Nobody praises a checkout that works. Stock numbers match the shelf, the promo code behaves, orders reach the warehouse, and the whole thing stays invisible. Then a big sale starts, a vendor pushes a bad update, or someone talks a help desk into a password reset. Here’s a look at the weak spots retailers tend to find late, a few well-known cases, what the industry is trying out, and a short list worth checking before peak season.
The Stack Nobody Looks At
Retail systems tend to pile up like stuff in a garage. A till bought years ago, a Shopify or Magento storefront, some warehouse software, a loyalty app from an agency that stopped answering emails. Plus a spreadsheet where the real stock numbers live, because everybody trusts it more than the system. Each part works fine. Between the parts it’s CSV exports and good will.
Another plugin is the usual answer, and once or twice that’s okay. After the third patch for the same gap, a team might get further with custom software development services, built in-house or by an outside studio that knows retail and omnichannel, with ERP, PIM and WMS integrations covered. One proper integration layer beats a drawer full of connectors, and it’s a lot easier to debug at 6 a.m. on a sale day.
Why It Goes Unnoticed
Dashboards show revenue and conversion. Hardly any show how many orders are stuck in the sync job, or how many webhooks failed overnight. Uptime only says the server is alive. It says nothing about whether the order actually reached the warehouse. A site can look perfectly healthy while a handful of payments quietly fail, and finance finds out at month end, usually with a sigh.
There’s also the human side. Everyone has learned to work around the glitch. The store assistant knows the stock count for that one product is always wrong. The marketer knows the discount code needs a second click. The workaround becomes folklore, and folklore never makes it into a bug ticket.
The Usual Suspects
- Stock drift: the website says three left, the shelf has none, and the customer gets a cancellation email two days later
- Promo codes that pile up in ways nobody planned, such as a welcome code plus a loyalty discount plus free shipping on a low-margin item
- Different prices in the app, online and at the till
- Returns handled in a tool the stock system never hears about
- One payment provider, one carrier API, no plan B
- Product data that lives in five places and agrees in none
- Alerts sent to the mailbox of someone who left in spring
Cases Worth Remembering
Traffic Spikes
Ticketmaster’s Eras Tour presale in 2022 buckled under demand its tests hadn’t predicted, and the public sale got cancelled. Amazon’s Prime Day 2018 opened with error pages and photos of dogs. Companies like these do run load tests. They just tend to model polite shoppers, while real ones refresh, open three tabs and bring bots along. Smaller shops get the same treatment on a smaller scale: one influencer post, and a checkout built for a quiet Tuesday falls over.
A Bad Update
CrowdStrike, July 2024. A faulty update froze Windows machines across airlines, hospitals and shops. In some stores the registers and self-checkouts went dark, and staff dug out paper and pens. Chains with a rehearsed offline mode coped far better. Plenty didn’t have one, because who rehearses for a problem that has never happened?
The Side Door
Target’s 2013 breach began with credentials taken from an HVAC contractor. M&S and the Co-op, hit in spring 2025, showed a newer version of the same story. Reports pointed to social engineering aimed at a third-party service desk, and online ordering at M&S stayed paused for a long stretch. A good firewall doesn’t help much when a convincing caller gets a password reset.
The Big-Bang Rewrite
Hershey switched on a new ERP just before Halloween in 1999 and struggled to ship candy during its busiest weeks. Lidl scrapped a long, costly SAP-based rollout in 2018. Replace everything at once and hope for the best: that pattern keeps failing. Module by module, with the old system running alongside, works better. It’s slower and far less exciting for the steering committee, but the shelves stay full.
The Quiet Data Mess
Not every failure makes the news. Plenty of retailers lose weeks to something as plain as product data. Sizes entered as “M”, “Medium” and “med.” Photos attached to the wrong variant. A supplier feed that changed its column order on a Friday. Cleaning that up is mostly integration work, the kind of job studios such as https://bintime.com/ take on with PIM and ERP feeds, and it pays back quickly. Nobody writes a post-mortem about bad data, yet it breaks marketplace listings, ad feeds and search results in one go.
What Retailers Are Testing
Cashierless Stores
Amazon withdrew Just Walk Out from its US Fresh grocery stores in 2024 amid reports that many transactions still needed human review, and moved toward Dash Carts. A demo that hides manual labor behind the curtain is still a demo. Self-checkout has had its own wobble: several chains have pulled back in some locations after theft and customer complaints, and Booths in the UK took most of its machines out in 2024.
Smart Shelves
Walmart has been rolling out electronic shelf labels, and Simbe’s Tally robots roam aisles at chains like Schnucks and BJ’s Wholesale, checking for gaps and wrong price tags. Unglamorous stuff. But a store manager notices when the price complaints stop, and a price change that used to take an evening of sticker printing now takes a click.
Agents
Salesforce pushes Agentforce. OpenAI and Stripe introduced the Agentic Commerce Protocol in late 2025, and Anthropic’s Model Context Protocol gives agents a standard way into inventory and order data. Good idea, provided the product data is clean. An agent can’t recommend a jacket whose size chart is a JPEG. Plugging an agent into a messy stack just automates the mess faster.
Breaking Things on Purpose
Netflix’s Chaos Monkey made chaos engineering famous. Retail teams copy it in a lighter form: a game day where the payment provider gets switched off in staging, to see what customers would actually get. Uncomfortable, yes. Still cheaper than finding out in November.
Who Owns the Problem?
Here’s a pattern that repeats from small shops to large chains. Marketing launches a campaign. Operations hears about it on the morning it starts. IT gets the angry call when the site slows down. Three teams, one failure, zero owners.
The retailers that cope well usually do something simple. They name one person per critical flow (checkout, stock sync, returns) and give that person the right to say “not before the sale.” They keep a short, plain-language runbook for the nasty scenarios: payment down, warehouse system down, store network down. And they run it past people who didn’t write it. A runbook that only its author can follow isn’t a runbook.
Terry Clark is the Publisher and Content Director of 365 Retail, with more than a decade of experience covering retail design, technology innovations, store openings and the wider retail industry. He also works closely with leading retailers, suppliers, agencies, events and industry awards across the UK.
