Skip to content
AI Strategy

How to Find the Right AI Use Cases in Your Business

Most AI projects that disappoint were chosen the wrong way round. Someone saw a tool, liked it, and went looking for a problem it could solve.

The projects that pay back start from the other end. They begin with a task that is already costing the business time or money, put a number on it, and only then ask whether AI can help. This guide sets out a simple way to do that, so you know what your first AI project should be before you speak to any vendor, including us.

Start with the business, not the AI

Good AI use cases hide in plain sight. Look for the places where people repeatedly read, move, check, rewrite or reconcile information. Copying figures from an email into a spreadsheet. Checking an invoice against a purchase order. Rewriting call notes into a client file. Reading a hundred enquiries to find the ten that matter.

Then measure the pain before you brainstorm the fix. A task that "takes ages" is not a use case yet. A task that takes eight hours a week is.

Five questions to ask your team

The people doing the work know where the time goes. Ask them directly. You will get better answers in a short conversation than in any survey.

What do you do repeatedly?The same steps, every day or every week, with only small changes.
What takes too long?The tasks people put off, or that eat the end of every day.
Where do mistakes happen?Rework, missed follow-ups and errors a client spots before you do.
Where are you copying data between systems?Anything that is typed in twice is a strong candidate.
Where are expensive people doing low-value work?Partners doing admin, senior accountants matching payments, clinicians writing up notes.

Put a number beside the pain

Take the answers and turn them into an AI opportunity map: one row per task, with how often it happens, how long it takes and what it costs in hours. This is the core of any AI opportunity assessment, and you can do it on a single sheet of paper.

TaskFrequencyTime eachCostAI potential
Qualify inbound leads50 / week10 mins8+ hrs / weekHIGH
Write up actions after client calls15 / week25 mins6+ hrs / weekHIGH
Chase overdue invoices30 / week5 mins2.5 hrs / weekMEDIUM
Negotiate a key client renewal2 / month2 hrs1 hr / weekLOW

The first row is the worked example from Jamie Walsh's founder briefing. The other rows are illustrative.

The first row alone is more than a full working day every week. At a loaded staff cost of €40 an hour, that is over €17,000 a year spent on one task, before you count the leads that went cold while they waited.

Notice what the map does. It turns a vague sense that "sales admin is a pain" into a figure you can compare against the cost of fixing it. It also shows you which tasks are not worth automating, which matters just as much.

How to judge AI potential

The last column is the one people find hardest. A simple rule helps: if a task starts with reading text, it is a strong candidate. If AI can read something, it can usually classify it, flag what matters and recommend a next step.

Strong candidates

  • Sorting and prioritising emails and enquiries
  • Reading invoices, contracts and postings to extract or check details
  • Matching records across two systems
  • Drafting standard documents, replies and reports
  • Turning call transcripts into action lists

Stays with people

  • Relationship work with clients and suppliers
  • Negotiation and anything that depends on trust
  • Final judgement on decisions that carry real consequences
  • Approving anything client-facing or financially material

An accounting practice we worked with started with an AI agent that read the text descriptions on accounting postings. Once the team saw that, they realised the same capability applied anywhere a judgement began with reading text, and they found the next uses themselves.

Pick the right first project

Your map will probably show several good candidates. The best first AI pilot is not always the biggest number. It is the one most likely to succeed and be seen to succeed. Look for three things:

  • Narrow enough to finish. One task with a clear start and end, not "improve sales".
  • Owned by one team. One manager can say yes and one group of people uses it.
  • Safe if it gets something wrong. A mistake is caught and corrected before it reaches a client or moves money.

If three departments have to agree before it can start, it is the wrong first project.

Run it in three stages

Do not guess what would help people. Ask, then build exactly what they asked for. Teams are often particular about how things should work, and they are usually right.

WEEKS 1–2

Ask

The task they name, not yours.

Start with the task the team says they want gone. It is more likely to be used, and they already know what good looks like.

WEEKS 3–8

Prove

A result they can feel.

Show it working in their own week. Give someone back the part of the job they resent and you have made the argument for you.

MONTH 3 ONWARDS

Spread

Champions do the selling.

A respected sceptic who comes round will convince the rest of the business faster than you can.

Measure it

Before the pilot goes live, record how long the task takes today. After it goes live, measure again. That baseline is what turns a good feeling into a business case for the next project.

The savings do not need to be dramatic. One hour saved per person per week is 52 hours a year, more than a full working week of capacity for every person it touches. Our guide to what AI costs an Irish business shows how to set that against the cost of the build to get a payback period.

Real examples of AI use cases

These are live systems, each of which started as a row on a map like the one above.

Finance & AR

Invoice reconciliation across Salesforce and Xero

An AI agent matches payments in Xero against open invoices in Salesforce, applies the agreed allocation rules and flags exceptions for a senior accountant to review.

Four days of senior accounting time freed every month
Professional services

Work-in-progress to invoice

The agent pulls WIP from the practice management system, applies each engagement's billing rules, drafts the invoice and routes it to a partner for approval.

Ten days off the cash conversion cycle on average
Founder productivity

Email triage agent

A founder's inbox of 200 messages a day is read, sorted by priority and drafted for routine replies. Only what needs the founder is escalated.

Eight hours back every week
Month-end close

Balance sheet reconciliation

Account tie-outs, intercompany matching and FX adjustments run automatically. The senior accountant reviews the result instead of doing the work.

Close compressed from days to hours
Construction

Construction hub

Five separate spreadsheets and shared folders brought into one platform, including AI that reads contract PDFs and builds the payment schedule for staff to check.

Payment schedules extracted in under 30 seconds

Two more from Jamie's own work show how wide the range is. In a healthcare practice, AI sped up clinician report writing and the case admin around each patient. Clinicians kept the part they trained for, the time with patients, and lost the paperwork they resented. And lead qualification, the first row of the map above, is one of the most common first projects we see.

Not every use case needs a custom build. Some are best handled with AI automation of a fixed process. Others need an AI agent that can read, decide and act within limits. Our explainer on business process automation covers where the line sits. The method for finding the task is the same either way.

Find your first use case

Finding the right first project is the step that decides whether AI pays for itself. Once you have it, the rest of the six-step AI readiness guide shows how to build on it.

If you would rather not do it alone, we will map the highest-value opportunities in your business. Book a free 30-minute call and we will tell you which task to start with, what it would cost and what you would get back.

Thirty minutes, on Dublin or Atlanta time.