Key takeaways
- Your most critical processes could be the worst place to start with agents.
- Culture decides whether an agent gets used before the technology does.
- Four questions separate the agent ideas worth backing.
Everyone is talking about AI agents, but with many pilots failing to progress, you’re unsure which idea to prioritise.
The pressure to do something with agents is real, so it’s tempting to focus on your most critical processes. But these are often the hardest to change, so months later, your project may have consumed budget yet changed little.
The ones that land best take repetitive tasks off teams and free skilled people to do the work that needs their judgement. The challenge is less about building and more about knowing which jobs to hand over.
But before you weigh up ideas, it helps to understand why some stall.
Why your culture decides more than technology
The reason an agent succeeds or fails often has less to do with the agent itself. It comes down to whether the people are ready to work differently.
Most teams are further ahead than their organisations recognise. People can see where an agent could help and are open to handing over the dull parts of their day. What holds them back is everything around them: the targets that still reward the old way of working and the managers who want familiar reports in the same format.
Drop an agent into that environment, and it will remain unused, or add further administration on top of the work it was meant to remove.
The blocker is rarely the technology but the conditions around it. That’s worth facing because the best idea cannot outrun a culture that isn’t ready for it.
Where do agents fit in your business?
This is not about having too few ideas. Most people can reel off a dozen things an agent might do. The challenge is filtering the genuinely useful candidates from those that seem good but don’t progress, and knowing how to judge between them before committing time and resources.
Agent pilots stall for predictable reasons, and we have written about what separates the ones that succeed. The patterns are consistent. Pilots that progress start from a real problem and a clear measure of success. The ones that fade lack clarity and tend to lead with the technology, hoping that value will emerge later.
The task is to find the best ideas and have a way to weigh them.
Start with the work people prefer to skip
The strongest candidates are often the tasks few people volunteer for.
Think about those recurring jobs that pull skilled people away from the work only they can do. Copying figures between systems, chasing approvals and manual data entry are persistent examples.
This work matters, but repetitive manual tasks like these rarely require the skills of the people currently performing them.
We saw this with a client whose team were spending dozens of hours a week keying hundreds of emailed orders into Dynamics 365. Each one involved a PDF or image that someone had to open and retype. We built a process that reads the attachments, extracts order details across several languages, and creates the records for a person to check rather than type. Boring data entry was removed without losing human oversight, allowing the business to recover around 3,000 hours and £40,000 in capacity over 12 months.
Adoption is the second prize. When an agent takes an unloved task off someone’s plate and hands time back to do more of what they enjoy, they’re more likely to become an advocate. Choose the right job, and the people involved might help you make the case for the next agent.
Decide the goal before you build
Before you commit to any idea, settle one thing: how will you know it worked?
When an agent launches without a metric attached, it may struggle to prove it made any difference.
Decide the measure up front. How long does the task take today? How often does it happen? What’s the error rate? How many hours does it consume, and what does it cost?
Capture that starting point before the agent goes live, so subsequent improvements are something you can take to a CFO or the board.
A clear baseline keeps everyone honest. It reveals when an idea is working, but just as helpful, when it is not.
Agent questions that separate the best ideas
Once you have a few options, you need a consistent way to compare them. Here are four questions for starters.
How often does it run? An agent that handles a task that runs several times every day will always return value faster than one built for infrequent events.
How close does it sit to money or risk? The nearer a process is to revenue, your customers, or something that carries real consequence if it goes wrong, the easier it is to justify the effort. That’s where critical processes demand attention, but always alongside the other three considerations.
Can it use your existing data? An agent that runs on information you already hold, which is in reasonable shape, can move now. Data that needs major clean-up or consolidation first is a separate project.
Can it prove itself with a number? If you can’t describe what better looks like in terms you already track, the value will always be arguable. The ideas worth backing are the ones where success is plain for everyone to see.
Other tests will matter depending on your situation, from the appetite of the team involved to the systems an agent would touch. But these four help sort the serious contenders from the concepts that only impress in the room.
The readiness test underneath it all
One last test sits beneath every idea.
A seemingly transformational agent built on shaky data or an undocumented process is a risk dressed up as a quick win. They follow the process they are given, gaps and all, and amplify whatever they find.
If the underlying work is unclear or if data is missing, inaccurate or incomplete, agents expose these cracks, and people stop trusting what they produce.
This is the difference between an appealing idea on the surface and a ready one. Working out which is which is the heart of our AI Readiness Assessment to prioritise the agents that will deliver measurable value.
Which idea first?
Start with the tasks people would gladly give up. Prioritise the jobs that happen often, sit close to something that matters, run on data you can trust, and show their value in a number. Then be honest about whether the process underneath is ready for an agent.
Get that choice right, and the result will do more than save a few hours. It earns the trust that makes the next one easier.
Ask yourself: which job taken off your team’s plate would free the most hours?
