It's Not the AI: Why Commercial Pharma Teams Aren't Seeing Gains From AI

Commercial pharma teams aren't seeing the AI gains they were promised. Here's the uncomfortable part: it's not the AI's fault.

The numbers are stark. Only 9% of life sciences leaders say they're getting significant returns from AI, and just 22% have managed to scale it at all.^[1] Yet McKinsey puts $18 to 30bn of AI's annual value in commercial functions specifically.^[2]

So the value is real. The models are extraordinary. And most commercial teams still aren't moving the needle. Why? Because we're all watching the wrong thing.

We're optimising for the thing we can't control

We wait for the next, more powerful model. The newest shiny tool. The release that finally changes everything. But here's what I keep coming back to: that's completely out of your control, and it's getting better for free anyway.

The model is a rising tide. Your data foundation decides whether you've got a boat.

The teams actually getting returns aren't the ones with the cleverest AI. They're the ones whose data was ready for it. Clean pipelines. Consistent structure. The right data, collected and stored in a way a machine can actually use.

That's the part nobody talks about, because it isn't glamorous. But it's the only part you control. And it's the one investment that compounds. Every time the models improve, you benefit more, but only if your foundation can carry it.

Start building the thing you can control

So before you wait for the next model, get the controllable things right:

• Pipelines: can you get data in, clean and current, without manual heroics every time?
• Structure: is it modelled consistently, so a machine can actually reason over it?
• Collection & storage: are you capturing the right data, in one place, governed?
• AI readiness: are your data models built to feed AI, not just fill dashboards?

Stop optimising for the thing you can't control. Start building the thing you can.

At Lumora, it's the pattern we see again and again: the teams winning with AI fixed their data foundation first. The model was never the hard part.