similar legislation in the UK and US placing new constraints on algorithmic pricing practices.

The Brand Risk Calculation

Dynamic pricing in fashion is not purely a technical problem. It is a brand problem. Consumer perception of price fairness matters enormously in categories where purchase decisions are partly aspirational. If a customer sees a coat at £280 on Monday and £220 on Wednesday, their trust in the retailer — and in the value of the product itself — is damaged, even if the price change was algorithmically justified.

This is why most sophisticated AI pricing systems in fashion operate with a concept of “price integrity windows” — minimum periods during which a product holds its current price, regardless of real-time signals. The algorithm might identify an optimal repricing opportunity in hour four of a product’s life, but the guardrail holds price for at least seven days to protect against the appearance of arbitrary or capricious pricing.

Transparency is emerging as both a regulatory requirement and a competitive differentiator. Brands that communicate their pricing logic — “limited stock,” “early-bird price,” “member exclusive” — convert the algorithm’s decision into a narrative the customer can understand and accept. Those that simply change prices silently invite suspicion.

The most advanced thinking in this space treats pricing not as an extraction problem — how much can we charge? — but as a relationship problem — what price sustains trust while optimising the business? The algorithm can answer the first question. The second requires human judgement operating alongside it.

The Physical Store Challenge

Everything described so far applies relatively cleanly to e-commerce. Physical retail is harder. A price tag on a rail cannot change in real time without significant operational cost. Printing and replacing labels across a large store takes hours and introduces error risk. Customer confusion when they see a different price at the till than on the hanger creates friction and reputational damage.

Electronic shelf labels (ESLs) are the infrastructure solution — digital price displays that can be updated remotely, synchronised with the e-commerce platform, and changed in seconds rather than hours. Decathlon, Primark, and several Zara formats have deployed ESL trials. But the capital expenditure — roughly £10–30 per label, multiplied by tens of thousands of SKUs per store — makes full deployment a significant investment.

A growing number of retailers are instead pursuing a hybrid approach: dynamic pricing on their app and website, with store prices updated on a weekly rather than continuous basis, informed by the same AI signals but operating on a more practical cadence. Customers who shop via the app can access real-time prices; store prices follow on a lag.

The longer-term trajectory is clear. As ESL costs fall and RFID-integrated inventory tracking becomes standard, the gap between digital and physical pricing intelligence will close. The store of 2030 will reprice as fluidly as its website does today.

What Executives Should Prioritise

For fashion executives evaluating or expanding AI pricing capabilities, the landscape of available solutions is broad — from purpose-built platforms like Revionics, Competera, and Prisync to custom-built internal systems at scale players. Choosing is less about technology than about organisational readiness.

The most common failure mode is not technical. It is cultural. Merchandising teams that have spent careers developing pricing intuition often resist algorithmic overrides — even when the data supports them. Change management, transparent model explanations, and a phased transition that demonstrates wins before removing human override authority are prerequisites for successful deployment.

Data infrastructure is the second prerequisite. AI pricing models are only as good as the data feeding them. Retailers without clean, unified SKU-level sell-through data — across channels, stores, and markets — will find their pricing algorithms making decisions on incomplete information. The investment in data quality and integration typically precedes the investment in pricing AI.

The payback, when the foundations are right, is significant. Industry benchmarks suggest AI-powered pricing deployments return 15–25% margin improvement within 18 months, with sell-through rates improving by 10–20%. In a sector where margin is perpetually under pressure, that is not incremental improvement — it is structural advantage.