How Boutique DX Builds a Schedule
Engineering · 2026-07-11
"AI scheduling" has become a phrase vendors put on everything. It deserves a precise answer: what actually happens between a manager clicking Generate and a month of shifts appearing? Here is how our engine works — including the parts that are deliberately not machine learning.
Scheduling is a constraint problem, not a text problem
A boutique schedule is a set of hard commitments. Labor law defines maximum hours and mandatory rest. Contracts define who works how much. Your house rules define coverage: a keyholder at opening, enough advisors at peak, a fitter present when alteration appointments are booked. Staff availability and requests define what people can and want to work.
These are logical constraints, and the correct tool for logical constraints is a constraint solver. Boutique DX formulates each boutique's month as a formal optimization model and solves it with industrial solvers — the Z3 theorem prover and Google's OR-Tools — the same class of technology used for airline crew rostering. The solver does not "hallucinate" a plausible-looking rota; it either satisfies your rules or tells you which rules conflict.
This distinction matters because a schedule that is 95% right is not 95% useful. One missed rest-period rule is a compliance incident; one Saturday with no keyholder is a boutique that cannot open.
Demand comes from forecasting — the actual machine learning
The solver needs to know how many people each hour requires, and that is where statistical learning earns its place. Boutique DX forecasts each boutique's traffic using time-series models (built on Prophet) trained on that boutique's own history, not a global average. The inputs are concrete:
- Footfall from in-store people-counting sensors, where installed
- Bookings and events already in the calendar — appointments are demand you can see in advance
- Seasonality at three levels: day-of-week, month, and local holiday calendars
- Weather, which moves boutique traffic more than most operators expect
Luxury boutiques are a distinct forecasting problem. A big-box store serves thousands of transactions a day, so its noise averages out. A boutique may serve forty clients a day, where one VIC appointment changes the afternoon. That is why bookings and events are first-class forecast inputs rather than noise, and why every boutique gets its own model.
Managers stay in charge
The engine proposes; the boutique disposes. Generated schedules arrive as drafts. Managers adjust freely, and every manual edit is checked against the same rule set in real time — the system warns when a change would break a rest rule or leave a coverage gap, instead of discovering it at month end. Staff request swaps and leave from the companion app, and accepted requests are re-solved into the schedule rather than patched over it.
We think of it as three layers with clear responsibilities: statistical models predict demand, a constraint solver arranges people, and humans make the judgment calls neither should make.
Why not just use an LLM?
We use large language models where language is the problem — summarizing a schedule, drafting an announcement. We do not use them to place shifts. A language model predicts plausible text; it cannot guarantee that no advisor works a close-then-open, that contracted hours are met, or that a schedule is even feasible. A solver can, and does, prove it.
The industry data suggests operators sense this. In Legion's 2025 State of the Hourly Workforce study, a majority of frontline managers said AI could make scheduling easier — yet only around one in ten actually uses auto-scheduling. The gap is trust. Our answer to the trust problem is not a bigger model; it is a system whose behavior can be inspected: named rules, explicit constraints, and warnings that cite the rule they protect.
What this looks like in practice
A monthly schedule for a boutique — constraints, requests, events, forecast and all — is generated in minutes, not the hours managers report spending weekly on manual rotas. The manager reviews a draft that already respects labor law and house rules, adjusts for the things only a human knows, and publishes to every advisor's phone.
If you would like to see the engine run on your own boutique's rules, book a demo — bring your most complicated month.
Boutique DX is the operating system for luxury boutiques: AI-powered scheduling, client appointments, queue management, events, and analytics in one platform.