Why two teams get different results from the same AI

The claim

Give two teams inside the same company the exact same AI tool, same model, same subscription, often even similar prompts, and their results can differ enormously. The usual explanation is that one team is just "better at AI." That explanation is incomplete enough to be actively misleading.

The incomplete explanation

"Better at prompting" implies the gap is a wordsmithing skill: one team knows the tricks, the other doesn't. That's a small part of the picture at best. It doesn't explain why the same person's output from the same tool improves dramatically the moment they're handed better material, with their prompt barely changed. If the skill were really about phrasing, that improvement wouldn't track material quality the way it consistently does.

The actual variable

The teams that get consistently good results tend to be the teams with tighter operational knowledge: clearer internal documentation, faster access to the specific facts of a given case, a shared understanding of exceptions and edge cases that doesn't live only in one person's head, which is context debt by another name when it's missing. None of that shows up as "AI skill" in any obvious way. It shows up as the material that gets fed to the model, explicitly in a written brief or implicitly in what the person typing the prompt already knows and includes without thinking about it.

Why this looks like an AI problem when it's an information problem

A team with messy internal knowledge gets mediocre AI output and often concludes the tool itself isn't very good for their use case. Meanwhile the team next door, doing ostensibly the same job with better internal documentation, gets consistently strong results from the identical tool. Both teams are right that their experience differs. Neither is right about why. The tool didn't change between the two rooms. What each room had to work with did.

What "good with AI" actually means

Teams that look like AI power users are usually the ones that had their operational knowledge in order before the technology arrived. They're now translating what they already had into a form a model can use. That's a sign of organisational capability, not a new AI skill, which is why telling a struggling team to "use AI more" doesn't help it catch up. The gap was there before the tools were.

The pattern repeats at scale

The same dynamic that separates two teams inside one company separates companies inside a sector. A company whose internal knowledge is scattered, undocumented, and exception-riddled will get worse AI output than a competitor with the same tools and cleaner internal material, and the gap will look, from the outside, like one company is simply "better at AI" than the other. It rarely is. It's usually further ahead on a problem that predates AI entirely.

Five patterns I keep seeing in large AI rollouts finds the same pattern one level up, across large organisations' rollouts rather than between two teams.