field-service-scheduling-policy-designer-query

Imported from forcedotcom/sf-skills (skills/field-service-scheduling-policy-designer-query/SKILL.md). Install upstream with npx skills add forcedotcom/sf-skills --skill field-service-scheduling-policy-designer-query. Copyright stays with the author.

Managing Sfs Scheduling Policy Designer

When to Use This Skill

Use this skill to reason about and design a complete Salesforce Field Service (SFS) scheduling policy — the policy's intent, work rules that filter candidate resources, relevance groups that scope rules and objectives, and service objectives with mathematically consistent optimization weight values. Trigger whenever the user mentions a scheduling policy, work rules, service objectives, relevance groups, In-Day Optimization, Commit Mode, field service optimization, penalty points, or wants to configure or tune a Salesforce Field Service scheduling policy. Walk the user through defining requirements (work rules), scoping (relevance groups), and trade-off questions to derive consistent weights for each objective, then emit a structured build spec that a separate data-layer skill turns into records. This skill designs the policy; it does NOT create, read, update, or delete any records.

Workflow

Salesforce Field Service – Scheduling Policy Designer (Business Logic, Condensed)

This skill helps the user design a complete Salesforce Field Service scheduling policy end to end and then hands the design off — as a structured build spec — delegating all record operations to the sfs-sobject-create skill. This skill owns what the policy should be and why; it never touches how records are created.

Interview sequence (REQUIRED): Run interviews IN THIS ORDER and make sure not to skip any question, completing each before moving to the next:

Phase 1: Scheduling Policy Interview — Collect policy settings (name...

Phase 2: Work Rule Design Interview — Ask one question...

Phase 3: Service Objective Interview — Group trade-off questions by...

Phase 4: The scheduling policy definition — name, description, In-Day...

Phase 5: Work rules — the hard filters that decide...

Phase 6: Relevance groups — optionally scope a work rule...

Phase 7: Service objectives + weights — the soft scoring that grades the surviving options. This is the deepest part of the skill

a conversational trade-off interview back-calculates weight values from the penalty math so all objectives are calibrated on a consistent scale.

Phase 8: The build spec — a structured, machine-readable summary...

Rules vs. objectives — the core mental model: Work rules are hard filters — they reject any resource or slot that violates them, and are always applied before objectives. Service objectives are soft scoring — they grade the candidates that survive the rules and never reject anyone. If a user wants "must have X," that's a work rule. If they want "prefer X," that's an objective. Rules shrink the candidate list; objectives rank what's left.

Terminology: Always call the product Field Service when talking to the user — never use the abbreviation "FSL," even though it appears in this skill's own internal notes as shorthand for the managed package.

Scope boundary — design only, no CRUD: This skill produces decisions and values, never records. It has no knowledge of Salesforce object/field API names, composite API structure, or record-creation order, and it must not attempt to create, read, update, or delete anything. When the design is complete, it emits the build spec and delegates all record operations to the sfs-sobject-create skill. If the user asks "now create it" / "deploy this," produce (or finalize) the build spec and hand off to sfs-sobject-create.

How to use this skill: If the user just wants weights, go straight to Conversation Flow (the trade-off interview) — that behavior is unchanged. If they want to design or reason about the whole policy, work through the sections in order: policy definition → work rules → relevance groups → objectives/weights → build spec.

Source of truth: The calculations and weight-derivation math in this skill are authoritative — they reflect the true internal SFS optimizer behavior and take precedence over Salesforce public documentation, which is less precise about the math. The documentation links in Live Documentation References are for background and behavioral details that may change over time — fetch them for current details, but never let them override the penalty formulas in this skill.

Defining a Scheduling Policy

A scheduling policy bundles work rules and service objectives together and tells the optimizer how to schedule. This skill decides the policy's settings and intent; the data-layer skill creates the actual record. When helping a user design one, collect these policy-level decisions first — each maps to a field the data-layer skill will populate.

How to run this section conversationally: Ask about each setting one at a time — Name, then In-Day Optimization vs. Global, then Commit Mode, then Description — never bundle multiple questions into a single message; wait for the user's answer before asking the next. Keep this section scoped strictly to the policy-level settings below — do not mention work rules, the mandatory Service Resource Availability rule, or any other rules-related framing notes here; that framing belongs only once the conversation reaches the Work Rules section that follows. This skill always builds a fully custom policy from the user's own requirements: never offer, suggest, or ask whether the user wants to start from one of Salesforce's standard starter policies or any other template (see Starting points below — background only, not a conversational option). Always call the product Field Service — never use the abbreviation "FSL" in any interaction with the user.

