The most common question commerce leaders ask me right now: where do we start with AI?
It's the wrong question.
The right question is: what is the state of your customer data? Because AI in commerce is not a technology project. It is a data quality project with a technology wrapper. And data quality problems are solved differently than technology problems. Slower, less glamorously, with more organizational friction.
Most companies treat AI implementation as a procurement decision. Find the right vendor, sign the contract, run a pilot, scale. This logic worked for SaaS. It does not work for AI. AI performance is directly proportional to the data it runs on. Most commerce organizations have data that is partial, inconsistent, siloed, or simply wrong.
I have seen this pattern enough times to call it consistent. A promising pilot delivers impressive results: curated data, controlled conditions, a use case selected because it would work. Then it tries to scale. Production data is messier. Edge cases emerge. The model makes recommendations that are technically correct but operationally useless. The project stalls.
Consultants move on. The internal team inherits something that partially works and is politically difficult to kill. The board is told the initiative is "in progress."
This is not a technology problem. It is an organizational honesty problem.
The data debt nobody wants on the roadmap
Nordic e-commerce companies have invested well in technology infrastructure over the last decade. The platforms are modern. The integrations mostly work. But the data layer has not kept pace. The actual quality, consistency, and governance of customer and product data has not been the visible problem, so it has not been treated as a problem at all.
A product catalogue with 15% incorrect attributes does not slow down the website. It does slow down an AI recommendation engine trying to understand product relationships.
The investment to fix this is real: time, headcount, unglamorous work. It competes with roadmap items that are more visible and easier to explain in a board meeting. So it gets deferred. The AI project reveals it.
Before the next vendor demo
Run an honest internal audit first. Customer identity: do you actually know who your customers are across touchpoints, or do you have a fragmented picture that looks clean in a dashboard but is not? Product data: is it complete, consistent, and structured in a way a machine can parse? Behavioral data: are you capturing what you need, and is it reliable?
This audit will not make a good slide. But it will tell you where your actual ceiling is before you discover it six months into a contract.
The companies moving fast on AI did not find a shortcut around this. They started the data work earlier, usually because something else forced them to: a platform migration, a compliance requirement, a replatforming. Their AI readiness is the consequence of that earlier work.
When the vendor shows you what the product can do, ask what data quality their solution requires to perform as shown. Ask what happens when the data does not meet those requirements.
The answer will tell you more than the demo will.