In our recruitment business, average time-to-fill has come down from around 60 days to around 35.
People assume that’s a technology story. It isn’t, or not mainly. The improvement came from changing how we manage and reuse information, and from designing a better process around the tools we already had. The system was the easy part. Deciding how people should work with it was the job.
I think about that a lot now that every business is asking how to bring in AI. Because the question I hear least is the one I’d ask first. Should this process exist at all?
Implementation is not the result
I’ve spent my career across systems and processes, finance, commercial operations and business support. One lesson keeps repeating. Technology is rarely the whole answer.
Take a CRM rollout. The usual scorecard says the system is live, the training is done and the spreadsheets have gone. That’s a list of implementation milestones. It says nothing about outcomes.
The outcomes are the questions that come afterwards. Are people spending less time on admin? Is information easier to find? Are decisions quicker? Has service improved? Can managers see what’s happening without asking their teams for another report?
If you can’t answer yes to those, you’ve bought software. You haven’t improved the business.
AI is the same test with higher stakes.
Most processes are accidents
Nobody designs a messy operation. It builds up.
A customer asks for something unusual, so you add a step. A new system arrives, so you keep the old one running alongside it. Something goes wrong once, so an approval gets added. Someone needs a quick fix, builds a spreadsheet, and it quietly becomes the way things are done.
Most businesses will recognise the results. A tracker that duplicates what the CRM already holds. A sign-off that exists because of an incident nobody remembers. A report that goes out every Friday to people who may or may not read it.
Each one was sensible when it started. Together, they leave capable people spending much of their week feeding processes that no longer earn their keep.
Then AI turns up, and the temptation is obvious. Automate the tracker, generate the report, speed up the approval.
You’ll save some time. But if the report doesn’t inform a decision, producing it faster creates nothing. If two teams enter the same data into two systems, AI can move it between them, yet removing the duplication would be far better. Automating a process nobody needs just means you do the wrong thing faster.
Simplify, standardise, then automate
In my experience, the biggest operational gains come from questioning what already exists, not from adding something new. The order I work in is deliberate.
Simplify first. Walk the process from start to finish. Which steps add value? Where do the delays sit? What is recorded twice? What exists only because it always has? Often the best improvement is deleting a step.
Standardise second. Once you know what the process should be, make it consistent. This isn’t about bureaucracy or removing judgement where it’s needed. It’s about one reliable way of working that people understand, can follow and can improve. Skip it, and you’re asking technology to cope with a pile of exceptions, workarounds and personal preferences. It won’t cope well.
Automate third. Only when the process is fit for purpose should you ask where technology can make it faster, more accurate or less dependent on manual effort. And automation doesn’t have to mean AI. A simple workflow, an integration between two systems you already own or a basic rule is often enough. The aim is the best operational outcome, not the highest AI count.
Thirty minutes saved is not thirty minutes gained
Say AI saves someone 30 minutes a day. That’s useful. But the more interesting question is where those 30 minutes go.
If they’re absorbed by the next unnecessary task, nothing has changed. If they go to supporting customers, solving harder problems or winning business, the organisation has genuinely moved forward.
This is why I think the real opportunity is bigger than saving time on individual tasks. For decades, businesses have organised themselves around the limits of their technology. People re-key data because systems don’t talk to each other. Managers spend hours building reports because information is hard to reach. Teams answer the same enquiries repeatedly because there was no reliable way to resolve them automatically.
AI lets us question all of that. But taking advantage means leaders have to revisit responsibilities, decision-making, performance measures and workflows alongside the technology. Handing people a new tool and leaving the old operating model untouched won’t do it.
Activity is not progress
There’s a growing temptation to show that you’re “doing AI”. Tools deployed, licences bought, staff trained. All of that is activity. None of it proves progress.
I’d start somewhere less glamorous, with the problems that are actually hurting performance. Customer response times are too slow. Managers can’t see what their teams are doing. Admin is crowding out higher-value work.
Once the problem is clear, there are three honest options. AI might be the right answer. Existing technology might already solve it. Or the process might simply need redesigning.
And when AI does go in, measure it against something the business cares about, whether that’s turnaround time, error rates, service quality, team capacity, customer satisfaction or cost to deliver. If the measure doesn’t connect to something that matters, you may be paying for technology that looks impressive and changes very little.
This is an operations problem
AI is easy to file under IT. I’d argue that’s a mistake.
Making it work depends on understanding how work moves between departments, where decisions really get made, how information is managed and what people need to do their jobs well. It also depends on people changing habits. The best tool in the world will fail if staff don’t trust it, if managers keep asking for manual workarounds, or if the data underneath is unreliable.
That’s why operational leaders have such a big part to play. They live with the day-to-day reality of running the business. They can tell a process that’s genuinely necessary from one that survives out of habit. They can see how an inefficiency in one team creates work for another. And they can connect changes in how people work to the commercial goals of the organisation.
Start with the problem
I’m optimistic about what AI can do for business operations. Used well, it can remove a huge amount of administrative friction, improve decisions and help organisations scale with more discipline.
But I don’t expect the biggest winners to be the businesses with the most tools. I expect them to be the ones willing to challenge how they operate. They’ll cut what’s unnecessary, deal with the inefficiencies everyone has learned to live with and redesign how work flows.
Because transformation isn’t about making old processes faster. It’s about building a better way of working.
So before you automate your next process, ask whether you need it at all. Sometimes the smartest thing you can do with AI is decide not to use it.