Core settings:

• Scheduling Policy Name — the display name for the policy. Tip: if In-Day Optimization is enabled, Salesforce recommends putting "In-Day" in the name so it's easy to identify when dispatching or optimizing.
• Description — a free-text description of the policy's intent. Good place to record the "Policy at a Glance" summary this skill generates in Step 3.
• In-Day Optimization (boolean) — when true, the policy uses in-day optimization instead of global optimization. Global runs for hours across the full horizon; in-day is time-boxed for last-minute changes (up to 5 minutes with Enhanced Scheduling and Optimization, up to 10 minutes without it) and can be triggered by dispatchers or a scheduled job.
• Commit Mode — governs what happens when a dispatcher (or a scheduling operation) changes the schedule while a global/in-day/resource optimization is already running and the two conflict. Two values:
  • Always Commit — apply the change even if it conflicts with the in-progress optimization.
  • Rollback — reject/undo the conflicting change to protect the optimization run.

Capture each of these as a value in the build spec's policy object — do not worry about how the record is stored; that is the data-layer skill's job.

Mandatory rules (every policy) — this is the one canonical statement of this rule; every other mention in this skill just points back here:

• Service Resource Availability — always include exactly one in the design and the build spec (mandatory: true), regardless of what the user asks for. The Field Service managed package does not create this rule automatically, so it must always be emitted for the data-layer skill to create. If the user gives no availability details, emit a baseline Service Resource Availability rule anyway (no breaks, no overtime/travel flags) so the policy is valid.
• Earliest Start Permitted and Due Date (Match Time rules) — do not include these in the build spec. The Field Service managed package creates them automatically when the scheduling policy is created. Do not design, emit, or ask about them; they are provisioned for you.

Starting points (background only — not a conversational option): Salesforce ships four standard starter policies — Customer First, High Intensity, Soft Boundaries, Emergency. This is background knowledge only; never suggest or offer these to the user as a jumping-off point — every policy this skill designs is built fresh from the user's own stated requirements, not modeled on a template.

For behavioral background on these settings, fetch: service.pfs_scheduling.htm (see Live Documentation References).

Work Rules

Work rules are hard filters. They refine the candidate list for a service appointment by rejecting any service resource that violates a rule. They are always applied before service objectives, and objectives only ever score the resources that survive the rules. If a requirement is "must," it's a work rule; if it's "prefer," it's an objective.

Database rules vs. Apex rules (why it matters for performance)

• Database rules filter at the SOQL-query level — disqualified resources never come back from the database, so they're cheap. They're aggregated into one query and applied in no particular order.
• Apex rules run after the initial query, iterating over every returned candidate (like a for-loop) to validate — this has a real CPU cost, compounded by objective grading that runs afterward.
• Guidance: include at least one database rule so Appointment Booking / Get Candidates start from a small candidate pool. Aim to narrow to roughly ~20 candidates with database rules before Apex rules and objectives run.

The 16 work rule types

Each row: name — engine (DB/Apex) — what it does. "DB w/ ESO" means the rule runs at the database level only when Enhanced Scheduling and Optimization is enabled, and as Apex otherwise.

Doc naming note: the docs' "Considerations" list once refers to "Enhanced Match" in the baseline-database-rules list — there is no rule type called Enhanced Match; it means Extended Match. Treat as a documentation typo.

Step 1 of building rules: define your requirements

Arrival Window Match Time rules (opt-in default pair). If the user wants appointments to honor customer arrival windows, ask them to confirm, and if they agree, emit two Match Time rules with these exact default settings — do not ask the user to fill these in, they are the standard arrival-window configuration:

Emit each as a Match Time work rule whose params carry those four values (keys serviceScheduleTimeProperty, serviceTimeOperator, serviceTimeProperty, passEmptyValues). These are separate from the mandatory Earliest Start Permitted / Due Date rules (never emitted — see Mandatory rules).

Interpreting and emitting break times (clock time vs. shift-start offset): A Service Resource Availability rule expresses breaks in one of two shapes, and this skill decides the shape and emits the values the data-layer skill needs — the data-layer skill never receives a clock time it has to convert. Choose the shape as follows:

• One fixed daily break at an absolute clock time (e.g. "30 minutes at 12:00 every day") → emit it as an absolute break: { "mode": "absolute", "startClock": "12:00", "durationMinutes": 30 }. No offset math, no working-day start needed.
• One or more breaks defined as an offset from the start of the working day (e.g. "a 30-minute lunch starting 3 hours into the shift") → emit each as an offset break, in minutes from the start of the working day: { "mode": "offset", "earliestStartOffsetMinutes": 180, "latestEndOffsetMinutes": 240, "durationMinutes": 30 }. All three of earliestStartOffsetMinutes, latestEndOffsetMinutes, and durationMinutes are mandatory for an offset break. When the user states the break in offset terms already (e.g. "3 hours after the start of day"), that offset is earliestStartOffsetMinutes (180) — do not ask for the working-day start; you don't need it.
• Breaks given as clock times but there is more than one (e.g. "15 minutes at 10:00 and 30 minutes at 12:00") → you must use the offset shape, which means converting each clock time into an offset from the start of the working day. You cannot do that without knowing when the day starts, so ask the user for the shift / working-day start, then compute earliestStartOffsetMinutes = (break start clock − day start) in minutes for each break. Never assume the day starts at midnight (that would turn "10:00" into a wrong 600-minute offset). (Only ask for the day start in this clock-time case — never when the break is already expressed as an offset.)

