Most modernisation projects don't fail dramatically - they just quietly underdeliver, and everyone moves on slightly disappointed. The technology works. The business case was solid. But nothing actually changed, because you automated the past instead of building the future.
What if the process shouldn't exist at all?
Everyone asks "how do we automate this process?" Wrong question.
The right question is "why does this process exist at all?"
Most legacy workflows weren't designed - they accumulated. Someone needed a workaround years ago, it became a process, and now three people spend their week feeding it. That workaround probably made sense at the time. The tools were different, the constraints were real, and it solved an immediate problem. But when was the last time anyone checked whether that context still holds?
AI doesn't just let you automate that process. It lets you kill it. But that requires stepping back and asking whether the problem you're solving should exist in the first place - and that's hard to do when you're deep in the logic of how things are.
Those three people spending their week on that process? They're not the problem. They're the opportunity. AI shouldn't just make their busywork faster - it should eliminate the busywork entirely and free them to do something that actually matters. If you were starting from zero today, would you build any of this, or would you give those people their week back?
First answers usually point to symptoms
Ask someone what's broken and they'll tell you. They'll be confident. They'll be specific. And they'll usually be describing a symptom rather than the cause.
We've seen this pattern repeatedly, and it almost always plays out the same way. The first answer is the frustration, the thing that's top of mind. The real problem is underneath, and it usually takes a few honest conversations to find it.
First conversation: they tell you what they think is wrong. Second conversation: you ask why that thing exists, and the story gets more complicated. Third conversation: you finally get to the constraint, the decision, the broken assumption that's actually causing the pain.
This is where most projects go wrong. Not in the build - in the definition. Skip this step and the result is a beautiful solution to the wrong problem. On time, on budget, and unused.
The people closest to the problem usually know what's actually broken. They've just never been asked the right questions, or given permission to answer honestly. That third conversation isn't just diagnostic - it's where you find out what your team has been working around for years, and what they could be doing instead if you got out of their way.
Requirements documents are theatre. Data and outcomes are real.
Traditional projects start with months of requirements gathering. Workshops, stakeholder interviews, specification documents that nobody reads. By the time you're done, half of it's already out of date.
AI doesn't work that way. You need exactly two things:
The data: What do you actually have? Not what you wish you had, not what's in the data dictionary - what's real, accessible, and clean enough to use?
The outcome: What decision or action should be different at the end? Not "insights" or "visibility" - what specific thing changes?
Data + outcome = requirements.
Napkin. Whiteboard. Doesn't matter. If you can't explain it simply, the problem isn't the format - it's the thinking.
This is why AI projects can move in weeks instead of quarters - if you let them. And that speed matters, not for its own sake, but because every month spent documenting the old way of working is a month your team stays stuck in it.
The real problem isn't the technology.
Most organisations are using AI at the surface. Write this email. Summarise this document. It's useful, but it doesn't change anything. It doesn't require anyone to step back and think about what they're actually trying to achieve.
When you really sit with a problem - when you take the time to understand the constraints, the workarounds, the accumulated logic of years of "that's just how we do it" - that's when AI stops being a tool and starts changing how you work. Not because it's magic, but because you've finally given it a real problem to solve.
But most businesses don't empower their people to think this way. There's no incentive to pause. No reward for questioning the process. Everyone's measured on output, so everyone keeps optimising the thing that might not need to exist.
Sometimes you have to go backwards to go forwards. It's like rugby - you can only pass behind you, but that's how you create the space to break through. The organisations that get the most from AI aren't the ones rushing to automate everything. They're the ones willing to step back, look at what they've built, and ask what their people should actually be doing with their time.
AI gives you license to engage creatively with your problems - to solve them differently, and free your people from processes that no longer need to exist.
Your people already know what's broken. They've been working around it for years. The question is whether they'll get permission to fix it properly - or whether the organisation will keep automating the old way and wondering why nothing feels different.
This is part of how we think at PolusAI. If this resonates, let's talk.