AI Agents Won't Wait for Your Overnight Stock Sync

I recently shared the story of the Christmas experiment: a seasonal brand briefed three agencies to build competing shops, same products, same trading window, same marketing budget. We cost double and generated ten times the revenue. At Codegarden, someone sensibly asked: what actually made the difference?

Part of the answer deserves its own post, as it's about to matter far more than it did then.

The ring-fenced stock problem

From what we understood of how the other two sites ran, their stock updated on a schedule, typically overnight. One of them, I think, was worse than that. It ran on a manual CSV refresh, so over a weekend the figures could sit untouched for a couple of days. To avoid overselling, each was effectively ring-fencing a small pot of stock, say ten units of a popular line. The website showed ten. Customers bought ten. And then the site showed nothing, even when there was plenty in the warehouse, because the feed didn't update until the next refresh.

In peak Christmas trading, on their best-selling products, those sites would have been closed for business hours at a time while holding stock nobody could see.

Live syncing every product all day is expensive, and most of that spend goes on slow movers that barely shift. There's a quieter cost too. Sync everything in real time and the storefront's uptime is now tied to the warehouse system behind it, so a bad morning on the ERP turns into a bad morning on the site, usually in the middle of your busiest trading.

Ours did something narrower. It watched for lines running low. Once a product dipped to where its sale rate might empty it before the next scheduled update, each sale queued a fresh count from the warehouse. The queuing mattered. Several sales of one fast-moving line could cluster into a single check, so a popular product wasn't firing a call per order. Most stock never triggered any of it. The lines that mattered in peak week stayed accurate to within a sale or two, and that's all it takes.

I'd be overstating it to pin the whole 10x on stock. Several things pulled in the same direction. Ours was a fully customised build where the others felt templated, the customer journey was tighter, and we ran win-back emails to people who'd browsed and left. Some of that is hard to isolate after the fact. Stock freshness is the one I keep coming back to, because the failure mode is so specific: a best seller showing as unavailable during the hours the warehouse was full of it, in the weeks that decide the year. That's a gap you can price. A nicer checkout is much harder to value.

Stale data doesn't announce itself, either. No error page, no complaint, nothing in the logs. The customer sees "out of stock", quietly buys elsewhere, and the end-of-season report just shows numbers a bit lower than expected. It behaves like a slow leak, which is exactly why it goes unfixed.

Now add agents

Here's why I'd tell this story with more urgency today than when it happened.

A growing share of buying journeys now involve an AI agent somewhere: a shopping assistant comparing options, an agent checking availability across suppliers, a procurement tool re-ordering automatically. However far and fast that trend runs, the direction is clear, and agents are even less forgiving than people.

A human shopper who really wants your product might ring up, check back tomorrow, or chance an order anyway. An agent won't. It reads your data, takes it at face value, and moves to the next supplier in milliseconds. Show "out of stock" while your warehouse is full and you've been silently filtered out of a sale you'd have won, with no signal it ever happened.

Accurate, current, structured data has gone from a customer experience nicety to the actual sales interface. For B2B businesses, where agents are likely to handle a lot of routine re-ordering first, this lands soonest.

If you're a business

Run a simple audit this week. Pick five products and compare what your website says against what your warehouse holds, checked at lunchtime while the shop is trading. If they disagree, work out the gap, then put a revenue number on it using your best sellers, because that's what the gap costs you in peak weeks.

And when you scope your next platform, ask explicitly: which data is live, which is scheduled, and how fast can a stock change reach the storefront? The answers shape the architecture, as we covered in the boring technology post, and now they shape your visibility to agentic buyers too.

If you're an agency or Umbraco partner

Sync design has always been treated as a technical detail, settled late, by developers, on instinct. Promote it. The real-time versus scheduled decision is a revenue decision, and the Christmas experiment is a useful story to borrow: same products, same marketing, a 10x gap, and stock freshness as one of the few differences you can point at and cost.

It doesn't have to mean live everything. The low-stock trigger we used on that build is a decent middle ground: scheduled updates for the long tail, and a fresh check against the warehouse whenever a fast-moving line gets close to running out. Cheap on the integration, accurate where it counts. The threshold for "fresh enough" is tightening on the stock side regardless (fifteen-minute syncs where overnight used to do, event-driven updates for the fastest lines), so ask clients about velocity and tolerance, design to the answer, and put the reasoning in the proposal, where it sells.

The takeaway

Stock data that lags reality is a silent revenue leak, and it's about to get louder. People forgive the odd stale number and move on. Agents don't. Treat data freshness as part of the sales proposition, not the plumbing, and you'll be one of the businesses an agent can buy from.

Tim Gaunt is the founder and CEO of TSD, a UK e-commerce agency that's been designing, building and supporting complex commerce platforms for over 20 years. This post is part of a series based on his Codegarden 2026 talk, The €200k Sweet Spot, and the questions that followed it.