Always emit latestEndOffsetMinutes for every offset break — ask for it directly, do not fabricate a window. The earliest start comes from what the user stated (an offset, or a converted clock time) — do not invent ± tolerance windows around it (no "±1 hour" / "±2 hour" options). If the user only gave a break start (or a duration with no stated finish-by), ask them plainly for the latest the break may end, phrased in the same terms they used: if they gave an offset ("starts 3 hours after start of day"), ask "what is the latest it may end, as time after the start of the day?" and convert (e.g. "4 hours after start" → 240); if they gave a clock time, ask for a clock time and convert with the day start. If the user says the break is fixed/pinned with no flex, set latestEndOffsetMinutes = earliestStartOffsetMinutes + durationMinutes. Never emit an offset break missing any of the three fields, and never guess the latest-end. If a break requirement is too involved to convert reliably, ask clarifying questions or recommend the manual approach rather than guessing.

Maximum Travel from Home — establish the cap type, not the unit. When a requirement caps how far a resource may travel from home, ask the user whether the cap is by distance or by travel time — the two are configured differently and the data-layer skill needs to know which. Emit the value and the type in params: { "maxTravelFromHome": 50, "maxTravelFromHomeType": "Distance" | "Travel Time" }. Interpret the user's phrasing to set the type — "50 miles"/"50 km" ⇒ Distance, "45 minutes" ⇒ Travel Time — but do not ask for or emit a distance unit (miles vs km): it is not part of the work-rule config; the unit is governed by the org's locale/distance settings elsewhere, so the rule stores only the number. For Travel Time the value is minutes. Never emit the cap value without its type.

Step 2 of building rules: per-rule parameter questions

Once the requirements funnel (Step 1) has selected which rules to build, ask the specifics for each selected rule. The Emit line gives the exact params keys to put in the build spec — use these key names verbatim so the data-layer skill can map them to fields. Only ask about rules the user actually selected; keep the rest out of the conversation.

All rule types below have a fully-defined question → param path — the data-layer fields are confirmed against the Field Service managed-package model.

• Service Resource Availability (always present) — ask: "Can work run into overtime?" (yes/no) · "Can resources travel outside working hours to/from home?" — if yes, ask up to how many minutes of travel are allowed from home (to the first job) and to home (from the last job); treat "no limit" as 120 (Field Service's effective unlimited ceiling), and if the answer is no, both are 0 · and the breaks interview (see Interpreting and emitting break times above). Emit: { "enableOvertime": true|false, "travelFromHomeMinutes": <number>, "travelToHomeMinutes": <number>, "breaks": [ … ] }. travelFromHomeMinutes/travelToHomeMinutes are minute values, not booleans — 0 = no travel outside working hours, 120 = effectively unlimited, or the specific cap the user gives. If the user gives no availability detail, emit a baseline rule (enableOvertime: false, travel-minute keys and breaks omitted).
• Match Time (arrival windows) — ask only "Should appointments honor customer arrival windows?" If yes, emit the two default rules from the Arrival Window table above (no further questions).
• Match Skills — gate first, don't assume it's needed. Ask: "Do work orders require resources with relevant skills or specific proficiency levels?" If no, do not emit a Match Skills rule at all. If yes, emit the rule and ask the follow-up: "Should the resource also meet a minimum skill level (proficiency), or is simply having the skill enough?" Emit: { "matchSkillLevel": true|false }. Do not ask about skill-type AND/OR logic — this skill deliberately does not configure skillTypeLogic.
• Match Fields — ask: "Which Service Appointment field must match which Service Resource field, and with what operator?" Emit: { "serviceProperty": "<appointment field API name>", "resourceProperty": "<resource field API name>", "booleanOperator": "=" }. The operator is one of the symbols =, >=, <=, >, < (default =). (These field names are the customer's own fields — pass them through as given.)
• Match Boolean — ask: "Which checkbox field on the resource must be true (or false)?" Emit: { "resourceProperty": "<resource field API name>", "value": true|false }. Remind the user there is a max of 5 Match Boolean rules per policy.
• Maximum Travel from Home — see the cap-type note above. Emit: { "maxTravelFromHome": <number>, "maxTravelFromHomeType": "Distance"|"Travel Time" }.
• Working Territories — ask: "Restrict resources to their primary and secondary service territory memberships?" This is the only territory-scoping rule the interview emits — there is no separate Match Territory question (by design, no interview path produces a standalone Match Territory rule). No user parameters — selecting the rule is enough. Emit: { "workingLocationEnablePrimary": true } (the standard fixed value; the data-layer skill maps it to {ns}Working_Location_Enable_Primary__c). (If a user explicitly needs primary/relocation-only scoping — the classic Match Territory behavior — advise them it must be configured manually; the interview does not emit it.)
• Service Crew Resources Availability — first ask the gating question: "Do you need to ensure crew-type resources get scheduled?" If no, do not emit this rule at all — there is nothing to create. If yes, emit the rule and ask its two parameters:
  • Consider Service Crew Membership (yes/no) — "Should the rule evaluate individual crew members' availability and skills, not just the crew record?" → considerCrewMembership.
  • Maximum Additional Service Resources (optional number) — "What is the maximum number of extra resources beyond the base crew that can be pulled in to meet the minimum crew size?" → maxAdditionalResources.

