There's a familiar pattern in business technology. Something new comes along, everyone gets excited, executives start asking how quickly they can implement it, and before long there's a project underway with a budget, a steering committee and a collection of impressive-looking PowerPoint slides.
At the moment, that something is artificial intelligence.
Don't get me wrong. I'm genuinely excited about AI. I use it, I see enormous potential in it, and I think it's going to fundamentally change how businesses operate. But I've also spent enough time working with organisations and their technology to recognise what happens when enthusiasm for a solution gets ahead of understanding the problem.
And AI is particularly good at making that mistake expensive.
We've seen this movie before
Before AI, there was robotic process automation. Before that, digital transformation. Before that, enterprise systems that promised to revolutionise how organisations worked. Every generation of technology arrives with the promise of greater efficiency, better customer experiences and lower operating costs.
And every generation produces businesses that spend a fortune implementing new technology, only to discover that the underlying problems haven't gone anywhere.
I've seen organisations invest significant time and money automating processes that probably shouldn't have existed in the first place. Processes with unnecessary approvals, duplicated data entry, confusing ownership and workarounds that have somehow become accepted business practice.
Nobody particularly likes these processes. Nobody can quite explain why they work the way they do. But because they've been around for years, everyone assumes there's probably a good reason.
Then someone suggests adding AI.
The result might be faster. It might even be cheaper. But it's still a badly designed process, now with considerably more sophisticated technology attached to it.
Just because you can automate something doesn't mean you should
Imagine a business where a customer request passes through five different teams before anyone can make a decision. Each team reviews the same information, asks a few questions and forwards it to the next person. Sometimes it gets sent backwards because somebody forgot to complete a field.
The business identifies this as an opportunity for AI. Perhaps an intelligent agent could read the request, classify it, summarise the information and send it to the right team.
That sounds useful. And it might be.
But here's the question I'd ask first: why does the request need to go through five teams?
Maybe the answer is regulatory. Maybe there are genuine risks that require different people to be involved. Or maybe the process evolved over ten years because nobody ever stopped to redesign it.
If you can eliminate three of those steps, clarify who owns the decision and simplify the information required, you might achieve a bigger improvement before writing a single line of code.
Then, once the process makes sense, AI might make it even better.
That's the order I'd prefer.
The hardest part is usually understanding what's actually happening
One of the first things I try to understand when looking at a business problem is the difference between how a process is supposed to work and how it actually works.
Those are frequently two very different things.
There's the documented process, usually represented by a neat flowchart. Then there's the real process, involving spreadsheets, emails, conversations, exceptions, unofficial shortcuts and the one person who knows how to fix everything when it goes wrong.
That person is often the most important part of the entire operation. Unfortunately, their knowledge usually lives entirely inside their head, and everyone gets nervous whenever they book annual leave.
If you introduce AI without understanding those realities, you're building a solution around an incomplete picture.
You might automate the official process while completely missing the reasons people work around it. You might remove a manual check that looks unnecessary but actually prevents a costly mistake. Or you might create a system that performs beautifully during a demonstration and falls apart the moment it encounters a real customer.
Good process improvement starts with curiosity. Talk to the people doing the work. Follow a transaction from beginning to end. Understand the exceptions, the frustrations and the decisions people make along the way.
The answers are rarely found in the original project brief.
AI doesn't magically fix bad data
Here's another uncomfortable truth: a surprising number of business problems that look like technology problems are actually data problems.
Information is duplicated across systems. Customer records don't match. Important fields are missing. Different departments have different definitions of the same thing. Someone's spreadsheet has quietly become the most reliable source of information in the organisation.
Now imagine asking an AI system to make decisions using that information.
AI can help identify inconsistencies, classify messy information and improve data quality. But it can't automatically know which of three conflicting customer records is correct, or whether an outdated business rule should still apply.
And when AI produces a confident answer based on unreliable information, the problem can become harder to spot.
A person might look at a spreadsheet and say, "Hang on, that doesn't look right."
An AI-generated answer can look remarkably polished while being completely wrong.
