Start With the Process Everyone Hates

Start with the process everyone hates.

Not the one that sounds impressive in a board presentation. Not the one that gets the CEO excited at a conference. The one that makes someone in your operations team quietly sigh every Tuesday afternoon.

Every organisation has one. The invoice matching that takes someone three days a month. The data reconciliation that makes your finance team consider career changes. The report that gets manually compiled every Monday morning from four different spreadsheets that haven't been properly reconciled since 2019.

That's where you start with AI. Not with a grand strategy. Not with a transformation roadmap. With the thing everyone hates doing.

I know this isn't what you want to hear. You've read the headlines. You've seen what ChatGPT can do. You've got a competitor who just announced their "AI-powered customer insights platform" and you're wondering if you're falling behind.

You're not. They're probably six months away from quietly shelving that project. Let me explain why.

Why the exciting projects fail

There's a pattern I see repeatedly, and it goes something like this.

A company gets excited about AI. Someone senior goes to a conference, comes back buzzing, calls a meeting. The team brainstorms ambitious projects. AI-powered customer insights. Predictive analytics for demand forecasting. Autonomous decision-making for supply chain optimisation. The words "digital transformation" get written on a whiteboard. Budget gets allocated. A vendor gets hired.

Six months later, they discover their data isn't clean. Their processes aren't properly defined. Nobody can agree on what success looks like. The people who actually understand the day-to-day operations weren't consulted. The project gets quietly downgraded, then paused, then forgotten about entirely. The vendor still gets paid.

This isn't anecdote. It's statistics.

RAND Corporation found that roughly 80% of AI projects fail. S&P Global reported that 42% of organisations that have deployed AI models end up abandoning them. McKinsey's research shows only 23% of organisations are successfully scaling AI beyond pilot projects.

Eighty percent failure. Let that sit for a moment.

These aren't small companies getting it wrong. These are organisations with dedicated AI teams, significant budgets, and access to the best technology available. They're failing because they're starting in the wrong place.

They're starting with the exciting project instead of the boring one.

The "start with what hurts" philosophy

The best first AI implementation in any organisation shares a few characteristics. It's tedious. It's well-defined. It's low-risk. And it's high-annoyance.

Find the most soul-crushing, repetitive process in your business. The one with clear inputs and clear outputs. The one where failure is cheap and success is immediately obvious. The one where, if you automate it, a specific person's life gets measurably better.

Not theoretically better. Not "strategically aligned" better. Actually, tangibly, I-no-longer-dread-Tuesdays better.

This is counterintuitive because we've been sold a narrative about AI as this transformative, revolutionary force that will reshape entire industries. And it will. Eventually. But it reshapes them one boring process at a time, starting with the ones that hurt the most and matter the least if something goes wrong.

How to find your first process

When I sit down with a business for the first time, I don't ask about their AI strategy. I ask what they hate doing.

Not the leadership team. The people who actually do the work. The operations team. The accounts team. The admin staff. The warehouse coordinator. The person who spends every Friday afternoon copying numbers from one system into another because the two systems don't talk to each other and nobody's had time to fix that since the company switched CRM providers in 2021.

Here's how you find your first process.

Walk the floor and talk to people. Not about AI. About their week. What took the longest? What was the most tedious? What do they wish they could just make disappear? You're not looking for the CEO's vision of the future. You're looking for the operations team's reality right now.

Look for the "Tuesday afternoon problem." Every organisation has recurring tasks that nobody enjoys. They happen on a predictable schedule. They follow roughly the same steps every time. They produce roughly the same output. And they make someone's eyes glaze over. That's your candidate.

Check three criteria. Is it repetitive? Does it follow clear rules? Is the data already digital, or could it easily be? If you get three yeses, you've probably found a good starting point.

Good first candidates tend to include invoice processing, data entry and reconciliation, report generation from multiple sources, email triage and routing, appointment scheduling, and compliance checking against known regulations.

Bad first candidates include anything described as "strategic decision-making," anything called "customer experience transformation," and anything that requires nuanced human judgment about ambiguous edge cases. Those projects aren't impossible. They're just terrible places to start.

The difference is simple. Good first candidates have a clear definition of "done." Bad first candidates have a definition of "done" that changes depending on who you ask and what day of the week it is.