Emit: { "considerCrewMembership": true|false, "maxAdditionalResources": <number> } (omit maxAdditionalResources if the user leaves it blank).

• Extended Match (advisory only — the data-layer skill will not build this) — Extended Match needs a custom junction object plus fields that no skill in this pair creates (that is schema DDL the user must own). So do not emit it as a normal work rule with mappable params. Instead, when a requirement calls for custom-criteria matching (e.g. serviceable postal codes, product lines), advise the user what to set up, and record it under prerequisites — do not put it in workRules[] for the data layer to create. The setup guidance to give the user: (1) create a junction object linking Service Resource to the matched object, with exactly two Master-Detail relationships (one to Service Resource, one to the matched object) — the packaged trigger requires exactly two or the rule fails; (2) a Service Appointment field of type Lookup that drives the match; (3) a reference field on the junction matched against it. Tell the user that once those exist, the Extended Match work rule must be created and configured manually in Setup (Field Service Settings → Scheduling → Work Rules), because pointing a rule at objects/fields that are created outside this design is beyond what the data-layer skill does. Capture the intended objects/fields in prerequisites as advisory text; do not emit a params mapping.
• Count Rule — ask: "Is there a maximum number or duration of appointments per resource per day?" If yes, have the user describe the limit in plain terms, then classify it into three values:
  • countBy — one of "Appointments" (a count of jobs, e.g. "no more than 8 jobs per tech per day"), "Hours" (a duration cap, e.g. "≤ 6 working hours of appointments per day"), or "Custom" (sum of a custom numeric field on the Service Appointment, e.g. "≤ 100 units on a truck").
  • maxValue — the numeric cap (for Hours, convert the limit to hours).
  • fieldHint — only for Custom (and optionally Hours): the Service Appointment field to sum. Set null for a plain Appointments count.

Emit: { "countBy": "Appointments"|"Hours"|"Custom", "maxValue": <number>, "fieldHint": "<SA field API name>"|null }. When countBy is Custom (or Hours against a specific field), add a prerequisite note: "Confirm the summed field <fieldHint> exists on Service Appointment." The Time Resolution is always Daily and the counted object is always Service Appointment — the data-layer skill sets both automatically, so do not ask about them. Reminder: up to 10 custom-field count rules per policy.

Selection-only rules — the rule record itself has no config fields; the behavior comes from data on other objects. For each: ask a plain gate question, emit {} if yes (or skip entirely if no, where noted), and record the referenced data as a prerequisite.

• Required Resources / Excluded Resources — ask: "Should the policy force assignment to a resource marked Required (or prevent one marked Excluded) on the work order / WOLI?" Emit {}. These key off Resource Preference records on the work order — note that as a prerequisite (the data must exist).
• Work Capacity (ESO only) — ask: "Cap work at what limit (Hours or Percentage of capacity), for which territory and which work-type/appointment attribute?" The limits themselves live on standard WorkCapacityLimit records per territory (a prerequisite set up outside the policy); adding the rule to the policy just enables enforcement. Capture the intent and note the prerequisite; the rule record itself has no config fields.
• Service Appointment Visiting Hours — gate only; do not ask for the windows. Ask: "Should appointments be restricted to customer visiting hours?" (yes/no). The windows are per-account data, not a single policy-wide value: they live on an OperatingHours record referenced from the Work Order (populated from the Account), so there is nothing to collect at design time. If yes, emit {} and record a generic prerequisite ("each account must have its visiting/operating hours populated; the Work Order's Visiting Hours must resolve from the account"). Never ask the user to state the allowed windows — they differ per account. (No relevance-group scoping.)
• TimeSlot Designated Work — gate only. Ask: "Do you need to reserve specific time slots / shifts for particular types of work?" (yes/no). Phrase it as "types of work," not "Work Type" — Work Type is a specific Salesforce object, and saying it implies the reservation can only key off that object, when the designation is broader. The slot↔work reservation lives on standard TimeSlot data (a prerequisite), so there is nothing to collect at design time. If yes, emit {} and record a generic prerequisite ("time slots must be configured to designate the intended work"). (No relevance-group scoping.) Do not ask the user to enumerate slot→work mappings — that is data setup.

For per-rule configuration fields and gotchas, fetch the specific rule sub-page under service.pfs_optimization_theory_work_rules.htm (see Live Documentation References).

Relevance Groups

A relevance group scopes a single work rule or service objective so it applies only to certain appointments or resources, instead of the whole policy. This lets one policy hold different logic for different work or resource types — e.g., different break/travel limits for part-time vs. full-time employees, or expedited scheduling for high-value accounts.

