Ask a leadership team what their first AI project should be and you will hear predictive analytics, or a customer chatbot. Ask the same people where the business actually loses time every week and the honest answer is almost always a spreadsheet.
There is a reason for the gap. The predictive project sounds like the future. The spreadsheet sounds like admin. But one of them has a measurable baseline, a willing user base and a payback you can calculate before you start, and the other does not.
What the spreadsheet is actually doing
In most growing Irish businesses there is at least one workbook performing a job well beyond what a spreadsheet should be asked to do. It bridges two systems that do not talk. It holds logic in formulas nobody has audited in years. It has a filename ending in something like _FINAL_v7_USE_THIS, and there are four near-identical copies in different folders.
It is also, usually, load-bearing. The month-end close depends on it. And it depends on one person.
Why it is the better first project
- The baseline already exists. You can count the hours spent, the errors found, the times the file has broken. Predictive projects have to invent their own success criteria.
- The users want it. Nobody enjoys the reconciliation workbook. Removing it makes you popular, which matters more for adoption than any business case.
- The scope is bounded. One process, one output, one owner. It can be delivered and proven inside a quarter.
- It reduces real risk. Key-person dependency and unauditable logic are genuine exposures, not theoretical ones.
What replacing it involves
Not, in the first instance, buying anything. It involves mapping what the workbook does, separating the parts that are genuine business logic from the parts that are compensating for something upstream, and fixing the upstream problem where you can.
Roughly half of what a bridging spreadsheet does should not need to happen at all. Automating it wholesale preserves the problem in a faster format.
What remains is usually a modest, well-scoped automation with a payback measured in months. It will not be the project anyone puts on a conference slide. It will be the one that makes the second project easy to approve.