If your team is feeling the pressure to “do something with AI”, start by finding the workflow friction that is already costing you time, money, and customer patience. The right answer might be AI, but it might just as easily be automation, integration, or a cleaner process, and this post will help you tell the difference.

If you have sat in a leadership meeting lately, you have probably heard some version of it: we need to do something with AI.

That pressure is real. Customers are hearing about it. Competitors are talking about it. Teams are being asked where the opportunities are, what the roadmap is, and whether they are already behind.

The awkward bit is that “do something with AI” is not actually a business problem. Slow turnaround is a business problem. Re-keying the same information three times is a business problem. Customers waiting for an answer because the truth is trapped across inboxes, spreadsheets, and separate systems is a business problem.

AI may help with some of those things. It may not be the first thing you need.

That is where a lot of businesses get themselves tangled up. The conversation starts at the solution end. People go looking for a copilot, an agent, a summariser, or a clever way to automate a screen, before they have worked out where the real drag sits in the workflow.

So money gets spent, something smart gets added, and yet the team still feels busy because the real bottleneck is still there.

The better starting question is less exciting, but far more useful: where is the friction that is already costing us time, money, and customer patience?

The problem is usually not an AI shortage

Most operational pain is not caused by a lack of intelligence. It is caused by work getting stuck, repeated, scattered, or quietly improvised.

Sometimes that shows up as repetitive manual work. Someone is copying details from an email into a CRM, then into a spreadsheet, then into a finance package, because no single step has been joined up properly. AI can sometimes help read the email faster, but it does not magically remove the duplication.

Sometimes the drag is in handoffs and status chasing. One person finishes their bit, another person needs to pick it up, and the whole thing slows down because the live state is not visible anywhere sensible. People start sending “just checking” emails. Customers do the same. Now the process is doing extra work simply to explain where the work is.

Sometimes the issue is re-keying and duplicated data. The information already exists, but not where it is needed. Staff become the integration layer. They copy, paste, compare, and correct, and the business pays for that in lost time and avoidable errors.

And sometimes the real workflow decisions are happening in inboxes and spreadsheets rather than inside a governed process. The important judgement calls are still being made, but the logic, ownership, and history are informal. That is when work starts to depend on memory, heroics, and “the way we usually do it”.

If any of that sounds familiar, the important point is this: you do not have an AI problem first. You have a friction problem first.

That is good news, because friction is easier to reason about than hype. You can observe it. You can describe it. You can score it. And once you can do that, it becomes much easier to decide whether the right answer is AI, automation, integration, or a simple redesign of the workflow itself.

What the wrong starting point looks like

When businesses start with the technology question, they tend to chase the most visible idea rather than the most expensive bottleneck.

The inbox gets an AI summariser, but nobody fixes the fact that three teams still need to chase each other for status afterwards. The team experiments with an agent, but the real cost is still that key details are being entered manually into multiple systems because nothing integrates cleanly. A manager asks for AI-driven insights, but the operation still has no reliable view of where work is live, blocked, waiting for approval, or quietly sitting in someone’s head.

That is why “not everything needs AI” is such a useful principle. It is not anti-AI. It is simply a reminder to match the tool to the actual drag. If the pain is mostly interpretation of messy text, AI might be a strong fit. If the pain is that the same data is being copied around by hand, integration or rules-based automation may do more for less. If the pain is that nobody can see or govern the flow, you probably need a clearer process before you need anything clever on top.

Two ordinary patterns that get mistaken for AI opportunities

The shared inbox that looks like an intake problem

At first glance, this one often looks perfect for AI. A shared inbox is full of incoming requests, and the team wants faster triage. So the instinct is to ask for classification, summaries, and suggested replies.

Those things can help, but they are rarely the whole answer. The bigger issue is often what happens after the email has been read. If ownership moves by forwarding messages around, if status sits in a spreadsheet, and if supporting information has to be hunted down across separate systems, the real bottleneck is not the reading. It is the handoff.

Before, the inbox feels busy and urgent because it is acting as a makeshift workflow tool. After, the workflow has its own visible state, ownership is clear, and AI can assist with intake inside that governed flow instead of trying to carry the whole operation on its back.

The repetitive admin task that looks like a thinking problem

This one usually starts with a team saying they are drowning in admin and asking whether AI could take some of it away. Sometimes it can. But often the work is not difficult because it needs intelligence. It is difficult because it is repetitive, split across systems, and badly connected.