How it works, and the division of labor

A relevance group is driven by a Boolean (true/false) field. Every standard or custom Boolean field on the Service Appointment and Service Territory Member objects is selectable in the rule/objective's Relevance Group dropdown. On a work rule or objective, you pick the limiting Boolean field; the rule/objective then applies only to records where that field is true. The two bases are just the two objects the Boolean can live on:

• Service Appointment basis — scopes by the appointment being evaluated (e.g., a formula checkbox "Platinum Account" that's true when the related account tier is Platinum). Use to target appointment types.
• Service Territory Member basis — scopes by the service territory member of the resource being evaluated (e.g., "Part-Time" / "Full-Time" checkboxes). Use to target resource populations. Only primary and relocation memberships are supported — not secondary.

In the design: decide the basis (Service Appointment vs. Service Territory Member) and the name of the Boolean field that identifies the subset, then record both on the rule/objective in the build spec (relevanceGroup: { basis, booleanField }). This skill only decides which field scopes which rule/objective — wiring the group onto the record is the data-layer skill's concern.

The Boolean field itself is a prerequisite the user must own, not something either skill in this pair creates: creating the field is a schema change (DDL), and setting its true/false values on records needs a flow, formula, trigger, or data load. So advise the user to, before deploying: (1) create one custom Boolean field on the appropriate object (Service Territory Member for a resource subset, Service Appointment for a work subset) per subset; (2) populate it true for exactly the records in that subset (via flow/formula/import), keeping subsets mutually exclusive where the rule type requires it. Then reference that field by name in the build spec so the data-layer skill can attach it as the relevance group.

Worked example (the canonical pattern)

To apply a Match Boolean rule (or any scoping) to only certain appointment types: identify (or have the user create) a Boolean field on the Service Appointment that is true for those appointments, and name it as the rule's relevance group in the build spec. The rule then applies only to appointments where the field is true. The same pattern with a Service Territory Member Boolean (e.g., Part-Time vs. Full-Time) lets you design two copies of the Maximum Travel from Home rule with different limits — one scoped to part-timers, one to full-timers.

For objectives, the classic example is combining ASAP with a relevance group: a formula checkbox "Platinum Account" on the appointment, an ASAP objective named "Expedite Platinum Accounts" scoped to it with a high weight, so platinum jobs schedule sooner (accepting more travel). Give scoped high-priority objectives a clearly higher weight, or the engine may prefer not to schedule the appointment at all because of its ASAP penalty.

Scoping any rule or objective to a subset (the general rule)

This applies to every work rule and every service objective — not just breaks or availability. Whenever a request says a rule or objective should apply to only certain resources or only certain work, rather than the whole policy, the mechanism is a relevance group. Watch for this signal and reach for it every time:

• "…applies to certain types of resources" (e.g. "resources in France", "part-time techs", "senior engineers", "the install crew") → create a Service Territory Member relevance group: a Boolean field on Service Territory Member that is true for exactly those resources, selected as the rule/objective's relevance group.
• "…applies to certain types of work" (e.g. "installation jobs", "platinum-account appointments", "emergency work orders", "jobs over 2 hours") → create a Service Appointment relevance group: a Boolean field on Service Appointment that is true for exactly those appointments, selected as the rule/objective's relevance group.

Always check whether the specific rule/objective supports the basis you need. Not every rule/objective can be scoped by Service Territory Member, and not every one can be scoped by Service Appointment. Before recommending a relevance group, consult the relevance-group documentation's support matrix (service.pfs_relevance_groups.htm) for that exact rule or objective, and consider only Enhanced Scheduling & Optimization (ES&O) support — ignore the legacy/non-Enhanced columns. If the needed basis isn't supported for that rule/objective under ES&O, say so and suggest the closest supported alternative instead of inventing one.

When two subsets each need a different configuration of the same rule/objective, design one instance per subset — never one merged instance. A single rule/objective instance applies to every record it covers, so you cannot express "France gets pattern A, everyone else gets pattern B" in one instance. Instead specify one instance per subset, each scoped by its own distinct Boolean field, and keep the groups mutually exclusive (for single-coverage rule types like Service Resource Availability, an overlap throws an error — see Key rules and gotchas). Example: "Resources in France get a 15-minute break at 10 AM and a lunch between 12 and 2; all other resources get a 30-minute break between 12 and 3" needs two Service Resource Availability rules — "Availability – France" scoped to a Break_Group_France__c STM checkbox, "Availability – Standard" scoped to a Break_Group_Standard__c STM checkbox — each carrying only its own group's breaks, with every resource landing in exactly one group.

When the scenario is genuinely too complex, recommend a manual design. If a request combines multiple subsets, ambiguous values, and relevance-group fields that don't exist yet, don't force a single merged instance into the spec. Walk the user through the relevance-group design above and let them refine it (or spec only the parts that are unambiguous), rather than emitting a plausible-looking but wrong merged rule/objective. Prefer asking clarifying questions (which subset? resources or work? what Boolean field identifies each? is that basis supported for this rule/objective under ES&O?) over guessing.

