Ops & Supply

How to close out the restaurant pos at end of night with AI

Ops & Supply3 AI tools7 steps6 friction points

Want an agent to handle this on your live data? See the Starch version →

Closing out the restaurant POS at end of night means reconciling cash drawers, verifying that card batch totals match what the system reports, accounting for voids and comps, tipping out servers, and producing a shift summary before the last person locks up. It's a sequence that happens every single night, takes 20–40 minutes when it goes smoothly, and can stretch much longer when a drawer is off or a batch fails to close.

The workflow feels like an AI problem because most of the friction is cognitive, not physical — comparing numbers across multiple reports, spotting discrepancies, remembering which steps come in which order, and writing up a clean end-of-night summary for the owner or manager on duty. If you could hand someone a stack of shift data and ask 'what's off and why,' that's exactly what people hope AI can do.

ChatGPT, Claude, and Gemini can genuinely help here — as thinking partners and document generators. You can paste in Z-report data and ask the model to check the math, describe a discrepancy and get a likely explanation, or ask it to draft a nightly summary in a format your manager expects. What they can't do is reach into your POS system, pull the numbers live, or remember what last Tuesday's close looked like.

Skip the manual work → See the Starch version

Ops & Supply3 AI tools7 steps6 friction points

AI walkthrough

How to do it with AI today

A practical walkthrough using ChatGPT, Claude, and other off-the-shelf LLMs — what they're good at, what you'll have to do by hand.

Tools that work for this

ChatGPTClaudeGemini

Step-by-step

1 Run your Z-report and cash drawer reconciliation report inside your POS (Toast, Square, Lightspeed, Clover, etc.) and export or screenshot the totals — net sales, cash expected, cash actual, credit card batch total, voids, comps, and tips.

2 Open ChatGPT or Claude and paste the raw numbers from your Z-report directly into the chat window. Include every line item — don't summarize first, because the model needs the actual figures to check the arithmetic.

3 Ask the model to reconcile the numbers: identify whether the cash drawer is over or short, whether the card batch total matches net card sales, and flag any line items that look unusual given the total cover count.

4 If a discrepancy appears, describe the context in a follow-up message — for example, 'we had two voids after 9pm and one comp for a manager meal, total $74' — and ask the model to account for those adjustments and re-check the reconciliation.

5 Ask the model to generate an end-of-night summary paragraph you can paste into Slack, email, or a shared Google Doc for the owner. Give it your preferred format: date, total sales, covers, cash over/short, card batch status, any open issues.

6 For tip-out calculations, paste in total tips collected and your house tip-out formula (e.g., 'servers tip out 3% of net sales to bussers and 1% to bar'), and ask the model to produce a per-person tip-out table.

7 Save the session as a browser bookmark or copy the prompts into a doc so you can repeat the same flow tomorrow night — the model has no memory of tonight's close.

Prompts you can copy

Here is my Z-report from tonight: [paste data]. Cash expected: $412. Cash counted: $389. Card batch total: $3,841. Net card sales per POS: $3,798. We had one void for $43 and one comp for $0. Reconcile this and tell me what's off.

We're $23 short in the cash drawer and I can't figure out why. Here's the breakdown of every cash transaction tonight: [paste]. What are the most likely causes and what should I check first?

Write an end-of-night shift summary for the owner. Format: Date, Total Sales, Cover Count, Cash Over/Short, Card Batch Status, Open Issues. Data: [paste your figures].

Our tip-out formula is: servers tip 3% of their net food sales to bussers, split evenly, and 1% to the bar. Here are tonight's server sales: [paste]. Calculate each server's tip-out amount and the total going to bussers and bar.

I need a repeatable end-of-night POS close checklist for a 60-seat full-service restaurant using Toast. Include Z-report steps, cash reconciliation, card batch close, tip-out, and manager sign-off. Make it a numbered list under 20 steps.

Reality check

Where this gets hard

The walkthrough above works — until your numbers change, the LLM hallucinates, or you have to re-paste everything next month.

No live connection to your POS — every session starts with manual copy-paste from the Z-report, and one missed field means the reconciliation is wrong from the start.

The model has no memory between sessions, so you re-explain your tip-out formula, your comp policy, and your reporting format from scratch every single night.

Output structure drifts run to run — the summary format you carefully prompted on Monday isn't guaranteed to match what you get Thursday, even with the same prompt.

When a number is off, the model can only reason about what you paste in; it can't pull transaction logs, check voided tickets, or cross-reference the batch against your payment processor independently.

Nothing is stored — if you need to compare tonight's close against last week's, you'd have to paste both sets of data into the same session manually and hope the context window holds.

Tip-out calculations are only as accurate as the data you paste; if you forget to include a server's side-work sales or a late add-on, the model has no way to catch the omission.

Tired of the friction?

Starch runs the whole workflow on live data — no copy-paste, no hallucinated numbers, no re-prompting next month.

See the Starch version →

Starch alternative

The same workflow on Starch

Starch is an agentic operating system — you describe the nightly close workflow you want, and an agent builds a persistent app that runs it against your live data every night, not a one-off prompt you re-run manually.

Connect your POS data through Starch's integration catalog — the agent queries it live each night, pulling actual sales, voids, comps, and cash figures without any copy-paste from you.

Connect Plaid once and Starch syncs your actual bank transactions on a schedule — so the Transaction Insights app can flag the moment a card batch deposit doesn't match what your POS reported, automatically.

Describe your nightly summary format in plain English — 'every night at midnight, generate a shift summary with total sales, cash over/short, card batch status, and any comp or void over $50, then post it to Slack' — and Starch builds and runs that automation continuously.

Your tip-out formula lives in the app, not in a prompt you retype — the agent applies it to each night's server sales and produces the distribution table without you re-explaining the rules.

Use the Project Management app to track recurring close issues — a drawer that's been short three nights running, a batch that failed twice this week — so patterns surface as tasks instead of disappearing into chat history.

Because the app persists, week-over-week comparison is built in — ask 'how does tonight's close compare to the same day last week?' and the agent has the data to answer without you digging through exports.

Get closed-beta access →

Toolkit