By Sagar Shankaran, Founder of CallSphere
Slot systems time-stamp every jackpot lockup and reset. That baseline turns a casino AI project into recovered theoretical win, proven in one quarter.
Key takeaways
What is your average lock-to-reset time on a handpay?
Not your guess. The number. Nearly every general manager I ask cannot answer it, and nearly every one of them is sitting on the data — because the slot system stamps the moment the machine locks up and the moment an attendant resets it. IGT Advantage, Aristocrat's Oasis 360, Konami SYNKROS: they all record it, and almost nobody pulls the report.
That is a shame, because the handpay is the best process in a casino to test an AI investment against. It happens thousands of times a month, it has a hard start and a hard stop already in your systems, and the cost of it running slow is countable in theoretical win — the only currency your slot director argues in.
Deloitte's State of AI in the Enterprise 2026 found 84% of organisations investing in AI report positive returns — and the pattern behind the ones that do is consistent and unglamorous: map one messy process, put a human in the review seat, prove time saved or errors reduced, and only then widen the scope. The handpay is that one messy process for a casino floor.
Sixty days, exported from your slot system, four columns: asset number, jackpot amount, lock timestamp, reset timestamp. Add the zone and the shift. That is your baseline, and you must capture it before anybody installs anything, because after the fact you will be arguing about whether the improvement was the software or the new swing-shift schedule.
What comes back usually surprises people. The average is bad but the tail is worse: a handful of handpays every week sitting at forty minutes because the attendant was across the floor on a jackpot over the dual-signature limit, or because the cage had to bring the money, or because the guest was a foreign national and nobody on shift was sure whether the paperwork was a W-2G or the form for a treaty country.
Proving that an AI investment paid for itself means taking one process you already time-stamp, capturing thirty to sixty days of it untouched, changing one thing, and comparing the same shift, the same zone and the same day of the week. Everything else is a vendor slide.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent for hospitality in your browser — 60 seconds, no signup.
While you are pulling it, pull two more figures: handpays per month, and average coin-in per minute on the banks where handpays actually happen. You will need both in a minute.
A $2,400 hit on a $5 machine at 8:12 p.m. The candle lights and the machine locks. Here is where the minutes go today: the attendant may or may not notice the candle, may or may not be on that end of the floor, has to walk to the machine, radio for the amount, take the guest's identification, walk to a bank or the cage for the money, complete the jackpot paperwork, get a floor supervisor for the second signature, pay, and reset.
With an agent watching the slot system's lockup events, the sequence starts differently. The lockup pages the nearest available attendant on her handheld with asset number, denomination, amount and zone — not a general call to the floor. At the same time it pulls the player's record and pre-fills the W-2G fields it already has from the card and from prior filings, and flags the jackpot as over the dual-signature limit so the supervisor is walking before he is asked.
flowchart TD
A["Machine locks on a $2,400 hit at 8:12pm"] --> B["Agent pages nearest attendant with asset and denomination"]
A --> C["Agent pre-fills W-2G from the players club record"]
B --> D["Attendant verifies identification at the machine"]
C --> D
D --> E{"Over the dual-signature limit?"}
E -->|Yes| F["Floor supervisor co-signs"]
E -->|No| G["Attendant pays and resets the machine"]
F --> G
G --> H["Lock-to-reset minutes logged against the baseline"]
Nothing in that sequence is clever. It is dispatch and paperwork — the two things that eat the fourteen minutes. The attendant still walks, still checks the licence, still counts the money into the guest's hand. She just does not spend four of those minutes finding out where she is going and copying a date of birth off a driver's licence into a form.
This is where most casino AI business cases go wrong. They count attendant hours saved, the finance director points out that you are not cutting slot attendants, and the project dies. Count the machine instead.
| Assumption | Figure |
|---|---|
| Slot machines on the floor | 1,100 |
| Handpays per month | 2,300 |
| Baseline average lock-to-reset | 14.2 min |
| Average after change | 9.4 min |
| Minutes saved per handpay | 4.8 |
| Machine-minutes recovered per month | 11,040 |
| Share where the guest keeps playing (assumed) | 70% = 7,728 min |
| Coin-in per minute in play on these banks (assumed) | $8.50 |
| Hold (assumed) | 7% → $0.60 theoretical win per minute |
| Theoretical win recovered per month | $4,637 |
| Attendant capacity returned | 184 hrs/month |
| Overtime share of that, assumed 25% at $29 loaded | $1,334 |
| Running cost, all in | $2,100 |
| Net per month | $3,871 |
| One-time setup $18,000, payback | 4.7 months |
Notice the 70%. That factor is the honest part of this table. Not every guest keeps playing after a $2,400 jackpot — some cash out and go to dinner, and counting them as recovered play would be lying to yourself. Set it conservatively and be ready to defend it.
