How to Split Tips Fairly With AI

Tip distribution is arithmetic wrapped around a legal rule, and those two halves need completely different treatment. You can split tips fairly with AI in the sense that it will build the points model, run the weekly calculation, catch the rounding errors and draft the written policy in plain language. What it must not decide is who is eligible to be in the pool, because that is set by law in your jurisdiction and a model asked to opine on it will produce a plausible answer that costs you a tribunal claim.

So the order matters. Establish eligibility from the legal source, then hand the model everything else.

The part that is not negotiable

Two jurisdictions, two specific rules worth knowing before you open a chat window.

United Kingdom

The Employment (Allocation of Tips) Act 2023 requires employers to allocate qualifying tips fairly, and the statutory Code of Practice on Fair and Transparent Distribution of Tips came into effect on 1 October 2024 for England, Scotland and Wales. Employers must have regard to the code when designing a tipping policy, and employment tribunal judges must consider it where relevant. Failing to follow it is not automatically proof of unfairness, but it is the document a tribunal will read next to your policy. Workers can bring a claim, and a tribunal can order a revised allocation or a payment.

United States

Under the Fair Labor Standards Act, managers and supervisors cannot share in a tip pool, and the test is duties rather than job title. Department of Labor opinion letter FLSA2026-13 addresses the awkward real case directly: a restaurant shift supervisor who meets the executive duties test may not participate in a tip pool even when that person also works bartending shifts or helps hosts and bussers. Such a supervisor may keep only tips received directly from customers for service they solely provided, and if tips are pooled among bartenders so they cannot be attributed to them alone, they may take no share. The letter also confirms pools composed solely of managers and supervisors are not permitted. Status under the primary duty test is determined on at least a workweek basis and does not change shift to shift.

Write your eligibility list down as a fact, from the source, before prompting. Then the model's job becomes arithmetic on a fixed roster, which is a thing it is reliably good at.

Build the points model to split tips fairly with AI

Most fair schemes weight by role and hours. Points per hour by role, multiplied by hours worked, summed, then the pot divided pro rata. The model is useful here for stress-testing weights rather than inventing them.

Our tip pool covers these roles. Eligibility is already decided
and is not up for discussion.

<roster>
Role            points/hour
server               1.0
bartender            1.0
host                 0.6
busser               0.6
kitchen (BOH)        0.5
</roster>

Weekly pot: 2,840.00
Hours worked this week:
  servers 196, bartenders 64, hosts 38, bussers 44, kitchen 150

1. Calculate total points, value per point, and the payout per
   role and per hour worked.
2. Show the arithmetic so I can check it.
3. Then tell me which role does best and worst per hour, and
   whether any role's share looks hard to defend to that team.

For those numbers the total comes to 196 + 64 + 22.8 + 26.4 + 75, which is 384.2 points, and the pot divides at roughly 7.39 per point. A server working 40 hours takes about 295.60. A kitchen hand working 40 hours takes about 147.80. That ratio is the conversation you are actually having with your team, and seeing it as a number rather than a principle is the point of the exercise.

The third instruction is the valuable one. A model is good at spotting that your host weighting produces an hourly share below the busser weighting once you account for the shift patterns hosts actually work, which is the sort of thing that surfaces as a grievance three months later.

Run the weekly calculation without a spreadsheet you cannot audit

Two sensible patterns depending on volume.

• Low volume: paste the hours each week and ask for the payout table plus the arithmetic. Keep the output. It is your record.

• Higher volume: ask the model for the spreadsheet formulas once, then never prompt again. The calculation does not change week to week, and a formula you can audit beats a chat you have to re-trust. We covered that handoff in prompting AI to write a spreadsheet formula.

Either way, insist on seeing the working. Tip arithmetic with uneven hours and fractional points is exactly where a model will produce a confidently rounded total that does not reconcile to the pot, and a total that is 40 short is a problem you want to find before payday rather than after.

Reconcile the pot against your card settlements rather than against your own notes. If card tips arrive separately from takings, using AI to read receipts and invoices is the faster way to pull weekly tip totals out of statements than typing them in.

Draft the written policy

The UK code expects a written policy, and having one is good practice regardless of jurisdiction. This is the task AI is best at here, because it is translation: turning a decision you have already made into language your staff will actually read.

Draft our tipping policy for staff, not for lawyers.

Facts:
- Eligible roles and weights: [paste roster]
- Pot includes: card tips, cash tips handed in, service charge
- Pot excludes: nothing
- Paid: with wages, the Friday after the week it was earned
- Who runs the calculation: [name], checked by [name]
- Managers and supervisors: not eligible, per [cite your rule]
- Disputes: raise with [name] within 14 days, written answer
  within 14 days

Constraints:
- Under 500 words. Plain English, reading age 12.
- State what happens to tips in the week someone leaves.
- State how the policy changes and with what notice.
- No legal disclaimers. No hedging language.
- End with a worked example using round numbers.

The worked example at the end is what makes the policy believable. A staff member who can see that a 40-hour server week returned about 295 last week will trust the scheme in a way that no paragraph about fairness achieves. If this policy is joining a wider staff document, keep the formatting consistent with it: see using AI to write an employee handbook.

One more thing worth generating while you are here: the explanation you give verbally when you roll the scheme out. It is a different register from the policy and it is where most of the pushback lands.

Where this interacts with scheduling

Points-per-hour schemes make the rota a pay decision, which it was not before. Whoever builds the schedule is now distributing tip income, and that changes how people read a quiet Tuesday. Worth being deliberate about: using AI to write a staff rota covers the scheduling side, and our AI for small business pillar has the wider set of operations tasks worth handing over.

Frequently asked questions

Can managers take a share of tips?

In the US, no, where the person meets the executive duties test, and that holds even on shifts when they are bartending. They may keep only tips a customer gave them directly for service they alone provided. Check the current Department of Labor guidance for your situation rather than relying on a summary, including this one.

Should kitchen staff be in the tip pool?

It is a choice rather than a rule in most places, and it is increasingly common. The UK code is concerned with whether your allocation is fair and transparent rather than mandating a particular split, so what matters is that you can explain the weighting and applied it consistently.

Can AI tell me whether my split is legal?

No, and this is the one question to keep away from it. It will answer fluently, the answer will be jurisdiction-blurred, and you will have no way to tell. Get eligibility from the statute, the statutory code or an adviser, then let the model do the arithmetic within those limits.

What about cash tips kept by the individual?

Treat them explicitly in the policy rather than leaving them unmentioned, because an unaddressed cash practice is where most disputes start. Whichever way you decide, write it down and apply it to everyone.

How often should the weighting be reviewed?

Once or twice a year, and after any real change in service model. A points table set when you had no table service rarely still makes sense two years later, and the review is a half-hour job once the calculation lives in a formula.