ZenotiHelp

Review AI scheduling's upstream inputs

ZTZenoti Docs Team Updated October 25, 2026 6 min readGuide

Availability: AI Employee Scheduler is in Beta. Your center is enrolled automatically by Zenoti β€” there is no self-enrollment or join step required.

Zenoti's AI Employee Scheduler builds its recommended roster from two distinct stages: a demand forecast and a schedule-generation pass. Understanding what each stage uses β€” and what it deliberately excludes β€” helps you assess why the system recommends the shifts it does and how to get the most accurate suggestions.

Audience: Center managers who oversee provider scheduling and want to understand how AI scheduling recommendations are generated.

Before you begin

  • Your center must be enrolled in the AI Employee Scheduler Beta.
  • Familiarize yourself with the Schedule Grid and the day-level coverage summary before reviewing inputs (see Use the schedule grid and Review the day-level coverage summary).
  • No additional configuration is required to view inputs β€” the forecast and schedule refresh automatically on a nightly cadence.

Steps

Understand the two-stage process

The AI Employee Scheduler operates in two sequential stages. The stages are independent: stage two consumes the output of stage one, but a manager-triggered recompute can regenerate stage two without re-running stage one.

Stage 1 β€” Demand forecast

The system builds a demand forecast from your center's own historical booking data. The forecast covers the next 28 days (four weeks) and refreshes nightly.

Stage 2 β€” Schedule generation

The system combines the demand forecast with your provider roster and center configuration to produce a recommended schedule. This stage also refreshes nightly and can be re-run on demand by a manager against the existing forecast, without restarting the demand-prediction process.

Note: A custom date-range picker for one-off forecast windows is not yet available. The forecast window is always the rolling 28-day horizon.


Review what the demand forecast uses

The demand forecast is built exclusively from your center's historical booking activity. It does not use prices, revenue, the current or previously published employee schedule, or any individual guest's personal details.

The forecast draws on ten categories of historical signals:

  1. Service and time of day β€” Which service was booked and in which part of the day it was requested.
  2. Calendar position β€” The day of the week, the specific date, the month, and whether the day falls on a weekend.
  3. Public holidays β€” Regional public holidays that affect booking patterns for your location.
  4. Recent slot activity β€” How many bookings occurred in the same time slot over recent weeks.
  5. Service popularity β€” How frequently the service has been booked overall.
  6. Slot stability β€” How consistent demand has been for this slot over time.
  7. Cancellations and no-shows β€” Recent cancellations and no-shows, which affect net realized demand.
  8. Waitlist requests β€” Guests who requested this slot but could not be booked, reflecting unmet demand.
  9. Membership and package appointments β€” Appointments redeemed using a membership benefit, membership credit, or package.
  10. Campaign-discounted appointments β€” Appointments booked with a campaign discount applied.

Important: The demand forecast is derived solely from your location's own booking history. It does not incorporate any other center's data, staff schedules, pricing, or individual guest identity.


Review what the schedule-generation stage uses

Once the demand forecast is available, the schedule-generation stage combines it with the following inputs to produce the recommended roster:

  • Demand forecast β€” The output from Stage 1 (see above).
  • Eligible provider roster β€” The providers who are candidates for scheduling at your center.
  • Each provider's past utilization β€” How fully each provider's available hours have historically been used. Higher past utilization influences which providers are selected and protected from shift reductions.
  • Maximum working hours β€” The maximum hours each provider is permitted to work, as configured for your center.
  • Leave days β€” Days each provider has approved leave.
  • Non-working days β€” Days each provider does not work based on their roster configuration.
  • Target utilization β€” Your center's configured target, which sets the balance between scheduled supply and predicted demand. Valid values range from 50% to 100%.

Note: All providers in the eligible roster are treated as interchangeable for demand purposes. Service-specific or skill-specific provider matching does not affect which provider the system recommends for a given slot.


Understand the nightly refresh and on-demand recompute

Both the demand forecast and the recommended schedule refresh automatically each night. If circumstances change during the day β€” for example, a provider submits a leave request after the nightly run β€” a manager can trigger an on-demand recompute of the schedule.

Important: An on-demand recompute regenerates the schedule recommendations using the forecast that is already on record. It does not re-run the demand-prediction pipeline. The demand forecast itself updates only on the nightly cadence.

For instructions on triggering a recompute, see Run an on-demand recompute.

Worked example β€” Manager regenerates suggestions after a leave request

A center manager opens the Schedule Grid on a Tuesday morning and sees that the nightly run has already produced suggestions for the coming four weeks. At 10:00 AM, a provider requests Thursday off. The manager approves the leave, then triggers an on-demand recompute. The system uses the demand forecast from the previous night's run and recalculates the recommended roster to account for the provider's Thursday absence. The Thursday day state updates to reflect the revised coverage. No new demand forecast is generated; only the schedule recommendations change.


Tips

  • Review the demand forecast inputs alongside the Schedule Grid's day-level coverage summary to understand why a day is showing as Understaffed or Overstaffed β€” the mismatch almost always traces to one of the ten historical signal categories or a change in the eligible provider roster.
  • If waitlist volume in your center is growing, the forecast will capture that unmet demand and may suggest additional staffing for high-waitlist slots.
  • When a provider's leave is approved late in the day, use the on-demand recompute to refresh schedule recommendations before the next nightly run rather than waiting overnight.
  • Check your center's target utilization setting if the recommended roster consistently feels too lean or too heavy β€” that single value has a direct effect on how the schedule-generation stage balances supply against the forecast.
  • Remember that the forecast reflects your center's own booking history. If your center has recently changed its service offering or operating hours significantly, allow several weeks for historical patterns to stabilize before treating forecast levels as fully representative.

Troubleshooting

The schedule recommendations seem unrelated to actual busy periods.

The demand forecast uses historical booking activity, not the current published schedule. If your center recently changed its hours, services, or staffing model significantly, the historical patterns in the forecast may lag behind current reality. Allow several weeks of booking history to accumulate under the new model. In the interim, use the Schedule Grid's day-level coverage summary to manually identify and adjust suggestions that look misaligned.

An on-demand recompute did not reflect a provider's new leave day.

Confirm that the leave day is fully saved and approved in the provider's roster before triggering the recompute. The schedule-generation stage reads approved leave at the time the recompute runs. If leave was not yet saved, rerun the recompute after confirming the leave record is in place.

The forecast appears the same day after day and does not seem to update.

The demand forecast refreshes on a nightly cadence, not in real time. If you trigger an on-demand recompute during the day, the demand forecast used is the one produced in the previous night's run. The forecast itself will update in the next nightly cycle. If you believe the nightly refresh has not run, contact Zenoti Support.


Field-level details

This article describes the upstream inputs to AI scheduling at a conceptual level. For field-level detail on individual forecast parameters, schedule-generation configuration, or provider roster attributes, contact Zenoti Support.

Was this article helpful?
Your feedback helps us improve our documentation.