Rules handle leave requests, balances and clash checks, AI answers questions and spots patterns, and people decide edge cases and sick leave follow-up.
Partly, and mostly not with AI. Tracking annual leave and absence is a rules job: a request comes in, the balance is checked, the team calendar is checked and an approval is recorded. Ordinary automation does that reliably. AI helps at the edges, by answering staff questions in plain English, summarising absence patterns and drafting messages, while people decide the awkward cases and handle sick leave with care.
The process: staff ask for annual leave, someone checks how much they have left and whether the team is covered, a manager approves or declines, the calendar and payroll are updated, and sick leave is recorded and followed up. Many Irish employers search for a holiday tracker or an annual leave tracker for exactly this. Those tools are good at the rules part, including the way leave accrues and how public holidays fall. What they do not do is judge whether refusing a request is fair, or how to talk to someone who has been out sick for several weeks. If you still run leave from a spreadsheet and an inbox, the biggest gain is simply moving to one system with fixed rules. AI is an add-on to that, not the foundation.
What each step needs
01● Standard automation
Receive the request through one channel
Requests turn up by email, chat, a form and a quick word at someone's desk, and half of them never reach whoever keeps the records. A simple form or HR system with required fields (dates, type of leave, half day or full day) puts every request in the same place and starts the approval flow automatically. There is nothing here for AI to decide. The value is that nothing is lost and nobody has to retype dates into a spreadsheet.
⛨Works when everyone agrees that a request only counts once it is in the system. Without that agreement, the old channels carry on next to the new one.
02● Standard automation
Check the balance and entitlement
Entitlement follows fixed logic: contract, hours worked, start date, carry-over, public holidays and the statutory minimum under Irish law. Software calculates it the same way every time and shows the employee and manager the balance before anyone approves. A language model is the wrong tool because it can produce a plausible number that is wrong, and a wrong balance becomes a dispute later. Keep the calculation in rules that someone has tested against a few real employees, including part-timers and people who joined mid-year.
⛨Works when the policy is written down and has few exceptions. Irregular hours, part-year workers and mid-year contract changes need to be tested before you trust the numbers.
03● Standard automation
Check team cover and clashes
A rule can flag when too many people in a team ask for the same days, when a key role has no cover or when a blackout period such as year-end applies. The system shows the manager the clash instead of the manager hunting through calendars. The rule only flags; it does not decide. Setting the minimum cover per team is a management choice, and once it is set the software applies it consistently.
⛨Works when minimum cover is defined per team or role. In small teams where everyone is critical, the flag is useful but most requests will still need a conversation.
04● Human review
Decide on clashes and borderline requests
Two people want the same week, someone asks for leave during a deadline, or a request falls just outside policy. These are decisions about fairness, trust and business need. A manager can ask AI to list the facts, such as who had which weeks last summer, but the choice and the conversation stay with a person. Handing it to a model means nobody can explain a decision to the person who was turned down.
⛨Applies to the small share of requests that clash or break a rule. Most requests are approved without a conversation once the rules are clear.
05● AI candidate
Answer staff questions about leave and absence
Staff ask the same things again and again: how many days do I have left, can I carry days over, what happens to my leave if I am sick while away. An assistant that answers only from the written policy and the person's own balance can reply in plain English at any hour and pass the question to HR when the policy is silent. It must quote the policy and never make up an answer about entitlement or the law.
⛨Works when the policy is current and kept in one place. If the policy is vague or contradictory, the assistant will repeat the confusion.
06● Keep human
Record sick leave and hold the return conversation
Recording the dates of an absence is a simple job for rules. The conversation around it is not: checking in with someone who has been unwell, agreeing adjustments and noticing when something deeper is going on. Sick leave data is health data, which the GDPR treats as special category information with strict limits on who sees it and how long it is kept. Keep the follow-up with a named manager or HR contact and keep health details out of general AI tools.
⛨Applies to every sick leave absence. Longer absences can involve medical certificates and workplace adjustments, so HR should own those decisions.
07● AI candidate
Report on absence and handle the leave year-end
Once the data is clean, AI is useful for turning it into a short monthly summary: where absence is rising, which teams have large unused balances, who is close to losing days when the leave year ends. It can draft the reminder emails for HR to review. Totals and carry-over stay with rules; AI describes and highlights. Treat a pattern as a reason to ask a question, not as proof of a problem with any person.
⛨Works when the absence data is accurate and the team is large enough for patterns to mean something. In a team of five, a summary reveals who was out sick and should be shared with care.
A sensible first experiment
Pick one team of ten to thirty people and run leave through a single form and system for one quarter. Measure the number of requests that needed a manual correction, the days between a request and its decision, and the questions HR still had to answer by hand. In the second half, add an assistant that answers balance and policy questions from the written policy, and track how many it resolved without HR. Keep it only if it quotes the policy correctly and the error rate stays close to zero. Check a sample of its answers every week against the policy.
The trap to avoid
The common mistake is to start with AI before the rules are clear. If the policy has unwritten exceptions, a tool will apply the written version and staff will notice. The second mistake is putting sick leave details into a general chatbot or a widely shared report. Decide first what the policy says and what stays private, and only then add automation on top.
Questions teams ask
Do I need AI for a holiday tracker?
No. Most of the work is fixed rules about balances, approvals and calendars, and an ordinary holiday tracker or leave system handles that well. AI is an optional extra for answering staff questions and summarising absence data. Start with the rules and add AI only if the volume of questions justifies it.
Is an annual leave tracker the same as absence management software?
An annual leave tracker usually covers requests and balances. Absence management software typically adds approval flows, team calendars, sick leave, carry-over rules and reports. For a small team a tracker may be enough; for several sites or contract types you will want the fuller system.
Can AI decide who gets the week off when two people ask?
It can list the facts, such as previous leave and team cover, but the decision should stay with a manager. These choices affect trust and fairness, and the person should be able to ask a human why. Use rules for a first-come default if you want one, and let a manager override it.
How should sick leave data be handled?
Treat it as sensitive. Health information is special category data under the GDPR. Limit access to named people, keep only what you need, and do not paste it into general AI tools. Check with your HR adviser or the Data Protection Commission guidance what applies to your records.
Illustrative workflow guidance by Arcgent. Each business needs its own assessment. No integration or savings claim has been verified for your systems.