If someone is reading a form, copying the same details into three places, checking whether something already exists, and then sending a templated update, the expensive part may not be the interpretation. The expensive part may be that the workflow was never joined up properly in the first place.

Before, a capable person keeps the process moving by stitching together disconnected steps. After, the repeatable parts are automated, the systems share the right context, and AI is used only where messy inputs or nuanced judgement genuinely make it worthwhile.

Find the hotspot before you shop for the answer

Once you stop asking “where can we use AI?” and start asking “where is the operational drag?”, the next step is not complicated.

Pick a workflow that people complain about, or one that customers feel directly. Then score the pain points inside it. A simple 1-5 score across frequency, effort, and business impact is often enough to make the next decision much clearer.

Frequency tells you how often the problem happens. Effort tells you how much time, concentration, or skilled attention it burns each time. Business impact tells you what happens when it goes wrong, slows down, or stays invisible for too long.

The hotspot with the highest score is not always the noisiest one. It is often the boring thing that happens all day, every day, quietly draining hours and frustrating people.

This is also where a lot of AI decisions get calmer. When you can point to one hotspot and describe the drag clearly, the solution space narrows very quickly. Repetitive rules-based work points towards automation. Scattered data points towards integration. Invisible ownership points towards process redesign. Messy inputs, classification, summarisation, and drafting are where AI starts to earn its keep.

AI is one option in the toolbox, not the headline act

This is the part people sometimes find disappointingly unglamorous, but it is what makes AI projects useful instead of expensive experiments.

Good AI decisions usually sit inside a better workflow design. Used well, AI can help with intake triage, summarising a long email thread, pulling structure out of messy documents, suggesting a next step, or drafting something for a human to approve. Used badly, it gets asked to compensate for missing ownership, disconnected systems, vague process rules, or no agreed source of truth.

The practical question is not “can AI do this?” It is “what part of this workflow genuinely needs interpretation, and what part simply needs tidying up?”

That distinction matters commercially as well as operationally. If you use AI where automation would do, you may increase cost without removing much friction. If you use AI where integration is the real answer, you may speed up one step while leaving the rest fragmented. The strongest result is usually a blend: humans own judgement, process handles repeatability, integration carries context, and AI helps where interpretation is genuinely needed.

That kind of exercise gives you a practical way to prioritise and keeps the conversation anchored in business outcomes rather than fashionable tooling.

What good looks like after that

Once teams do this properly, the shift is usually less dramatic than people expect, but much more useful.

They stop talking about AI as though it has to justify itself everywhere. Instead, they start seeing the operation more clearly. They know where the repeated work lives. They know which delays are customer-visible. They know where staff are acting as the glue between systems, and where a process is still depending too much on memory, goodwill, or heroics.

That creates better decisions straight away.

A low-value re-keying problem might be solved with integration before anyone touches AI. A messy intake process might get a cleaner workflow and then add AI-assisted triage on top. A status-chasing problem might turn out to be mostly about visibility, ownership, and a single live source of truth rather than any shortage of intelligence.

This is the bit that often gets missed in the broader AI conversation. The value is rarely in saying “we used AI”. The value is in reducing delay, reducing admin, improving consistency, and freeing skilled people up to focus on judgement rather than manual glue work.

If you can shorten turnaround, reduce avoidable touches, lower error rates, and make live status easier to see, the commercial case becomes much easier to explain. That is true whether the enabling move was AI, automation, integration, better workflow design, or a sensible mix of all four.

A simple rule of thumb for choosing the fix

If you want an even quicker way to sense-check a hotspot, use this.

When the work is predictable and should happen the same way every time, think automation.

When the information exists but is trapped in the wrong place, think integration.

When people are constantly asking who owns this, what stage it is at, or what happens next, think process redesign or workflow orchestration.

When the input is messy, text-heavy, variable, or hard to interpret consistently, think AI.

And when the answer seems to be “all of the above”, that is often a sign that the workflow itself needs to be looked at end-to-end rather than patched at one point of pain.

If you are feeling the AI pressure right now

You do not need to ignore AI. You also do not need to start there.

Start with the friction. Find the tasks that repeat, the handoffs that stall, the data that gets copied, and the decisions that are living in someone’s inbox or spreadsheet. Score them. Prioritise them. Then decide whether the answer is AI, automation, integration, or a cleaner process.

If you want a practical way to do that, download our Operational Hotspot ROI Workbook below. It is designed to help you spot operational hotspots, score their impact, and make better decisions about where to invest first.

Because the best AI decision is rarely “where can we add AI?”.

It is usually “where is the friction, and what is the right fix for that?”