PwC published a study this year that should be required reading for every business leader thinking about AI.

The headline finding: three-quarters of the economic value being generated by artificial intelligence is flowing to just 20 percent of companies.

That means four out of five businesses are investing in AI and getting marginal returns. The fifth is pulling ahead at a pace that is going to be very hard to close.

I have spent the last two years working with businesses at different stages of AI adoption, and I have watched this divide form in real time. The question I keep coming back to is not “why is AI hard” but “what specifically are the top 20 percent doing that the other 80 percent are not?”

I think the answer is clearer than most people want to admit.

The wrong question is costing you

Most businesses I talk to are asking one of two questions about AI:

“Which tool should we buy?” Or: “How much does it cost?”

Both are the wrong starting point. And the research backs this up.

PwC’s analysis found that the AI leaders they studied were not more likely to have bigger AI budgets or access to better technology. The separating factor was strategic orientation. AI leaders are approximately two to three times more likely to use AI to identify and pursue growth opportunities and to reinvent their business model.

The laggards, the 80 percent, are using AI primarily to cut costs and improve efficiency in existing processes. That is not wrong. Efficiency gains are real. But they are the floor, not the ceiling. And they are the gains that are easiest for competitors to replicate once they catch up.

The leaders are using AI to do things that were previously impossible for a business their size. They are entering new markets with a fraction of the headcount it would have taken before. They are building products that personalise at scale. They are capturing intelligence from every customer interaction and feeding it back into the business in ways that compound over time.

Efficiency gains plateau. Capability gains compound.

Why the gap is widening, not closing

Here is what makes the current moment particularly important: the gap between AI leaders and laggards is not staying constant. It is growing.

Stanford’s AI Index 2026 found that enterprise AI adoption has crossed the kind of inflection point that the internet reached around year three of mainstream adoption. Deployment is accelerating, capability is increasing, and the organisations that established strong foundations early are now running experiments and deploying applications at a pace that latecomers cannot easily replicate.

The reason is structural. Building with AI is not like buying software. It is not a subscription you turn on and immediately access the full capability. Real AI deployment requires data infrastructure, process clarity, skilled people, and accumulated operational learnings. These take time to build. And the companies that started building eighteen months ago are eighteen months ahead on all of them.

BCG research puts a concrete number on this divergence. Companies that approached AI as a growth and reinvention tool, rather than purely a cost tool, are now generating returns that are two to three times higher than peers in the same industries with comparable AI investments.

Same budget. Two to three times the outcome. The difference is what they aimed at.

The three things AI leaders do differently

After working through enough deployments to see patterns, and reading enough research to pressure-test those patterns, I keep coming back to three operational differences between the top 20 percent and everyone else.

They built data foundations before building AI applications.

This is the most common place I see businesses stumble. They want to deploy an AI agent, so they try to connect it to their systems, and they discover that the data is a mess. Incomplete records, inconsistent formats, information siloed in spreadsheets, customer data in four different systems that do not talk to each other.

AI is not magic. It works with what you give it. A business with clean, connected, well-structured data can deploy capable AI in weeks. A business with fragmented data infrastructure spends months just getting to the starting line.

AI leaders invested in data quality and data architecture before the AI wave hit. Not because they predicted it. Because good data is good business practice regardless of AI. But it turns out that investment is now one of the most valuable things they did.

They treated AI as a capability, not a project.

Most businesses approach AI as a discrete initiative. They form a task force, run a pilot, write a report, present to the board. The pilot either fails quietly or succeeds and then sits in a deck while everyone debates next steps.

AI leaders treat it differently. They embed AI into how teams work. They measure outputs, not activities. They give people permission to experiment within guardrails and share what they learn. They move from “we ran an AI pilot” to “we run our business with AI” as a posture.

This sounds like a cultural observation, and it is. But it has practical consequences. A business that runs AI as a project will always be in pilot mode. A business that runs AI as a capability keeps getting better at it, week by week, month by month.

They upskilled broadly, not just in IT.

Deloitte’s 2026 enterprise AI research found that 70 percent of companies are now deploying generative AI in at least one function. But the businesses seeing the highest returns are deploying it across functions, not just in the functions where the IT team has influence.

The difference is workforce readiness. AI leaders invested in raising the AI literacy of their entire workforce, not just their technical staff. The marketing team can prompt effectively. The operations manager can evaluate which workflows are candidates for automation. The CFO can read an AI impact report and ask the right questions.

When only the technical team understands AI, you get technically sound deployments that do not get used. When the whole team has enough literacy to participate, you get adoption.

This is something we have seen clearly through EDNA Learn. Businesses where leadership has invested in team-wide data and AI literacy consistently make better AI decisions, adopt tools faster, and get measurably better ROI from the same technology. The research on what data-literate organisations actually achieve makes the case plainly — the performance gap between data-driven and data-blind businesses is large and widening. Knowledge is not just educational. It is structural. It changes who participates in decisions and how quickly the organisation moves.

What the 80 percent can actually do

If you are reading this and recognising yourself in the laggard group, the path is not mysterious. It is just work.

The first step is an honest assessment of your data foundations. Not an aspirational one. What data do you actually have? How clean is it? Where does it live? Who owns it? If you cannot answer those questions confidently, that is where you start.

The second step is getting at least one real AI deployment into production. Not a chatbot on your website that you set up with a free tool. An actual workflow that your team uses every day, that replaced or augmented something they were doing manually, and that you are measuring. If you are not sure whether your foundations are ready, there are three specific prerequisites most businesses skip — checking those first will save you a failed deployment. One real deployment teaches you more than twelve pilots.

The third step is building literacy. This is the one most leaders underestimate. Your team’s ability to evaluate, use, and improve AI tools is a bottleneck that no external vendor can solve for you. Getting your team started with data and AI does not have to be a year-long programme — there is a practical path that produces visible results in weeks. Training is not overhead. At this point, it is infrastructure.

The fourth step, and the one I think is most underutilised, is getting external expertise on the strategic layer. Most business leaders are not AI strategy experts. They should not have to be. But they do need access to someone who can look at their business, understand where AI creates asymmetric value, and help them sequence the investments correctly so they are not wasting money on the wrong things in the wrong order.

This is exactly what we built Omni Advisory to do. Not to replace your internal team. To give you the strategic context that makes your team’s decisions better.

The window for catching up is still open

I want to be honest with you: the gap is real, and it is growing. If you have been waiting for AI to “mature” before committing, you have already paid a cost. The leaders are not going to slow down for you to catch up.

But the window is not closed. The businesses I have seen go from AI laggard to AI leader in the shortest time share one characteristic: they stopped waiting for certainty and started building with urgency.

AI is not going to get simpler in a way that makes the decision easier. The tools will improve. But the competitive landscape will also intensify. The advantage of moving now is not that you get the best technology. It is that you start accumulating the data, the experience, and the institutional knowledge that turns good technology into lasting advantage.

The 80/20 split is not permanent. But it will not change on its own.

---

If you are trying to figure out where your business sits in this landscape, and what the highest-leverage moves are given your specific situation, that is exactly the conversation we have with business leaders at the start of every Omni Advisory engagement. Book a discovery call and we can talk through what actually makes sense for where you are right now.

And if the data foundation and team literacy piece is where you want to start, EDNA Learn has built programs specifically for business teams, not just individual analysts. The fastest path to AI returns starts with knowing what you are working with.