The Question That Separates the 20% from the 80%
74% of AI's economic gains go to just 20% of companies. The gap isn't budget or talent. It's the question they started with.
I've been building AI features for over a year now.
Some moved numbers. Some didn't.
For a while I assumed the difference was execution — better prompting, better models, better integration. But the pattern that kept emerging wasn't about any of that. The features that moved numbers and the features that didn't started with different questions.
The ones that didn't: "How do we do this faster?"
The ones that did: "What can we now do that we couldn't do before?"
Both questions produce working software. Only one produces compounding growth.
What the data says
PwC's 2026 AI Performance Study puts a number on this: 74% of all AI economic gains go to just 20% of companies. The other 80% are spending real money, running real pilots, building real features — and watching the results land somewhere between modest and invisible.
The gap isn't budget. Companies in the bottom 80% aren't underfunding AI. The gap isn't talent — the tools have democratized fast enough that a small, technically capable team can build almost anything. It isn't even the model. The same foundation models are available to everyone.
The gap is the question they started with.
The 20% ask: what becomes possible now that wasn't possible before? They look for the revenue line that didn't exist, the customer segment they couldn't serve, the product capability that changes the value proposition entirely.
The 80% ask: what do we already do, and how do we do it faster? They find a workflow, automate it, measure the time saved, and declare success. The time savings are real. The business impact is limited.
Why the efficiency question is seductive
The efficiency question is easier to ask because it has an obvious answer. You have a process. You can measure how long it takes. You can calculate the time saved. The ROI is clean and immediate.
The growth question is harder. It requires imagining something that doesn't exist yet — a customer you can't currently reach, a product you can't currently ship, a decision you currently can't make fast enough to matter. There's no obvious workflow to point at. The ROI isn't calculable upfront.
That asymmetry explains almost everything about how AI adoption plays out inside organizations. The projects that get greenlit are the ones with clear efficiency metrics and predictable returns. The projects that compound are the ones that couldn't be justified on a spreadsheet before they worked.
What this looks like in practice
At Rapid Data, building Rapid Next, I see this distinction constantly. The efficiency version of our AI roadmap would be: take the things our users already do — data entry, document generation, appointment scheduling — and make them faster. Automate the routine. Reduce clicks.
That's valuable. Our users would notice. Time savings in a busy funeral home are real and meaningful.
But the growth version asks a different question: what can a funeral home now do that they genuinely couldn't before we built this? Can they serve families who aren't local? Can they coordinate across multiple locations without additional headcount? Can a small operator compete on service quality with a chain that has five times their staff?
Those questions are harder to answer. The features are harder to spec. The success metrics are less obvious. But the ceiling is completely different.
Efficiency has a ceiling. You can only compress a process so far before you've removed all the slack and there's nothing left to automate. Growth compounds — each new capability enables the next one, and the gap between you and competitors who are only optimizing widens every quarter.
The practical implication
This isn't an argument against efficiency. Reducing friction, eliminating manual work, compressing cycle times — all of it matters, especially in operationally intensive businesses where people are spending hours on administrative tasks that could be automated.
The argument is about sequencing and proportion.
If your entire AI roadmap is efficiency projects, you're building a slightly leaner version of what you already are. If your roadmap includes growth projects — features that open something genuinely new — you're building a different company.
The 20% in PwC's study aren't ignoring efficiency. They're not sacrificing operational improvements for some theoretical growth upside. They're doing both. The difference is that they reserved some of their roadmap for the harder question.
The efficiency question is where most AI programs start. The growth question is where the compounding begins.
If you're building AI right now: look at your roadmap. Count the efficiency projects and the growth projects. If the ratio is 10:0, you already know what to do next.