The cheapest way to be wrong

Edition 047 · 3 September 2026

This post initially appeared on CEO & Founder, Jamie Huskisson’s newsletter, Breakthrough Commerce.

Being wrong is supposed to be expensive. That’s the assumption sitting underneath most ecommerce roadmaps, and it’s not an unreasonable one to hold. Retailers replatform, get it wrong, and are staring at administration a few months later because the new site converts worse than the old one and there’s no budget left to fix it. Digital transformation programmes get shelved halfway through and take a round of redundancies with them on the way out. Learn that lesson enough times and the conclusion writes itself: be careful, get it right first time, because being wrong will cost you the business.

De-risking ecommerce through smaller, reversible bets

That’s not the theme that ran through Meet Magento UK this year, though it kept surfacing anyway. Across three separate sessions, four speakers who’d never compared notes said the same thing five different ways: the cost of being wrong is something you control, not something you’re stuck absorbing. That’s the belief we built JH around from the very start, so there was something satisfying about sitting in the audience and watching it turn up on stage, unprompted, from people who’d never heard us say it.

Paul Ryazanov, CEO of MageCloud, told the story that got the biggest laugh in the room. There’s a website called ZagorGuitar.com that still looks like it was designed in 1985, beige background, no parallax in sight, exactly as it’s always been. A few years ago the owner tried to modernise it. Moved to Shopify, picked a template loaded with effects, went live. Sales dropped fifty per cent overnight. He rolled it back that same day and never touched the site again. It’s still there, still stuck in 1985, because the owner learned the expensive way what the rest of the room spent two days explaining how to avoid.

Ryazanov’s own answer is to take the risk off the client’s side of the table entirely. His agency doesn’t invoice until the work is delivered, and he pays for convert.com, the testing platform his team runs on, out of his own margin, because he’s confident enough in the result to carry that cost himself. His checkout guarantee is just as blunt: if it doesn’t convert at least fifteen per cent better than what’s live, the client doesn’t pay. The work behind that guarantee is granular. His team splits checkout visitors into five behavioural types (regular buyers, discount hunters, review readers, hesitant browsers, comparison shoppers) and serves each one a different widget at the one point in the funnel Shopify won’t let you touch and Magento will.

Maris Skujins, who runs digital commerce strategy at Scandiweb, made the same case from the architecture side. There’s always a route to where you’re going, he said, and there’s rarely only one. His team recently took on a UK B2B retailer that had to get off a legacy custom platform by a fixed date, big business, complicated processes, no room to phase it. But where the deadline allows, his default is to rip the machine’s process out and replace it piece by piece rather than commit to one irreversible migration. Same destination, reached through moves small enough to reverse if one of them goes wrong.

Hugo Wilson ilson, senior experimentation consultant at Conversion , argued for testing as the thing that stops your own confidence lying to you. “You’re making a small bet to understand the learning,” he told the room, and the alternative is worse than most teams admit. He sees clients who redesign a hundred things at once and walk away certain they’ve made a hundred improvements. In practice it’s closer to thirty real wins, fifty that changed nothing, and twenty quietly making things worse, with no way to tell which twenty. A big, unbroken launch doesn’t just risk failure. It destroys the evidence you’d need to learn from it either way.

Matthew Lawson, Chief Digital Officer at Ribble Cycles, had the two stories that tied the whole thing together. The first was a recovery. He backed a new front-end framework for a headless rebuild, and it was deprecated the day the site went live. He didn’t have the goodwill to go back to the board and ask for a second replatform, so he went to his hosting partner instead, and together they built a way to run two front ends at once, migrating page by page as the new one matured. The recovery from a bad bet was itself a small bet, not a second big one.

The second story was smaller still. His CEO built a store-locator app using Claude, in an evening, to see which branches overlapped and where the real gaps in the network sat. Nobody’s opening or closing stores off the back of a weekend prototype, and that was never the point. It gave the CEO a way to show the board what he meant instead of trying to describe it, and an evening’s build did more convincing than a slide deck ever would have.

None of them had planned to say the same thing. They landed on it separately because it’s true regardless of which part of the business you’re standing in, and because the expensive version of wrong is sitting in plain sight at every retailer who replatformed once and never got the chance to try again. Run three to five bets instead of seventy-five, by all means. Just make sure each one is small enough to survive being wrong.

The size of the bet is the risk control. Not the confidence behind it.

One more thing. We’re in the process of launching video from the conference on our YouTube channel, so you can watch all of these talks in full. We’ve also got an ebook of this year’s sessions on the way, and you can sign up to get it the moment it lands.

De-risking ecommerce is what we do. Get in touch if you’re ready to talk.