Stop Guessing Where AI Can Help

Why this matters. The most useful AI idea may be hiding in a task you do every week. You are more likely to find it by examining the work than by trying another tool.

The problem with starting from a tool

Perhaps you have asked an AI assistant to draft an email, summarise a document or suggest social media ideas. Those can be useful tasks. But they do not tell you whether you are applying AI where it matters most.

A small business owner might spend an hour preparing each customer proposal, while also using AI to polish a few sentences on their website. The writing help is pleasant. The proposal process might be a bigger opportunity because it happens repeatedly, involves several steps and takes time away from customers.

The question is not, "What else can this tool do?" It is, "Where does my work become slow, repetitive, inconsistent or difficult to manage?" That shift is the foundation of this course.

You do not need to understand automation software or write code to begin. You need a clear picture of your work, an honest account of the problems and a way to compare possible improvements.

Think like an assessor

An AI opportunity assessment is a structured review of the way work gets done. It identifies processes that might benefit from AI, then considers whether an improvement is realistic and worthwhile.

The source example behind this course described an assessment process that began with conversations about operational problems. The assessors asked what work people wanted off their plates, what software they used and where tasks became cumbersome. They then turned those answers into a ranked list of opportunities and demonstrated possible solutions using examples of real work.

You can use the same underlying approach without interviewing a large team. If you work alone, you are both the person doing the work and the person describing the problem. If you manage a small team, ask colleagues where time disappears and what requires repeated checking.

A useful assessment does four things:

1. It describes what happens now.
2. It identifies why that matters.
3. It proposes possible changes.
4. It tests whether the changes actually help.

Notice what is missing: a promise that AI can solve everything. Some problems are caused by unclear responsibilities, missing information or unnecessary approval steps. AI may not be the right answer. Finding that out is still a good result.

Start with outcomes people care about

A business rarely needs AI for its own sake. It needs to serve customers, deliver work, control costs and keep promises. An individual may simply want fewer repetitive chores and more time for meaningful work.

Choose an outcome before choosing a technology. Common outcomes include:

• Time: Spend fewer minutes on a recurring task.
• Consistency: Produce work that follows the same agreed steps each time.
• Capacity: Handle more enquiries or projects without creating chaos.
• Accuracy: Reduce avoidable omissions, while keeping appropriate human checks.
• Experience: Make it easier for customers or colleagues to get what they need.

For example, a consultant who manually turns meeting notes into follow-up tasks might want a reliable first draft of those tasks. The outcome is not "use an AI note-taking system". It is "reduce the time between a meeting and a clear follow-up, without missing commitments".

A coach may repeatedly prepare session summaries. A retailer may copy order details between systems. A tradesperson may turn inspection notes into quotations. Each example starts with a workflow rather than a product.

Find the friction in ordinary work

Think about the last seven working days. What did you repeat? What did you postpone? Where did you switch between several documents or systems? Which task required you to look up the same information again and again?

You are looking for friction, not necessarily a dramatic crisis. A task that takes ten minutes but happens many times a week may deserve attention. A difficult task that happens once a year may not.

Try this short exercise before reading further. Write down three tasks that match at least one of these descriptions: repetitive, slow, easy to forget, dependent on copying information, or difficult to delegate. Do not worry yet about whether AI can handle them.

In the workbook, use the section Stop Guessing Where AI Can Help to capture your three candidates and the one that bothers you most.

Now ask your AI assistant to help you examine the list. Remove customer names, personal details and confidential information first.

I want to identify useful AI opportunities in my work, not buy tools for their own sake. My role or business is [YOUR ROLE OR BUSINESS TYPE]. These recurring tasks cause friction: [TASK 1], [TASK 2], [TASK 3]. Ask me up to five practical questions, one at a time, to understand their frequency, time required, inputs, outputs and risks. Do not assume AI is the right answer. After my answers, suggest which task I should investigate first and explain why.

The prompt is designed to start a conversation. Its first answer is not a diagnosis. Your own observations provide the evidence.

A practical example: from report to proposal

Consider a business that receives inspection reports and uses them to prepare priced proposals. A person must read the report, identify quantities, consult pricing information, enter figures into a spreadsheet and prepare the final proposal.

This is a useful assessment candidate because the steps are repeated, the inputs and outputs are recognisable, and there is a clear way to compare current and proposed approaches.

In the source material, a similar workflow was used to demonstrate how AI might extract information from a report and assist with pricing. The demonstration was persuasive because it used examples of the work itself rather than a generic promise about automation.

But a demonstration is not the same as a dependable system. Reading a diagram incorrectly, missing a pricing exception or using an outdated rate could create an expensive mistake. Any proposed improvement would need testing against representative examples and human approval before customer-facing use.

For your own work, the equivalent may be much smaller. You might turn enquiry notes into a draft response, or take a set of anonymised meeting notes and produce a checklist. The principle is the same: start with a specific input, a defined process and an output you can inspect.

Separate opportunity from evidence

It is tempting to claim that AI will save a certain number of hours before you have tested anything. Resist that temptation.

At the assessment stage, there are three different kinds of statement:

Observed: "Preparing one proposal currently takes about an hour." Even this should be checked against more than one example if possible.

Estimated: "A redesigned process might take fifteen minutes." That is a hypothesis until you test it.

Verified: "Across five comparable trials, the new process took an average of fifteen minutes including human checking." That is stronger evidence, although it still may not predict every future case.

The original analysis highlighted an important caution: a faster proposal process does not automatically mean more sales. A business may have limited demand, a low conversion rate or insufficient capacity to deliver extra jobs. We will return to this distinction when we estimate value in Chapter 6.

For now, use plain labels such as known, estimated and not yet tested beside your claims. That habit makes your assessment more credible and your decisions easier.

What makes an opportunity worth exploring?

Not every repetitive task should be automated. Before you choose a candidate, ask five questions.

Does it happen often enough? Frequency influences the value of time savings. A monthly task can still matter, but the benefit must justify the effort.

Can you describe the steps? If nobody can explain the current process, mapping it may be the first improvement.

Are the inputs available and appropriate to use? A workflow that depends on private customer records needs stricter controls than one using public information or invented examples.

Can a person check the result? For an early experiment, prefer work where mistakes are visible and reversible.

Would an improvement matter? Saving five minutes on an unimportant task may be less valuable than reducing errors in a critical handover.

These questions do not produce a final ranking. They help you avoid starting with an idea that is exciting but poorly defined.

Your first ten-minute assessment

Choose one of your three tasks. Describe it in five short sentences:

1. The task starts when something happens.
2. I receive or look at these inputs.
3. I currently follow these steps.
4. The finished output is used for this purpose.
5. The frustrating part is this.

For example: "The task starts when a prospective client sends an enquiry. I read the message and check my service notes. I draft a reply, check availability and record the next action. The output is a helpful response and a clear follow-up. The frustrating part is repeating the same checks and sometimes forgetting to record the follow-up."

That description is far more useful than "I want an AI agent for my business". It gives you something specific to improve and measure.

Record your five sentences in the workbook. You will return to them when mapping the workflow in Chapter 4.

What you should take into Chapter 2

You do not need a perfect assessment today. You need one well-described problem and the willingness to examine it before choosing a solution.

Your first shortlist should contain three possible tasks. For one of them, you should know what triggers it, what goes into it, what comes out and why it is frustrating. You should also be clear about which statements are observations and which are guesses.

In the next chapter, we will turn that initial description into a practical improvement objective. This matters because a clear objective keeps your AI experiments connected to something useful rather than becoming a collection of disconnected tricks.

If you are unsure how ready your wider business is for AI, the AI Readiness Check is the next place to begin.