The quick win cascade

Here's what happens when you get that first boring project right.

The person whose Tuesday afternoons just got freed up starts talking. Not because you asked them to. Because they're genuinely relieved. They tell their team. Their team tells other teams. Someone from another department comes over and says, "I heard you automated the reconciliation thing. We've got this process where we have to..."

And suddenly you've got demand.

This is how AI adoption actually works in practice. Not through top-down mandates and transformation programmes. Not through strategy documents and steering committees. Through one person, in one team, having a measurably better week.

That person becomes your internal champion. They didn't sign up for it. They don't have "AI Champion" in their job title. They just had a problem that got solved, and now they're living proof that this stuff actually works.

Confidence builds. Budget follows. The second project is easier than the first, because you've already proven the concept, you understand your data better, and you've got someone in the organisation who will vouch for the approach.

After the third project, something shifts. People stop waiting for you to suggest things. They start bringing you their problems. "Could AI do this?" becomes a question you hear in the kitchen, not just in the boardroom.

This is the quick win cascade, and it is the only reliable way I've seen AI adoption scale in small and mid-sized organisations. The companies that try to skip straight to enterprise-wide transformation almost always stall. The companies that start small and let momentum build almost always succeed.

Scaling from one to many

Once you've got three or four successful implementations under your belt, the game changes.

You start to see patterns across processes. The same data quality issues come up repeatedly, which means you fix them once and multiple projects benefit. You build internal knowledge about what works and what doesn't. You develop a sense for which processes are ripe for automation and which ones need more preparation.

This is also when you can start thinking about more sophisticated approaches. Agentic AI — systems that can handle multi-step processes autonomously, making decisions and taking actions across different tools — becomes realistic once you've already mapped your processes, cleaned your data, and built organisational confidence through smaller wins.

But you earn the right to do the ambitious stuff by proving you can do the boring stuff first. Not the other way around.

What not to do

A few things to avoid, learned the hard way.

Don't start with a "comprehensive AI strategy." Start with a project. A strategy without evidence is just a PowerPoint deck. Deliver something first. The strategy will emerge from what you learn.

Don't buy a platform before you've proved a use case. I've seen organisations spend six figures on AI platforms before they've successfully automated a single process. That's like buying a Formula 1 car before you've passed your driving test.

Don't skip governance, even for small projects. Yes, even for the boring invoice processing automation. Who's responsible when it makes a mistake? How do you check its work? What data is it accessing? These questions are easier to answer for small, low-risk projects — that's part of why you start there. The EU AI Act is already shaping how organisations need to think about this, and getting governance right early is infinitely easier than retrofitting it later.

Don't measure success by technology sophistication. Nobody cares what model you're running or how clever your prompt engineering is. Measure time saved. Measure errors reduced. Measure the number of people who got their Friday afternoons back. Those are the metrics that unlock budget for the next project.

Don't forget the humans. The person whose job just changed needs support, not just a new tool. Automation anxiety is real. If someone's been doing the Tuesday reconciliation for three years and you suddenly automate it, they need to know what their role looks like now. The answer should be "better" — more interesting work, more time for things that actually require human judgment — but you need to have that conversation explicitly. Not after the fact. Before.

The unremarkable beginning

Every successful AI implementation I've been part of started with something boring. Something nobody would write a case study about. Something that, from the outside, looks profoundly unimpressive.

An invoice processing workflow that used to take three days now takes three hours. A weekly report that used to require someone to manually pull data from four systems now generates itself. An email triage process that used to eat two hours every morning now routes 80% of incoming messages automatically.

None of these made headlines. None of them were "transformative" in the way that word gets used at conferences. But each one freed up a real person to do more interesting work. Each one proved to a sceptical organisation that AI isn't just hype. And each one created the momentum for the next project, and the one after that.

The best AI implementation in your business won't be the one that impresses your board. It'll be the one that makes someone's work life slightly less miserable on a Tuesday afternoon.

And from that unremarkable beginning, everything else follows.

The grand AI strategy that everyone talks about? It doesn't come from a consultant's slide deck or a leadership offsite. It comes from a string of small, boring, undeniable wins that build on each other until one day you look up and realise your organisation has fundamentally changed how it works.

But it started with the process everyone hated.

It always starts there.