Notice also what is not in the table: no headcount reduction. Slot attendant coverage is a service level, and on a busy Saturday you want every one of them. The labour line only counts if you can actually show overtime falling.
Regional properties see February and March lift from tax refunds, and everybody sees a September change when football starts. A new bank of high-denomination machines moves your handpay count and your lock-to-reset average all by itself. Compare October to June and someone will take the claim apart in four minutes.
So do it the way you would test a new game: pick two zones with similar denomination mix and similar handpay volume, change one, leave the other alone, and compare them to each other in the same weeks. Report by shift, because graveyard and swing are different worlds and the average across all three hides both.
And watch the second number, not just the first. Average lock-to-reset is the headline. The percentage of handpays over twenty minutes is what guests actually complain about, and it is often the number that moves most.
Still reading? Stop comparing — try CallSphere live.
See the hospitality AI agent handle a real call — complete, industry-specific, and live in your browser. No signup.
Nothing here removes a person from the parts that matter. The attendant verifies the identification against the person sitting at the machine. The guest signs the tax form. The second signature over your dual-signature limit is a control your internal controls require, and a machine does not get to be either signatory. The physical reset is done by a badged employee.
Two more places to keep hands off. Anything that smells like a tilt, a tampering, or a disputed award goes to surveillance and to a supervisor, not into a faster payout path — speed is exactly wrong there. And foreign national paperwork is a judgment call about tax status and treaty position that your cage and your finance office should be making, not a form filled in from a guess.
One further caution: a pre-filled form is still your filing. If the players club record has a stale address, the W-2G goes out with a stale address. Building the habit of checking the pre-filled fields against the identification in hand is a one-line addition to the attendant's routine, and it needs to be in the written procedure from day one.
You do not need a vendor for the first step. Ask your slot performance analyst for sixty days of lockup and reset timestamps by zone and shift, plus handpay counts, plus coin-in per minute on those banks. That is a report request, not a project.
Then hold the meeting with three people in the room: the slot director, the finance director and the compliance officer. Slot knows whether the minutes are real. Finance decides the continuation factor and the theoretical win figure. Compliance says what has to stay in human hands. If those three agree on the baseline before anyone sees a demonstration, you will be able to prove or disprove the whole thing in one quarter.
Yes, and the mess is itself a finding. If lock and reset timestamps are unreliable on some banks, you have discovered that you cannot measure service on those banks at all — which matters far beyond this project. Start with the zones where the data is clean and expand.
Then your handpay volume changes and your baseline resets. The long-standing $1,200 slot reporting line has been the subject of industry lobbying for years, and if it moves, re-baseline before comparing anything. That is a good argument for keeping the measurement in place permanently rather than running it once.
Do not report it by attendant name. Report by zone and shift. The moment a per-person leaderboard appears, you will get faster paperwork and worse identification checks, which is the opposite of what you want, and your union representative will have a legitimate complaint.
If average lock-to-reset in the test zone does not beat the control zone by at least three minutes, stop. That is the threshold at which the theoretical win recovered stops covering the running cost at any believable continuation factor. Writing that stopping rule down before you start is the discipline that separates the 84% who report a return from the ones who quietly keep paying.
A last note on the calls this generates. Every large jackpot produces phone traffic — the guest calling back about the tax form, the mailed copy, the players club points, the room offer that followed. CallSphere builds AI voice and chat agents that answer business phone lines and web chat, book appointments and capture leads around the clock, which is a reasonable place to put the same measure-first discipline: baseline your abandoned call rate for a month before you change anything, then compare. The arithmetic works the same way on the phone as it does on the floor.
Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
Casino surveillance footage cannot leave the property. On-premises AI in 2026 cuts a 90-minute disputed handpay review to nine, with no video sent to a vendor.
The pre-season exterior survey eats four weeks of May and never gets finished. What autonomous inspection routes change for coastal vacation rental managers.
What an unplanned coach breakdown costs a tour operator, the engine readings that warn you first, and how to test condition-based monitoring on two coaches.
When per-seat AI pricing stops making sense for a host agency or tour operator, what running an open model actually costs in people, and the data case for it.
ROH, BAR, DQQB, roll-in shower: the lodging shorthand general AI mangles on the phone, what 2026 trade-trained models fixed, and what bad bookings really cost.
12,000 objects at 15 records a day is 800 working days. How AI agents working in parallel cut a museum's collections-online backlog to about five months.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI