… 11 unchanged lines …
  - Batch runs are configured as **Scope** (which resources and requirements), **Goal** (objectives and constraints) and **Plan** (the combination of the two). Interactive runs pick a goal from the settings panel — Maximise Utilisation or Front-load High-Priority Work.
  - Behaviour is driven by two data settings people forget: **Optimization Method** on each booking status (Optimize, Do Not Move, Ignore — defaulting to Do Not Move) and **Priority Value** 1–100 on priorities. The agent ignores the old **Level of Importance** field entirely.
  - Bookable resources must have **Display on schedule board** set to Yes to be considered.
- - It does not support Crew, Equipment, Pool or Facility resource types, requirement groups, multiday requirements, requirement fulfilment preferences, or RSO-specific fields. Maximum three breaks per working day — four or more throws an error.
+ - It does not support Equipment, Pool or Facility resource types, requirement groups, multiday requirements, requirement fulfilment preferences, or RSO-specific fields. Maximum three breaks per working day — four or more throws an error. **Crew resources are now supported** — Microsoft removed Crew from the unsupported list in the 3 Sept 2026 documentation refresh, so a client you previously ruled out on crews is worth re-checking.
  - Available in the same Azure regions as Field Service, excluding Azure Government and China. Online only. Prompt detection has been tested in English.
+ - **Plans can now run on a recurrence** — hourly, daily, weekly or monthly, with a minimum interval of 30 minutes — as well as on demand via **Run optimization plan**. A plan still runs only one optimization request at a time, so a recurrence tighter than the run duration queues rather than parallelises.
+ - **Requirement dependencies** are an always-enforced constraint: where two requirements have a defined dependency, the agent preserves their order and you can't turn that off.
+ - A **Prefer Existing Bookings** objective is available in a custom goal, which weights the schedule towards leaving current bookings alone. It's the softer alternative to per-booking restrictions.
+ - Per-booking overrides exist too — see [[feature-soa-booking-level-restrictions]] for **Optimization behavior** and **Restrict to**, which beat both the status-level method and requirement-level constraints.
  - Results expire: older than 72 hours, or past the start of their time range, and you have to re-run.
  
  ## When to use / skip
    … 1 unchanged line …
  
  It is not a replacement for Resource Scheduling Optimization. RSO is GA, it's licensed as an add-in, it runs at proper scale overnight, and it handles resource types and scenarios the agent flatly doesn't. If a client's requirement is "optimise 400 engineers every night", the answer is still RSO. The agent's batch mode tops out at 50 resources, and that ceiling decides most conversations before any other consideration.
  
- Skip it for now on any project that can't absorb preview risk, that uses crews or facilities meaningfully, that has multiday requirements, or that has no appetite for a consumption-based cost line. And skip it if the client's booking statuses and priority values have never been curated — the agent will produce nonsense from bad configuration data and the client will blame the AI.
+ Skip it for now on any project that can't absorb preview risk, that uses facilities meaningfully, that has multiday requirements, or that has no appetite for a consumption-based cost line. And skip it if the client's booking statuses and priority values have never been curated — the agent will produce nonsense from bad configuration data and the client will blame the AI.
  
  ## Configuration decisions
  - Interactive only, batch only, or both — and if batch, who owns the scopes, goals and plans as living configuration rather than a one-off setup task.
    … 7 unchanged lines …
  - The agent runs as a team identity. If the client uses custom tables or column-level security, the **Scheduling Operations Agent Team** needs matching roles or optimisations quietly return worse results with no obvious error.
  - Booking statuses default to Do Not Move when Optimization Method is unset. A first run that "does nothing" is almost always this.
  - Level of Importance is ignored. Clients who spent a previous project tuning it will need to redo that work in Priority Value.
+ - Recurrence makes an unattended agent out of an interactive one. A recurring plan with **Apply method** set to apply automatically will rewrite the schedule board overnight with nobody watching — decide that deliberately rather than inheriting it from a plan someone set up to test.
  - Truncation is silent-ish. Exceed the limits and the agent drops lower-priority records — open requirements first — and tells you records were truncated rather than refusing to run.
  - Cost scales with resources per run, so an enthusiastic dispatcher running batch plans repeatedly is a budget event. Set capacity alerts early.
  - Suggested results expire after 72 hours or once their time range starts, so a suggestion left over lunch may need re-running.
    … 4 unchanged lines …
  - Demo interactive optimisation with a real overrun scenario. Batch mode demos badly and its limits invite the wrong comparison with RSO.
  - Do the booking status and priority data work before the first demo. Nothing kills confidence faster than an optimisation that returns the same schedule.
  - Get the Copilot Credits conversation done with whoever owns the tenant budget, in the same meeting where you agree the pilot scope.
- - Before any pilot, confirm the client's resource types. Crews and facilities are common in maintenance-heavy estates and are simply out of scope today.
+ - Before any pilot, confirm the client's resource types. Facilities, pools and equipment remain out of scope; crews came into scope in September 2026, so re-test rather than trusting an earlier assessment — and confirm it in the target tenant, because this is preview documentation moving faster than release notes.
  
  ---
  
- *Worth revisiting when the agent reaches general availability, or when the 50-resource batch ceiling moves.*
+ *Refreshed 19 Sept 2026 for plan recurrence, Crew resource support, requirement dependencies and the Prefer Existing Bookings objective. Worth revisiting when the agent reaches general availability, or when the 50-resource batch ceiling moves.*