The fastest way to waste money on AI

Over the last 18 months, we've seen organisations rush to adopt AI in much the same way they rushed to adopt cloud, mobile apps, and digital transformation before it.

The pattern is remarkably consistent - a new model is released, a competitor announces an AI initiative, an executive sees a compelling demo online. Suddenly, teams are under pressure to ‘do something with AI.’

The problem is that most organisations start with the technology rather than the business problem they're trying to solve. Companies become fascinated by models, agents, automation, and the latest capabilities. They start discussing which AI tools to use before they've defined what success actually looks like. That's usually where value starts to disappear.

AI is not the strategy

One of the biggest misconceptions in the market right now is treating AI adoption as a strategy in itself. It isn't. AI is simply a capability.

Nobody ever said, ‘Our strategy is databases’ or ‘Our strategy is APIs.’ They were tools used to achieve business outcomes - AI should be viewed in exactly the same way. Yet many AI initiatives begin with questions like: How can we use AI? Where can we deploy agents? Which model should we choose? What can we automate? They're reasonable questions, but they're being asked in the wrong order.

The first question should always be: What's preventing us from growing, operating efficiently, or serving customers better? Only once you've identified the constraint should you determine whether AI is the right solution.

Start with friction, not features

The highest-performing AI projects we've worked on all started with a very clear business problem. A sales team spending too much time on administrative work, a customer service function overwhelmed by repetitive enquiries, a marketing team struggling to scale content production, an operations team buried in manual workflows. The common thread isn't AI, but friction.

When teams identify genuine operational bottlenecks first, it becomes much easier to determine where AI can create measurable impact. That's important because AI projects consume real resources. Beyond model costs, organisations are investing engineering time, product resource, implementation effort, governance, testing, maintenance and change management. If you're making those investments, the outcome should be significant enough to justify them.

The agent gold rush

We're currently seeing a huge wave of enthusiasm around AI agents and some of that excitement is justified. Agentic systems will undoubtedly become a major part of how organisations operate over the coming years. But there's also a growing tendency to build agents simply because agents are the fashionable thing to build.

An agent that automates a process nobody cares about is still a bad investment, an agent that saves five minutes per week isn't transformative, an agent that introduces complexity without improving outcomes is simply another piece of software to maintain.

The most successful organisations aren't asking where they can deploy agents but instead where human expertise is being constrained by inefficient systems, fragmented information, or repetitive work. That's a very different conversation.

Measure outcomes, not activity

One of the challenges with AI is that it creates an illusion of progress - it's easy to count prompts, automations, agents, workflows, and deployments, it's much harder to measure whether any of those things have materially improved the business.

The organisations generating the greatest return from AI are typically focused on a different set of metrics: Revenue growth, time saved, operational efficiency, customer satisfaction, employee productivity and cost reduction. Those metrics force clarity. They ensure AI remains tied to business value rather than becoming a technology experiment searching for a justification.

The organisations that will thrive

As AI becomes increasingly accessible, the technology itself becomes less of a differentiator. Every company will have access to powerful models, every company will be able to build agents, every company will be able to automate workflows. The advantage won't come from having AI, it will come from understanding where to apply it.

The organisations that win won't necessarily be the ones building the most. They'll be the ones that understand their business deeply enough to identify the highest-value problems and are disciplined enough to focus their efforts there.

Because the fastest way to waste money on AI is to start with the technology. The fastest way to create value is to start with the constraint.