Start with one workflow, not a roadmap
The fastest way to prove AI works in your business is to pick one workflow, automate it properly, and let the results speak.
There’s a pattern we see with businesses that stall on AI. They start with strategy. They commission a roadmap. They map out twelve months of initiatives across six departments. Then momentum stalls.
The roadmap can become a document that sits in a shared drive. Too big to start, too abstract to act on. Meanwhile, the team goes back to doing everything manually.
Why roadmaps stall
A roadmap tries to solve the whole problem at once. It identifies every possible use case, prioritises them, assigns timelines, and estimates ROI. It feels productive because it’s thorough.
But thoroughness is the enemy of momentum. When everything is planned, it can be hard to know where to begin. When the scope is twelve months, the start date tends to be “next quarter.”
And often the biggest challenge: no one in the business has seen AI work yet. They’re being asked to commit to a programme based on potential, not proof.
One workflow can change the conversation
Pick one workflow. Not the most complex one. Not the most important one. Pick the one where the win will be visible and the risk is low.
Something that runs frequently, takes a predictable amount of time, and produces an output people actually use.
Build a system that handles it. Not a prototype. Not a demo. A real system that runs in production, with real inputs and real outputs. Something the team uses every day.
When that system is working, the conversation often shifts from “could AI work here?” to “where else can we do this?”
How to pick the right workflow
Four criteria:
Frequency. The workflow should run at least weekly. Daily is better. The more often it runs, the faster the value compounds and the faster people get used to it.
Visibility. The output should be something people see and use. A report that goes to the leadership team. A document that goes to clients. Something where the improvement is obvious.
Pain. The current process should be genuinely tedious. If people don’t mind doing it manually, the motivation to adopt the new system is weak. Pick something people complain about.
Containment. The workflow should be self-contained enough that you can automate it without touching five other systems. Fewer dependencies means faster delivery and fewer things that can go wrong.
What “properly” means
This isn’t a hack or a workaround. Building one workflow properly means:
Defined inputs. The system knows exactly what data it needs and where to get it. (For more on what this architecture looks like, see what an AI system actually looks like.)
Structured processing. The AI steps are designed with clear prompts, validation, and error handling.
Useful output. The result goes somewhere useful in a format people can act on.
Human review. Someone checks the output before it goes further, at least initially.
Iteration path. There’s a way to improve the system based on feedback without rebuilding it.
The snowball effect
One working system creates demand for the next one. The team sees the time saved. They experience better output. They start asking “can we do this for X too?”
That’s when the roadmap writes itself. Not from a strategy document, but from people who have seen what’s possible and want more of it.
Start small. Start real. Start with one. And once it’s running, measure the value it creates so the next conversation is backed by evidence.