Key rules and gotchas

• Relevance groups must be mutually exclusive. If two rules with relevance groups overlap, the more restrictive one wins — and for Service Resource Availability, an overlap throws an error. Each resource must be covered by exactly one Service Resource Availability rule.
• Additive rule types — can appear multiple times and legitimately cover the same resources/appointments: Count Rule, Extended Match, Match Boolean, Match Fields, Match Time.
• Single-coverage rule types — a resource/appointment must be covered by at most one instance at a time: Match Skills, Match Territory, Maximum Travel from Home, Required Resources, Service Crew Resource Availability, Service Resource Availability, Working Territories. (Also: don't cover a resource by both Match Territory and Working Territories at once.)
• All-or-nothing rules — no relevance groups: TimeSlot Designated Work and Service Appointment Visiting Hours. Also Work Capacity doesn't support relevance groups.
• Objectives can be repeated with relevance groups too, but a record must not meet the criteria for two objectives of the same type at once — except Resource Priority, which can legitimately apply twice to the same appointments if each instance points to a different resource priority field (e.g., "Primary Priority" and "Secondary Priority"/"Tenure").
• Support for whether a rule/objective can be scoped by appointment vs. territory member differs between the Enhanced and non-Enhanced engines — verify the support tables when scoping. Group Nearby and Same Site objectives do not support relevance groups at all.

For the full support matrices and setup steps, fetch: service.pfs_relevance_groups.htm (see Live Documentation References).

Background: How SFS Scoring Works

The optimizer assigns penalty points to each candidate schedule. Lower total penalty = better schedule. Each service objective contributes penalty points based on its weight and its own scale (the worst-case scenario for that objective).

Minimize Travel weight is the anchor — its value is set by the user and all other weights are derived relative to it.

The general formula, common to every objective below (stated once here so it isn't re-derived nine times):

penaltyPerViolation = max( 1, roundingFn( (1000 × weight) / scale ) ) × finalMultiplier
total_penalty       = ceil( violations / granularity ) × penaltyPerViolation

Every objective applies the same ×1000 internal multiplier, then its own scale, rounding function, and final multiplier. Two consequences follow, and are not repeated under each objective below:

• The ×1000 multiplier is uniform, so it cancels out of every crossover/derivation equation between two objectives — this is why the simple continuous approximations (penalty ≈ units × weight/scale) used throughout the trade-off interview stay valid, and why every "derive weight_X from weight_Y" formula below is clean of any ×1000 term. Where the rounding function is not exact for integer weights (ASAP's round, Skill Level/Preference's roundInt against a scale that doesn't divide evenly), a small amount of drift is possible — flagged per objective under Precision notes where it matters.
• The max(1, …) floor guarantees every included objective has some effect even at a very low weight. Each objective's Precision notes flag the specific weight range where this floor actually binds.

The one exception to the uniform multiplier is Same Site, whose final multiplier (×0.01) nets to an effective ×10 rather than ×1000 — called out explicitly under its formula below, because it's the one place the "just divide by the other objective's rate" shortcut doesn't apply without adjustment.

Penalty Formulas by Objective

Use these formulas throughout the conversation to show math and back-calculate weights. Each objective states its scale, rounding function, and final multiplier, then its worked example and precision caveats — the general mechanics (×1000 multiplier, max(1,…) floor, why derivations stay valid) are covered once above and not repeated here.

Minimize Travel (anchor)

• Scale: 120 minutes (2 hours = default MaxGrade__c). Rounding: round5 (5 decimal places). Final multiplier: × 1/60 (converts the per-minute rate to per-second).

  penaltyPerViolation_travel = max( 1, round5( (1000 × weight_travel) / 120 ) ) × (1/60)

  With weight_travel = 1000: round5(1000×1000/120) = 8333.33333 → × 1/60 = 138.88889 pts/second.

• Penalty for X seconds of travel: travel_penalty(X_seconds) = ceil( X_seconds / 1 ) × penaltyPerViolation_travel (whole-second granularity — unlike ASAP's whole-minute granularity).

Example — 30 minutes (1800 seconds) of travel: 1800 × 138.88889 = 250,000 pts. At the scale boundary, 2 hours (7200 seconds) = 7200 × 138.88889 = 1,000,000 pts.

• Continuous approximation (for weight derivation): travel_penalty(X_minutes) ≈ X × (weight_travel / 120) — with weight_travel = 1000, ≈ 8.333 pts/min.

• Precision notes: round5 preserves much higher precision than ASAP's integer round. Travel is per-second; Overtime (same scale) is per-minute — a 61-second trip costs more than a 60-second one, unlike Overtime. The max(1,…) floor only binds below weight ≈0.12, irrelevant since Travel is fixed at 1000.

