Open any HR-tech demo from the last twelve months and you will be told that AI now writes your leave policies. Type a prompt, get a six-page document, paste it into your handbook, done. We’ve watched this play out across the industry and concluded one thing: nobody reads those policies, and the AI that wrote them solved the wrong problem.
The actual question employees and managers ask, dozens of times a week, is much smaller and much harder: why does my leave balance show this number?
That’s where Leave Balance is putting its AI. Not in policy generation — in balance explanation.
The problem with leave balance math
A modern leave balance is the output of at least eight inputs:
- Annual entitlement (statutory + contractual)
- Accrual rate (monthly, per pay period, anniversary-based)
- Carry-over from last year (with caps, expiry dates, exceptions)
- Pro-rating for joiners and leavers
- Adjustments (TOIL, comp leave, manual credits)
- Pending requests (counted or not, depending on policy)
- Encashment events
- Year-end roll-over rules
Each of those is correct on its own. Together, they produce a single number that looks right or looks wrong depending on whether you can hold all eight rules in your head at once. Most people can’t, which is why “why is my balance only 12.4 days?” is the most-reported HR support question in our customer base.
What “AI explain leave balance” actually does
When an employee opens their balance page, a one-paragraph explanation appears above the numbers:
You have 12.4 days available. You started the year with 8 carried over from 2025 (3 of those expire 30 June). You’ve accrued 6.6 days since January at 1.67 days per month, and you’ve taken 2.2 days. Two days are pending approval and not yet deducted.
That paragraph is generated from the same data the numbers come from — no extra source of truth, no risk of the explanation diverging from reality.
Admins get a different version of the same feature: when a support ticket arrives asking “why is X’s balance Y?”, the admin can pull up the employee’s record and get the breakdown immediately, with the policy clauses that drove each line cited.
leave emails? Track your employee's leave with Leave Balance
Why this is the right place for AI
LLMs are good at three things that are exactly relevant here: structuring numerical context, translating policy language into plain English, and personalising explanation to the specific data in front of them. They are not, despite the marketing, good at writing legally-defensible HR policy from a one-line prompt.
The corollary: the cheapest, highest-leverage place to use AI in an HR product is the place where ten variables collapse into one number that someone has to defend. That’s leave balance math. It’s not a flashy demo, but it eliminates the single most common HR support ticket.
What it doesn’t do
We’re being deliberate about scope:
- It doesn’t decide policy. Every rule still comes from your configured policies. The AI explains what the rules produced; it doesn’t change them.
- It doesn’t approve or reject leave. Approval chains stay human.
- It doesn’t write the policy document. That’s a job for your handbook, not your leave tool.
- It doesn’t hallucinate balances. The numbers come from the deterministic accrual engine. The AI only narrates them.
Where it sits in the bigger picture
This is the first AI feature we’ve shipped in Leave Balance. There will be more, but they’ll all share one principle: AI sits on top of deterministic data, not in place of it. The system of record is still the same boring Postgres database with the same audit trail — the AI is just a better translator between that data and the human who has to act on it.
If you’ve ever spent twenty minutes reconstructing an employee’s balance for a payroll dispute, this is for you.
leave emails? Track your employee's leave with Leave Balance
The bigger point
The HR-AI hype cycle is going to overshoot before it settles. The features that survive will be the ones that make a daily, repetitive task easier — not the ones that automate decisions that should stay with humans. Explaining a leave balance is a perfect fit. Generating a leave policy isn’t.
We’ll keep choosing the first kind of feature.