The fastest way to waste money on AI is to start with AI. Start with the work instead.
There is a version of AI adoption that looks impressive from a distance and achieves very little up close.
A business buys a new platform. A team launches a pilot. A dashboard appears. Someone announces an “AI strategy”. For a few months, there is energy around the initiative. Then the novelty fades and the underlying work is still slow, fragmented and dependent on the same people moving information between the same systems.
That is not transformation. It is decoration.
The useful starting point is not “Where can we use AI?” It is “Where is the organisation wasting time, losing information, repeating decisions, creating avoidable risk or asking capable people to perform work a system should handle?”
Once the problem is understood, AI may be part of the answer. Sometimes it will not be.
Start with the operating problem
The strongest automation opportunities are normally visible before anyone talks about models.
A team is copying information from email into a CRM. A manager is rebuilding the same report every week. A recruiter is searching five systems to understand one candidate. A finance team is reconciling information that should already agree. A customer waits because an internal hand-off depends on someone remembering to do it.
These are operating problems.
If AI can remove the friction, good. If a workflow rule, better data structure or simpler process solves it more reliably, use that instead.
Technology should earn its place.
Measure capacity, not novelty
I care much more about capacity created than features deployed.
Did the change reduce manual handling? Did it shorten a cycle? Did it improve response time? Did it give a senior person back several hours each week? Did it improve consistency? Did it make a decision easier to audit? Did it allow the same team to handle more work without becoming more stressed?
Those are meaningful outcomes.
A useful AI system should leave the organisation more capable than it was before. It should not merely create another layer that needs monitoring, prompting and explaining.
Keep humans where judgement matters
Automation is most valuable when it removes work that does not deserve human attention.
That does not mean removing people from important decisions. It means protecting their attention for decisions that actually require context, judgement, empathy, commercial instinct or accountability.
The goal is not maximum automation.
The goal is a better division of labour between people and systems.
A good system can gather information, structure it, identify patterns, draft options, enforce routine controls and surface exceptions. A person can then spend more time deciding what matters.
That is a much stronger operating model than asking people to behave like slow databases.
Design for the messy organisation you actually have
Most businesses do not operate inside clean diagrams.
They have old systems, inconsistent data, exceptions, informal workarounds, customers who do not follow the process and teams with different levels of technical confidence.
That reality matters.
An AI workflow that works only when every input is perfect is not an operating system. It is a demonstration.
Real systems need fallbacks, permissions, logging, human review, error handling and clear ownership. They need to fit the organisation that exists while helping it become better.
That is less exciting than a demo. It is also where most of the value lives.
Build trust through usefulness
People adopt systems that make their work easier.
They resist systems that create more administration, obscure accountability or make them feel monitored without helping them.
The quickest route to trust is usefulness.
Solve something irritating. Remove a repetitive task. Make an answer easier to find. Reduce duplicated work. Give a team a tool that makes Monday morning noticeably better.
Then build from there.
AI does not need to be everywhere.
It needs to be in the places where it makes the business work better.
That is the standard I think organisations should use: not whether the business can say it is using AI, but whether the business is measurably more capable because of it.