ASAP

• Scale: 43,200 minutes (30 days = 2,592,000 seconds). Rounding: round (integer).

  penaltyPerViolation_asap = max( 1, round( (1000 × weight_asap) / 43200 ) )

• Penalty for X seconds in the future: asap_penalty(X_seconds) = ceil( X_seconds / 60 ) × penaltyPerViolation_asap (whole-minute granularity).
• Continuous approximation: asap_penalty(Y_minutes) ≈ Y × (weight_asap / 43200).
• Precision notes: weights below ~22 are dead — round(1000×21/43200) = round(0.486) = 0, clamped to 1 by the floor, so any weight 1–21 produces the identical 1 pt/min rate. Above that, drift is small: weight 250 → round(250000/43200) = 6 → effective weight 6×43200/1000 = 259.2 (3.7% drift). Validate final weights with the integer formula when precision matters.

Same Site

• Scale: 1 (binary — kept together or split). Rounding: round5. Final multiplier: × 0.01 — this is the exception flagged above: it nets to an effective ×10, not ×1000.

  penaltyPerViolation_same_site = max( 1, round5( (1000 × weight_same_site) / 1 ) ) × 0.01

  With weight_same_site = 50: round5(50000) = 50000 → × 0.01 = 500 pts/violation.
• Total penalty: same_site_penalty = ceil( X_violations / 1 ) × penaltyPerViolation_same_site. Example — 1 split (w=50): 1 × 500 = 500 pts.
• Equivalence to the old conceptual formula: New: w × 10 × X vs. Old: w × X — the new internal formula is 10× the old one (not 1000×, because of the ×0.01 above). Compared internally against Travel (w=1000 → 138.89 pts/sec) or against ASAP, the crossover math still reduces to the same continuous form because both sides carry the same ×1000-then-scale pattern:

  same_site_penalty = asap_penalty(D_equiv)  →  weight_same_site = D_equiv × (weight_asap / 43200)

• Precision notes: round5 is exact for integer weights, so no meaningful precision loss. The effective multiplier is ×10 here — remember this is the one objective where it differs from the ×1000 baseline. Flat per-event penalty regardless of any time dimension.

Minimize Overtime

• Scale: 120 minutes (same as Travel). Rounding: round5. Final multiplier: × 1.0 (per-minute — no further division, unlike Travel's ÷60).

  penaltyPerViolation_overtime = max( 1, round5( (1000 × weight_overtime) / 120 ) ) × 1.0

  With weight_overtime = 1000: 8333.33333 pts/minute.
• Penalty for X seconds of overtime: overtime_penalty(X_seconds) = ceil( X_seconds / 60 ) × penaltyPerViolation_overtime. Example — 30 minutes (1800 seconds): ceil(1800/60) × 8333.33333 = 30 × 8333.33333 = 250,000 pts.
• Continuous approximation: overtime_penalty(Z_minutes) ≈ Z × (weight_overtime / 120).
• Precision notes: shares Travel's scale and round5 precision — the rates are directly comparable — but Overtime groups into whole minutes (ceil(seconds/60)) rather than Travel's whole seconds, so a 61-second block costs the same as a 120-second block (both = 2 minutes). Floor binds below weight ≈0.12, unlikely in practice.

Preferred Resource

• Scale: 1 (binary — honored or violated, no partial credit). Rounding: roundInt. Final multiplier: × 1.0.

  penaltyPerViolation_preferred = max( 1, roundInt( 100 × 10.0 × weight_preferred ) ) × 1.0

  (100 × 10.0 = 1000 — the same internal multiplier, just decomposed.) With weight_preferred = 375: roundInt(375000) = 375,000 pts/violation.
• Total penalty: preferred_penalty = ceil( X_violations / 1 ) × penaltyPerViolation_preferred. Example — 1 violation (w=375): 375,000 pts.
• Continuous approximation: exact for integer weights (roundInt(1000×w) = 1000×w, zero rounding error): preferred_resource_penalty = X_violations × 1000 × weight_preferred.
• Precision notes: zero rounding error for integer weights. Floor only matters below weight 0.001, irrelevant in practice. Flat per-event penalty — no time-based granularity.

Resource Priority

• Scale: 10 (priority values 0–10; 0 = best candidate, 10 = lowest priority, null = lowest). No rounding function — raw decimal.

  penaltyPerViolation_resource_priority = max( 1, (weight_resource_priority / 10.0) × 1000 )

  With weight_resource_priority = 1875: (1875/10.0) × 1000 = 187,500 pts per priority point.
• Total penalty for a resource with priority P: resource_priority_penalty = ceil( P / 1 ) × penaltyPerViolation_resource_priority — P is used directly as the violation count (the ÷10 scale division is already baked into the rate). Example — P=5 (w=1875): 5 × 187,500 = 937,500 pts. P=0 (best): 0 pts. P=10 (worst): 10 × 187,500 = 1,875,000 pts.
• Equivalence to old formula: New: P × (w/10) × 1000 = Old: w × (P/10), identical up to the uniform ×1000 — the ÷10 is just applied at rate-building time instead of per-resource at evaluation time.
• Derivation formula: weight_resource_priority = T_equiv × weight_travel / (12 × (P_low - P_high)).
• Precision notes: full floating-point precision, no rounding. Floor only matters below weight 0.01, irrelevant. Penalty scales linearly — priority 5 incurs 50% of the maximum penalty.

