There’s a pattern We keep seeing across ecommerce teams now a days. Somebody at the leadership level gets excited about AI — reads a few case studies, watches a demo, maybe attends a conference — and within a few weeks the company is paying for three or four AI tool subscriptions. Copywriting. Personalization. Demand forecasting. Customer support automation. The seats get bought. The Slack announcement goes out. And then, mostly, nothing changes.
Not because the tools are bad, but because the teams buying them haven’t done the work that makes those tools useful.
I watched this play out with a mid-size apparel brand last year. They brought in an AI tool to generate product descriptions at scale — reasonable problem to solve, they had thousands of SKUs and a small copy team that was perpetually behind. The tool worked fine in the demo, in production, it started pulling from their existing product data, which turned out to be a mess. Inconsistent material fields, missing sizes, wrong attribute tagging that varied by whoever had done the data entry that week. The AI didn’t know any of this it just filled in the gaps confidently. Wrong fabric compositions, inaccurate sizing guidance.
The descriptions went live, and the return rate started climbing before anyone connected the two. By the time the team traced it back, they’d already generated and published several hundred listings.
The AI wasn’t the problem it was doing exactly what it was built to do — generate fluent, confident-sounding copy from the inputs it was given. The problem was the inputs. Automation scales whatever you feed it, including confident-sounding errors, and it does it faster than any human team would have.
The brands I’ve seen actually extract value from these tools share one thing in common: they did boring work before they did exciting work. Cleaned their product data, standardized attribute tagging, built proper customer segmentation, mapped their actual workflows instead of assuming the AI would figure it out. None of that is interesting to talk about.
Nobody’s writing case studies about the three months a team spent fixing their PIM before touching a single AI integration. But that’s where the results came from.
The AI, when it finally arrived, wasn’t the investment. It was the reward for the infrastructure work most teams keep deferring.
What makes this tricky is that the tools are genuinely good enough now that you can get impressive-looking output almost immediately. A demo always works and the first week usually looks promising. So teams assume the hard part is the procurement decision — picking the right vendor, negotiating the contract, getting IT to approve the integration.
BUT, the hard part is actually everything that comes after. The data pipelines, edge cases, places where the AI confidently does something wrong and nobody catches it for two weeks because the output volume is too high to manually review.
I’m not convinced the bottleneck is the AI.