Sears is gone, Toys R Us survives as a brand name inside Macy’s, and Shein can turn a trending sketch into a product page in days. Retail innovation covers everything between a customer’s phone and the warehouse floor: apps, data, automation, AI, checkout. Stores that adapted kept their shoppers. The others kept their old systems. Here is what works, what is still being tested, and why the back end decides who wins.
Flashy on the front, fragile behind it
Retail conferences are full of good demos. Cashierless stores, virtual fitting rooms, a chatbot that knows your shoe size. Meanwhile the same company’s checkout dies on Black Friday and returns are processed in a spreadsheet somebody emailed around in March.
The reason is usually dull. The shiny layer sits on an ERP from 2008, a product database where “blue” is spelled four different ways, and a warehouse tool that one person really understands. Retailers who look into legacy modernisation services before the next big feature launch tend to avoid the awkward board meeting where a pilot works on a laptop and falls apart in production.
What the stack has to handle
- One customer identity across web, app, store and email
- Stock that updates live, not overnight
- Product data clean enough for search, filters and AI assistants
- Checkout with a handful of payment methods
- Returns that don’t take three emails and a phone call
- APIs for marketplaces, partners and, more and more, bots
Ever used click and collect where the app promised the item and the shelf disagreed? That’s an integration problem dressed up as a customer experience problem.
Who builds all this
Few retailers employ hundreds of engineers, so most lean on outside teams for specific pieces. Bintime, a Dutch-founded engineering company with teams in Europe, publishes cases that show the usual pattern. Eden Technology got a custom ERP, PIM and online shop over roughly five years. Futurumshop, a Dutch cycling retailer, has a personalized webshop, mobile apps and a loyalty program that were modernised piece by piece while the shop kept selling bikes daily. Retail and omnichannel sit first on the company’s list of industries, followed by brands, B2B and rental.
Not glamorous work. But updates that ship while the till keeps ringing are what most retailers actually need.
What’s new on the shelf
Retail technology comes in waves, and some of them recede fast.
Checkout without checkout
Amazon Go opened to the public in Seattle in 2018, with cameras and sensors tracking what people picked up. Just Walk Out later reached airports, stadiums and other companies’ stores. Then Amazon pulled it from its own Fresh grocery stores in 2024, and press coverage pointed to a lot of human review behind the curtain. Great demo, shaky economics.
Smaller ideas held up better: scan-and-go apps, weight-sensing self-checkout, and electronic shelf labels at chains like Walmart and Kroger, which quietly fixed price mismatches.
Assistants and visual shopping
Amazon’s Rufus answers questions such as “what do I need for a first camping trip?” inside the app, and Walmart has Sparky. Google’s try-on in Search puts clothes on different body types. IKEA’s Kreativ scans a room and drops furniture into it. Sephora has let people test lipstick through a camera feed since the late 2010s. Fashion is the obvious target, because wrong-size returns are one of its most expensive habits.
Agents as shoppers
Strange but real. In 2025 OpenAI launched Instant Checkout in ChatGPT with Etsy and Shopify merchants, built on the Agentic Commerce Protocol it developed with Stripe. Google published its own Agent Payments Protocol, AP2, and Perplexity tested in-chat purchases through PayPal. When a bot compares prices and pays, the homepage loses some importance. The product feed, live stock and honest delivery dates take over as the storefront.
Supply chain as a weapon
Zara tags garments with RFID to see what is really on the floor. Shein produces in small batches and reorders whatever sells. Alibaba’s Hema supermarkets ship online orders straight from the store. Drone drops have been tested by Wing and Walmart in Texas and Arkansas, and Starship robots deliver coffee around campuses.
Still in the lab
- Product pages rewritten for each visitor
- Voice agents that take phone orders
- Personal “taste twins” trained on one shopper’s history
- Smart mirrors and in-store assistants, piloted by Zara and Uniqlo
- Digital product passports, pushed by the EU’s ecodesign rules
A few will stick. The rest will join the Messenger chatbot graveyard of 2016.
Stores changed jobs
Someone declares the physical shop dead every couple of years. Yet Walmart fills many online grocery orders from ordinary stores, and IKEA keeps opening small city-center formats beside its blue warehouses. Patagonia’s Worn Wear and Zalando’s pre-owned section turned returns into a second revenue line.
One store can be a showroom, a pickup point, a mini-warehouse and a returns desk at once. That works only when every system agrees on what is inside the building. Often they don’t.
Where innovation trips
- Pilots that behave with 500 products and collapse at 500,000
- Chatbots that invent return policies
- Personalization that feels like being watched. Target’s 2012 pregnancy-coupon story still gets retold for a reason
- Data silos between e-commerce, stores and finance
- Integration work that never made it into the budget
The boring culprit
Ask engineers inside retail what blocks them and the AI model barely comes up. They name the old order system, the product database, the missing API, the pricing rules living in a script nobody documented. Familiar?
A full rewrite is the tempting fix and the dangerous one. TSB’s 2018 migration in the UK left customers locked out of their accounts for weeks, and retail has its own version: replatformed webshops that lost search traffic overnight because somebody forgot the redirects.
Cutting it in slices
Martin Fowler popularized a calmer route, the strangler fig pattern. New services take over one function at a time, say search, cart, pricing or stock, while the old system keeps running. Both live side by side until the new one earns trust.
Bintime’s modernization offer follows that logic. An assessment and migration plan arrive in the first few weeks, then the work runs in phases over a year or two, so a business approves one phase at a time. Scope covers replatforming, splitting monoliths, data pipelines, and connections to ERP, PIM, WMS and CRM systems, delivered from the EU under ISO 27001 and GDPR.
What a retailer can do this quarter
- Audit product data. Missing attributes break search, filters and AI at the same time.
- Map the integrations. Who talks to whom, how often, and what happens when it fails.
- Fix one journey. Checkout, returns or click and collect.
- Set a stock-accuracy goal. If the app says “in stock”, it should be true.
- Pilot one AI use case on real data. Search or product enrichment before try-on gimmicks.
- Write things down as the work goes. Otherwise the new system becomes next year’s legacy.