I've become increasingly interested in AI from an operational perspective, and not because I think every business needs to become an "AI-first" company. I'm much more interested in the less exciting question: where does AI actually make the work better?
There is a lot of enthusiasm around using AI to automate things that businesses currently do manually. Some of that is genuinely useful. But I've also seen enough operational problems to be cautious about automating something simply because it can be automated. If the underlying process is unclear, AI doesn't necessarily solve the problem. It can just make the wrong process faster.
That distinction matters because most businesses don't have a shortage of technology. They have a shortage of clarity.
They may have information spread across different systems, processes that exist mostly in people's heads, unclear ownership, duplicated work and teams spending time on tasks that nobody has stopped to question. Adding AI to that environment can certainly produce impressive results in the short term, but it can also make it harder to see what the underlying problem actually was.
Start with the work, not the AI
I think the most useful way to introduce AI into a business is to start with the work that is already happening.
What takes too long? What gets repeated constantly? Where are people copying information from one place to another? What requires someone to read through a large amount of information before making a relatively straightforward decision? Where are people spending time producing something that follows a fairly predictable pattern?
Those are much more useful questions than asking, "Where can we use AI?"
The difference might sound small, but it changes the conversation completely.
If someone on a team spends several hours each week summarising customer feedback, there may be a genuine opportunity to use AI to reduce the manual work. If a project manager spends time turning meeting notes into action items, that's another obvious place to experiment. If a support team repeatedly has to find information across several documents before answering a common question, there may be an opportunity to improve both the knowledge system and the way AI accesses it.
The technology becomes secondary.
The work comes first.
Automation doesn't fix unclear processes
One of the things that makes me cautious about automation is that businesses sometimes automate before they understand what they're automating.
Imagine a process that currently involves five steps, three people and a lot of back-and-forth. It is slow, but everyone has developed their own way of making it work.
An AI tool might allow you to automate part of that process immediately.
That sounds like progress.
But if nobody has asked why the five steps exist in the first place, you may simply have automated one part of a process that shouldn't have been designed that way.
I've seen similar problems without AI. A business introduces a new project management platform because the existing system feels messy, but the same unclear ownership and inconsistent processes simply appear in the new platform. The company now has a newer tool and the same operational problem.
AI doesn't change that principle.
Before automating something, I want to understand what the process is trying to achieve, who owns it, what information it depends on and where it tends to break.
Then we can decide whether AI actually improves it.
Some of the best opportunities are surprisingly boring
The AI applications I find most interesting aren't necessarily the ones that make the best demos.
They're often small improvements to work that people currently have to repeat over and over again.
A support team could use AI to help categorise incoming enquiries. A project manager could use it to turn a long meeting transcript into a useful summary and action list. An operations team could use it to help identify recurring themes in customer feedback. A company could make a large internal knowledge base easier for employees to search and understand.
None of those examples sounds particularly revolutionary.
They can still save a significant amount of time.
That matters because operational improvements compound. If something saves one person twenty minutes a day, it might not feel significant. If it saves twenty people twenty minutes a day, the picture changes. And if it also reduces the number of errors or interruptions involved in the task, the benefit becomes larger again.
That's where I think AI is particularly interesting for operations.
It doesn't always need to replace a person to be valuable.
Sometimes it just needs to give that person some of their time back.
Customer support is an obvious place to see both the opportunity and the risk
I've spent a lot of time working in customer support operations, so this is an area where I think the distinction becomes particularly clear.
There are plenty of repetitive support tasks that AI can potentially help with. Categorising tickets, identifying common questions, summarising customer histories, suggesting responses and helping agents find relevant information are all areas where reducing manual work can be useful.
But the existence of a repetitive task doesn't automatically mean it should be fully automated.
Customer interactions can contain context that isn't obvious from the wording of the message. A customer may have contacted support several times. A previous response may have created confusion. A refund request might be technically straightforward but part of a much larger customer experience problem.
An AI system can help an agent process information faster. That doesn't mean it should automatically make every decision.
For me, that's an important distinction.
The goal should be to give people better information and reduce unnecessary work while keeping appropriate human judgement where it matters.
AI also exposes how good your information is
There's another side of AI that I think businesses sometimes underestimate.
AI is only as useful as the information and context it has access to.
If your documentation is outdated, inconsistent or scattered across five different places, an AI system doesn't magically turn that into a reliable knowledge base.
In some ways, AI makes the underlying quality of your information more important.
This is actually a useful thing.
If a business starts trying to build an internal AI assistant and discovers that nobody can agree on which process is current, that's not really an AI problem. It's an operational problem that the AI project has exposed.
The same thing happens when teams have different versions of the same information or when nobody knows who owns a particular process.
Before you can expect technology to make information easier to use, you need to know what the information actually is.
That is one reason I think operations and AI are going to become increasingly connected. The businesses that get the most value from AI won't necessarily be the ones with the most impressive tools. They'll often be the ones that understand their own work well enough to know where those tools can genuinely help.
People still need to own the outcome
There is a tendency in some conversations about AI to talk about automation as though the ultimate goal is to remove people from the process completely.
I'm not convinced that's always the right objective.
Businesses still need people who understand the context behind decisions, know what the customer actually needs, can spot when something doesn't make sense and are willing to take responsibility for the outcome.
AI can make someone much faster without making them less accountable.
In fact, I think that is a more useful way to think about it.
If AI can help an operations manager analyse information in minutes rather than hours, that manager now has more time to think. If it can help a support agent find the right information quickly, the agent can spend more of the interaction actually helping the customer. If it can reduce the amount of administrative work a project manager has to do, they can spend more time managing the project.
The human role changes.
It doesn't necessarily disappear.
The businesses that benefit most will probably be the ones that experiment sensibly
I don't think anyone needs to have their entire AI strategy figured out before they start using AI.
Technology is changing too quickly for that to be realistic.
I'd rather see a business identify a small, clearly defined problem, try something, measure whether it actually helped and then decide what to do next.
Did it save time? Did the quality improve? Did it introduce new errors? Did people actually use it? Did it make the process easier or did it simply move the work somewhere else?
Those questions are much more useful than whether the AI tool is impressive.
I've always liked working in environments where you can test something, learn from it and adjust rather than spending months designing the perfect solution before anyone uses it. That was part of the appeal of the sprint-based approach we used at Melewi, and I think the same principle applies to AI.
Experiment, but give the experiment a purpose.
AI should create capacity, not just more output
For me, this is ultimately the test.
If AI allows a business to produce more reports, more emails, more content and more tasks without improving the underlying operation, I'm not sure how much has actually been gained.
More output isn't necessarily more value.
The more interesting possibility is that AI gives people capacity to do the things that require judgement, relationships, creativity and attention.
That's the opportunity I see.
Not replacing every person. Not putting AI into every workflow because everyone else is doing it. Not building complicated automation just so the business can say it uses AI.
Instead, looking carefully at how the business works and asking where technology can remove friction, improve access to information and give good people more time to do work that actually requires them.
If the business is already operating well, AI can make some parts of it considerably better.
If the business is operationally messy, I'd fix that first.
Because AI can make a good process faster. It can also make a bad process faster.
And the second one is usually much more expensive.