Before handing important work to an intelligent system, businesses need to understand the quality of the information feeding it, who owns that information and how errors will be identified and corrected.
It's not particularly glamorous work. It doesn't make for an exciting keynote presentation.
It is, however, rather important.
Sometimes the best AI solution is no AI at all
This might be an unpopular opinion in a world where everything apparently needs an AI-powered assistant, but not every business problem requires artificial intelligence.
Sometimes a better form is enough.
Sometimes the answer is integrating two systems that should have been talking to each other years ago.
Sometimes it's removing an approval, clarifying a policy or giving employees access to information they already need.
And sometimes a straightforward automation rule will do the job more reliably, more cheaply and with less complexity than an AI model.
The question shouldn't be, "Where can we use AI?"
It should be, "What's stopping us from achieving the outcome we want, and what's the simplest effective way to fix it?"
If AI is the best answer, fantastic. Use it.
But if you already know exactly what should happen every time, you may not need a system that makes predictions. You might just need a system that follows the rules.
There's no prize for using the most fashionable technology to solve the simplest problem.
Where AI genuinely changes the game
For all of that, I don't want to undersell what AI makes possible.
There are plenty of business activities that have traditionally been difficult or expensive to automate because they involve interpreting information, understanding context or dealing with variation.
Reading and classifying documents. Extracting useful information from unstructured text. Summarising complex material. Helping employees find answers across large collections of information. Supporting customer service teams with relevant context. Identifying patterns that might otherwise go unnoticed.
These are areas where AI can make a meaningful difference.
It can also change how quickly organisations experiment. Ideas that once required substantial development effort can sometimes be tested in days. Teams can build prototypes, explore different approaches and learn what works before committing to a large investment.
That's enormously valuable.
But the most interesting opportunities usually emerge when you understand the business well enough to recognise where AI can remove friction, improve decisions or create a genuinely better experience.
The technology creates possibilities. Understanding the business tells you which possibilities are worth pursuing.
Don't forget the humans
There's another part of AI adoption that doesn't receive nearly enough attention: the people expected to work with it.
Businesses can become so focused on what the technology can do that they forget to consider what changes for the employees using it.
Who is responsible when an AI-generated recommendation is wrong? When should someone review a decision? What happens when the system encounters something it hasn't seen before? Do employees understand its limitations, or are they expected to trust whatever appears on the screen?
And perhaps most importantly, have the people doing the work been involved in designing the solution?
I've always believed that the people closest to a process often have the clearest understanding of what's wrong with it. They know where time is wasted, which rules make no sense and what customers actually complain about.
Leaving them out because the project is supposedly about technology is a great way to build something nobody wants to use.
AI adoption isn't just a technical implementation. It's a change to how people work, make decisions and take responsibility.
If you ignore that, the technology can be impressive while the outcome is disappointing.
Start with the problem, not the presentation
If I were looking at introducing AI into a business, I'd start with a few fairly ordinary questions.
What are we actually trying to improve? How does the process work today? Where does the time go? What causes errors or frustration? What would success look like, and how would we measure it?
Then I'd ask what could be simplified or eliminated before considering automation.
Only after that would I look at where AI might add value, what data it would need, what risks it introduces and how we'd know whether it was working.
None of this sounds particularly revolutionary. That's because it isn't.
It's the same disciplined thinking that should sit behind any good business improvement initiative. AI doesn't make those fundamentals less important. If anything, it makes them more important because the technology allows us to move so quickly.
And moving quickly in the wrong direction is still moving in the wrong direction.
The opportunity is bigger than AI
I think we're entering a genuinely exciting period for business technology. AI will create opportunities to rethink processes, products and entire business models in ways that weren't practical even a few years ago.
But the organisations that benefit most won't necessarily be the ones that deploy the most AI tools or announce the most ambitious strategies.
They'll be the ones willing to question how their businesses actually work.
They'll simplify what doesn't need to be complicated, fix the foundations that aren't working and use technology where it genuinely improves the outcome.
Because the goal was never to have more AI.
The goal was to build a better business.