Where is AI already making you money and where is it just making you feel modern?

I gave a talk on this in Edinburgh the other day to a brilliant group of people at SMT Network.

Fast forward to the end to find out which types of AI your organisation uses today and where you might be falling behind.

Most leaders can point to one or two places where AI is genuinely working - and they typically looks like a marketing team using generative tools to produce campaign copy faster or. A customer service chatbot handling the straightforward queries.

But when you push beyond those examples, the conversation gets vague. There is a sense that AI should be doing more, but a lack of clarity about what, exactly, that means.

When we talk about AI as though it is one thing, we make it almost impossible to have a useful conversation about where it fits in a business. It is like talking about "transport" without distinguishing between a bicycle and a container ship. Both move things from A to B, but you would not use them interchangeably.

AI is not one technology. It is four distinct capabilities, each solving different problems at different levels of maturity.

Once you understand the four categories, the strategic picture becomes dramatically clearer.

Generative AI

The one everyone knows. It creates things: text, images, code and video and is the most visible category because it produces output you can immediately see and evaluate.

When JPMorgan uses AI to generate thousands of personalised ad copy variants, or when a professional services firm uses it to draft a first-pass client proposal in minutes rather than days, that is generative AI at work.

The barrier to entry is low while the time to value is fast and most organisations have already started here whether they planned to or not.

Predictive AI

Less glamorous but often more valuable. It forecasts outcomes from historical data. When a retailer predicts demand at store level by analysing weather patterns, local events and economic indicators, that is predictive AI.

When a bank identifies customers likely to leave 60 days before they do, enabling targeted retention offers, that is predictive AI.

It tends to deliver the highest measurable ROI of any category, precisely because it connects directly to revenue and cost decisions. Yet it rarely makes the headlines because there is nothing visually exciting about a probability score.

Traditional machine learning (ML)

The workhorse. It classifies, recommends and detects patterns.

Every time your bank flags a suspicious transaction, that is ML. Every time Netflix suggests something you actually want to watch or a manufacturer catches a defect on a production line before it ships, that is ML.

It is already embedded in most enterprise software, often invisibly. The strategic question is not whether you are using it, because you almost certainly are. The question is whether you are using it deliberately or just benefiting from whatever your software vendors have baked in.

Agentic AI

The most ingnored but expected to be the most valuable. Agentic AI takes actions.

It moves across systems, makes decisions and executes multi-step tasks autonomously.

When Klarna's AI assistant handles two-thirds of all customer service conversations end to end, processing refunds and updating orders without a human ever touching the case, that is agentic AI.

When an investment firm deploys an AI that independently gathers data from multiple sources, builds financial models and produces analyst-grade research, that is agentic AI. It is the category with the most transformative potential and the window to build institutional capability around it is open right now.

Co-Intelligence

Ethan Mollick, a professor at Wharton who has studied AI adoption more rigorously than almost anyone, argues that the biggest mistake companies make is treating AI as something to deploy later, once they understand it better.

His research shows that organisations which let people experiment broadly with AI, even messily, consistently outperform those that try to create a top-down strategy first. The strategy, he says, emerges from the experimentation.

Experimentation without a map leads to a lot of activity in one quadrant and blindness in the other three.

Most organisations are over-invested in generative AI because it is the easiest to access and the most fun to play with. Meanwhile, their competitors may be quietly building predictive capabilities that connect directly to margin improvement, or testing agentic workflows that will fundamentally change how work gets done.

Mollick also makes the point that whatever limitations you see in AI today will be gone within 18 months.

Every investment decision and every strategic choice about AI needs to account for the fact that capabilities are improving at a pace that makes most planning horizons obsolete before they are implemented.

The companies that wait for AI to be "ready" will find that their competitors have already built the muscle memory and institutional knowledge that takes years to develop.

So the practical question for any CEO reading this is not "are we using AI?" because you almost certainly are. The question is: how are you using all four types?

Do you know which quadrant your biggest competitor is investing in? And are you building the organisational capability to absorb the next wave, or are you still congratulating yourself on the chatbot you launched last year?

Now tell me what you think of this tool I made for you with Claude: