Finding Your First AI Use Case

"We know we should be doing something with AI, but we do not know what." This is the most common opening line we hear, and it is a more honest starting point than most. The businesses that struggle are not the ones admitting they do not know — they are the ones who picked a project because it sounded impressive. This guide is a practical method for finding a first AI use case inside your own operation, using nothing but a pen, a week of attention, and questions you can already answer.

Why "We Do Not Know Where to Start" Is So Common

The first cause is that the examples available to you come from other people's industries. Conference talks and vendor case studies describe large companies with large data sets, and the gap between that and a fifty-person business in Metro Manila is wide enough that the examples produce admiration rather than ideas.

The second cause is that nobody has mapped the actual work. Most businesses know what their departments are called and roughly what they are responsible for, but not where the hours actually go — which is a different question, and the only one that matters here.

The third cause is that the question being asked has no answerable form. "How should we use AI?" cannot be answered because it has no boundary. Replace it with "which task in this company eats the most hours for the least judgement?" and the same people who were stuck will usually produce three candidates in ten minutes.

Why the Usual Approaches Fail

Copying a competitor fails because you can see their output but not their inputs. You do not know what data they had, what their staff tolerated, or whether the initiative is actually working. Imitation here means adopting the visible half of someone else's decision.

Asking a vendor to identify your use case fails for a structural reason rather than a dishonest one: a vendor's map of possible problems is shaped by the product they sell. Their advice may be entirely sincere and still lead you to the problem their tool happens to fit.

Starting with your most painful problem is the most sympathetic mistake and still a mistake. The biggest pain is usually the most complex, most political, and least well-defined thing in the business — which makes it the worst possible first project. Save it for when you have shipped something.

Related: How AI Consulting Actually Works for Philippine Businesses: The Five Stages From Strategy to Deployment explains this in detail.

A Five-Question Method for Finding the Candidate

The first question is where the hours go. For one week, ask a few staff to note roughly how long tasks take — not precisely, and not as a performance exercise. Almost every business is surprised by at least one item on the resulting list. You are looking for volume, not importance.

The second question is what gets re-typed. Anywhere a person copies information from one screen into another screen is a place where the business is paying a salary to be a bridge between two systems. These tasks are usually invisible because nobody considers them work.

The third question is what waits for one person. If approvals, translations, or checks all queue behind a single individual, that person is a bottleneck — and often the constraint is not their judgement but the drafting they do before the judgement. That drafting is frequently a good AI candidate; the judgement is not.

The fourth question is what gets answered repeatedly. Look at the last month of customer enquiries and count them by type. In most businesses a handful of question types make up more than half the volume, and staff answer them from memory every day.

The fifth question is what would be cheap to get wrong. Of your candidates, which one causes minor inconvenience rather than real damage if the output is wrong? That is where you start — not because it is the most valuable, but because it is where you can learn without risk.

How to Run the First Project

The first step is to write the current process down in plain language: who does it, what they look at, what they produce, how long it takes. If you cannot describe it in a paragraph, it is too big to be a first project — pick a smaller piece of it.

The second step is deliberately unglamorous: have one person do the task with an AI assistant manually, copying and pasting, with no integration and no purchased tools. This feels too primitive to be a real pilot, and it is exactly the right first move. It costs almost nothing and it answers the only question that matters — does AI actually help with this specific task, on our actual messy data?

The third step is measurement, and it must include quality, not only time. Faster output that a manager has to correct is not a saving. Compare against how the work looked before, and ask the person doing it whether it was better or merely different.

The fourth step is a decision with only two options: embed it in the workflow properly, or stop. The failure mode here is neither — the pilot that neither ships nor dies, and quietly consumes attention for months. Set the decision date before you start.

Related: How to Measure ROI on AI in a Philippine Business: What Owners Should Count, and What They Should Ignore explains this in detail.

What This Approach Is Worth

The first return is that you get a real answer quickly. Two or three weeks of this method produces something better than any amount of strategy discussion: evidence from your own operation about whether AI helps with a specific task you actually perform.

The second return is that being wrong is cheap. Because the first project is chosen for low damage and run manually before any purchase, a candidate that does not work costs a week of one person's partial attention. Compare that with a twelve-month platform commitment made on a hunch.

The third return is internal confidence, which is underrated. One completed, measured project changes how the next proposal is received. Staff who watched a small thing work are far more willing to participate in a larger one, and management that has seen a real number is far more willing to fund it.

Related: How to Choose an AI Development Partner on Upwork: A Practical Vetting Guide for Philippine Businesses explains this in detail.

FAQ

Q: What if the week of tracking shows nothing obvious?

A: That is a useful result, not a failure — it usually means the work is genuinely varied and judgement-heavy, which is worth knowing before you spend money. In that case, look at the enquiry log instead; repeated customer questions are the most reliable candidate in businesses where internal work resists standardisation.

Q: Should we hire a consultant to find our use case?

A: You can, but do the five questions first regardless. The answers are things only your own staff know, and arriving at a conversation with them makes any outside engagement far shorter and cheaper. A consultant who cannot work from your own findings is not adding much.

Q: Our data is messy. Should we clean it first?

A: Not before the manual pilot. Cleaning data ahead of knowing what you need is a common way to spend six months and start nothing. The pilot will tell you which specific data actually needs attention, which is usually far less than everything.

Q: How do we stop staff from feeling threatened by this?

A: Involve them in the first two steps and be direct about the purpose. The exercise looks for tasks that consume hours without requiring judgement, which is generally the part of the job people least enjoy. Framing it as reclaiming time rather than reducing headcount is both more accurate for a first project and more likely to get honest answers.

Q: Is a small business too small for this?

A: No — smaller operations often get clearer results, because the work is less fragmented and one person can see the whole process. The method scales down to a handful of staff without modification.

Start With Hours, Not With Technology

The businesses that get value from AI are rarely the ones with the best technology strategy. They are the ones that looked honestly at where their hours go, picked something small and forgiving, tried it by hand before buying anything, and made a real decision at the end. None of that requires knowing anything about AI in advance. It requires a week of attention and a willingness to start with something unimpressive.

At PH AI Works, we help Philippine businesses and Japanese firms operating here run exactly this process — mapping the actual work, choosing a first candidate, and testing it before any commitment. If you know you should be doing something with AI but cannot name what, our free consultation starts with the five questions above rather than a product.

References

• Department of Trade and Industry (DTI) Philippines
• Department of Information and Communications Technology (DICT) Philippines