Almost every growing company starts tracking time off the same way. One tab in a spreadsheet, a row for each person, a column for the annual allowance, a column for days taken and a column that subtracts one from the other. It works in January. By October, two or three people have a balance that they dispute, and the person who keeps the sheet is spending an afternoon a month reconstructing what happened.

The sheet did not break. It was built to hold a number, and the number is derived from events that the sheet never recorded.

What the balance depends on.

A leave balance looks like arithmetic and is really a small set of rules applied over time. It depends on how fast time accrues, which may vary with length of service. It depends on any waiting period before a new hire can use it, and on a cap that stops it growing past a limit. It depends on what happens at the end of the year: a carryover up to a ceiling, a payout, or loss. It depends on whether time is taken in whole days, half days or hours, and on what happens when a public holiday falls inside a period of leave. It depends on unpaid leave, on adjustments for a mistake in a previous month, and on rules that differ by state or country.

A policy with six or seven of these moving parts is not a column. It is a program, and in a spreadsheet the program is a set of formulas that one person wrote and one person understands.

How the drift happens.

Drift comes from three habits that are natural and fatal.

The first is typing the balance. After a request is approved, someone subtracts the days and types the new figure. The figure is now a stored value with no history. If a day is subtracted twice or not at all, the error is invisible until someone adds it up.

The second is editing a formula mid-year. The company changes its accrual rule in July, so the formula is updated. The balances for the first half of the year, which were computed under the old rule, are now recomputed under the new one, or are not, depending on how the sheet was built.

The third is keeping the policy in a person's head. The person who maintains the sheet knows that one employee's carryover was agreed as an exception, that another started mid-month and accrued from the first full month, and that a third took a half day that was never logged. None of it is in the sheet. When that person is on leave, the sheet is a puzzle.

None of these is carelessness. They are what happens when a derived figure is stored as if it were a fact.

The fix: a ledger, not a balance.

The reliable design is to record events and compute the balance from them. A ledger has one row per event with a date, a person, a type and a quantity: an accrual, an approved leave request, a carryover, a payout, an adjustment with a reason. The balance is never typed. It is the sum of the person's rows up to a date.

Three properties follow.

  • The balance can be explained. Any balance traces to its rows, so a dispute is settled by reading a list instead of by memory.

  • History does not change when the rule does. A new accrual rule creates new accrual rows from its start date. Old rows stand.

  • An error is correctable and visible. A mistake is fixed by an adjustment row with a reason, and the original stays in the record.

This works in a spreadsheet, in an HR tool and in a custom system. The design matters more than the product. The same reasoning applies to any figure that is derived from events, including the stock count, which we discuss in other pieces on how counts drift between systems.

Reconcile it to payroll.

A balance matters because it becomes money: a payout at the end of the year or on termination, and a liability on the books while it is unused. Payroll and the leave record are two systems that describe the same facts, and they disagree more often than companies expect.

Once a month, compare the leave taken in the ledger with the leave paid in payroll for the same period. They should match. Where they do not, one of them is wrong, and the difference is usually a request approved in one place and not entered in the other. A monthly reconciliation takes minutes when the ledger is clean, and it is far cheaper than finding the discrepancy when someone leaves.

An illustration, with made-up numbers.

Imagine a company of forty people with a spreadsheet balance for each. In July, the company changes its accrual from fifteen days a year to eighteen for people with over two years of service. The sheet's formula is edited in place. For the eleven people affected, the first half of the year is now recomputed at the higher rate, adding about a day and a half each to their balances.

Nobody notices until a person leaves in October. Payroll pays out the balance on the sheet, which includes a day and a half they had not accrued yet under the old rule. Across the eleven people, the sheet overstates the liability by about sixteen and a half days, and one payout is wrong in a way that is hard to take back.

With a ledger, the rule change creates new accrual rows from the first of July. The first-half rows stand. The October balance is the sum of the rows up to the leaving date, and the payout matches.

Questions to ask of any tool.

If you outgrow the sheet and look at an HR tool, test it against the ledger idea before you look at the screens. Can you see, for any person, every event that makes up the balance, with dates? When you change an accrual rule, does it apply from a date, or does it rewrite history? Can you record an adjustment with a reason without overwriting a number? Can you export the events to compare with payroll?

A tool that answers yes to all four will let you explain any balance. A tool that stores a balance and offers a pretty calendar will move the drift into a different system. The calendar is a convenience and the ledger is the control, so choose on the second.

The first thing to try.

Choose three people with the longest service and rebuild their balances from the start of the year by hand from the approved requests. Compare what you get with what the sheet says. If they match, the sheet is healthier than most. If they do not, the gaps will show you which of the three habits is the cause. The rebuilt list is also the first version of a ledger. For where a spreadsheet stops being enough, see HRIS for a small business, and for the entry end of the same record, day-one failures on a spreadsheet.

Where we come in.

We start with a no-risk discovery: we rebuild a sample of balances from your requests, show where your current sheet or tool drifts and why, and propose the ledger and the payroll reconciliation, and you pay for the discovery only if you go on. A build comes with a guaranteed estimate, so if we estimate low, we absorb the difference. After go-live, Managed Technical Operations runs the monthly reconciliation and flags the differences.

A balance you can explain is one you do not have to argue about.

Find out what a connected system would change in your business.

In a no-risk discovery we look at how your work moves between your systems today and show what one connected system would do differently. You pay only if you proceed. Or see how we approach it.

Book the no-risk discovery

More on the same problem:

Where we're not the right answer

We'll tell you if we're a fit. If we're not, we'll tell you that too.

  • Your current stack works and nobody wants to change it
  • You want licences resold at a discount and nothing else
  • Your internal team owns the operating model and is not handing it over
  • You want hours of configuration work and nothing run for you: that is on our services pages, and it is not a managed solution
How we start

Most of our best clients come to us with a feeling, not a plan.

"Something isn't working." "We're outgrowing our tools." "We're afraid to make the wrong move." No-Risk Discovery is a short, practical conversation that gets you clarity before you commit to anything big. We'll tell you if we're a fit. If we're not, we'll tell you that too.

Talk to us

Have a question about this?

Say who you are and ask. A person replies within one business day.

A new client, or thinking about becoming one
What are you looking for?
An agentic solution (AI OS) is licensed and sold with managed services. An integrated solution is built for a professional services fee. A Service is professional work, done by the hour, the month or the project. Not sure yet is a fine answer.

Agentic solutions (AI OS)

Integrated solutions

Services

Platforms You Use or Are Considering

Your details are never shared or sold.