The problem
Employees and clients ask the same things: why is my tax code different, how is holiday pay worked out, what is this pension deduction. If you answer 80 such emails a month at 5 minutes each, that's 400 minutes, 6.7 hours a month.
What the AI does
Use ChatGPT, Claude or Gemini to draft a bank of generic reply paragraphs on the most common questions, from your own explanations and public guidance you paste in. A payroll manager reads and approves each one. When a query arrives, you pick the template and add the person's details yourself. For tax code questions, templates point the employee to HMRC. Never paste payslips or personal data into the assistant.
A worked example
A bureau in Sheffield with 20 clients gets about 80 payslip and pay queries a month. It builds templates with an assistant and has the payroll manager approve them.
Before
5 minutes to write each of 80 replies: 400 minutes, 6.7 hours a month
After
About 3 minutes to choose, fill in and send: 240 minutes, 4 hours a month
What it's worth
About 2.7 hours a month saved (2 minutes x 80 emails = 160 minutes), worth around £80 if staff time is worth £30 an hour. Cost is free to about £20 a month for one person; free plans have usage limits.
How to set it up
- List the ten questions you answer most often.
- Ask ChatGPT, Claude or Gemini for a short plain-English reply to each, with no personal data.
- Have your payroll manager check each against the payroll software and HMRC guidance.
- Save them as email snippets.
- Add the person's details yourself and read each reply before sending.
What goes wrong
- Payslips and pay data are personal data. Don't paste them into a consumer AI plan.
- Tax code and pension questions can need a specific answer; the template should point to HMRC or the pension provider where it can't decide.
- A template that gives the wrong holiday pay rule gets repeated. Review them when the rules change.
- Free assistant plans have usage limits, and paid plans cost between about £4 and £20 a month depending on the vendor.