Skill Level

• Scale: 10 (skill level values on a 10-point scale). Rounding: roundInt.

  penaltyPerViolation_skill_level = max( 1, roundInt( (1000 × weight_skill_level) / 10 ) )

  With weight_skill_level = 63: roundInt(63000/10) = 6300 pts per skill-level point.
• Total penalty for skill level S: skill_level_penalty = ceil( S / 1 ) × penaltyPerViolation_skill_level — S is used directly as the violation count. Example — S=4 (w=63): 4 × 6300 = 25,200 pts. S=8 (w=63): 8 × 6300 = 50,400 pts.
• Relationship to conceptual formula: S × weight becomes S × (1000 × weight)/10 internally; for integer weights (1000×w)/10 is exact, so roundInt introduces zero error.
• Derivation formula: weight_skill_level = T_equiv × weight_travel / (120 × (S_high - S_low)).
• Mode changes interpretation, not the formula: the formula is identical regardless of mode. Least Qualified — normal logic, lower penalty = preferred (S=4's 25,200 beats S=8's 50,400). Most Qualified — SFS inverts the preference for this objective only, so higher penalty = preferred (S=8's 50,400 beats S=4's 25,200 despite the higher score). If multiple skill requirements exist, SFS averages the skill level scores across them. If the appointment has no skill requirements, Skill Level has no impact. Ask the user which mode they're using before deriving the weight — the trade-off framing differs.
• Precision notes: zero rounding error for integer weights (exact division by 10 into 1000×w). Floor negligible below weight ≈0.01. Scales linearly with S.

Skill Preference

• Scale: 10 (skill priority values 1–10; 1 = highest preference, 10 = lowest; null defaults to 10). Rounding: roundInt.

  penaltyPerViolation_skill_preference = max( 1, roundInt( (1000 × weight_skill_preference) / 10 ) )

  With weight_skill_preference = 750: roundInt(750000/10) = 75,000 pts per skill-priority point.
• Total penalty for skill priority SP: skill_preference_penalty = ceil( SP / 1 ) × penaltyPerViolation_skill_preference. Example — SP=1 (w=750): 75,000 pts. SP=6: 450,000 pts. SP=10: 750,000 pts.
• Relationship to conceptual formula: weight × (SP/10) becomes SP × (1000×weight)/10 internally — the ÷10 that was per-skill (SP/10) at evaluation time is now baked into the rate, with SP used undivided as the violation count. Algebraically equivalent.
• Derivation formula: weight_skill_preference = T_equiv × weight_travel / (12 × (SP_low_pref - SP_high_pref)).
• Only applies when skills share the same Skill Type and the Match Skills work rule uses "At Least One Skill Matches (OR)" — no effect under AND matching. A blank Skill Priority defaults to 10.
• Precision notes: zero rounding error for integer weights. Floor negligible. Scales linearly with SP.

Group Nearby

• Scale: 1 (binary — in cluster or not). Rounding: round5. Final multiplier: × 1.0 — the full ×1000 internal space is preserved here (unlike Same Site's ×0.01).

  penaltyPerViolation_group_nearby = max( 1, round5( (1000 × weight_group_nearby) / 1 ) ) × 1.0

  With weight_group_nearby = 167: round5(167000) = 167,000 pts/violation.
• Total penalty: group_nearby_penalty = ceil( X_violations / 1 ) × penaltyPerViolation_group_nearby. Example — 1 violation (w=167): 167,000 pts.
• Equivalence to old formula: New: w × 1000 × X vs. Old: w × X — exactly 1000× the old formula (contrast with Same Site's 10×, since Group Nearby applies no ×0.01 adjustment).
• Derivation formula: weight_group_nearby = T_equiv × (weight_travel / 120).
• Precision notes: negligible precision loss. Flat per-event penalty — each appointment outside its cluster costs the same regardless of how far outside.

Minimize Gaps

• Purpose: Reduces idle time between a resource's consecutive appointments within a shift — a soft objective that's penalized for idle stretches but won't reject an otherwise-good schedule outright.
• Scale: 1 (each qualifying idle gap = 1 violation, counted per route/shift). Configurable minimum gap duration: between 30 minutes and 24 hours — gaps shorter than the minimum aren't counted. Rounding: round (no scale divisor — effective scale = 1).

  penaltyPerViolation_gaps = max( 1, round( 1000 × weight_gaps ) )

  With weight_gaps = 50: round(50000) = 50,000 pts per gap violation.
• Total penalty: gaps_penalty = Σ over all routes/shifts [ clump_counter(route) × penaltyPerViolation_gaps ] — clump_counter(route) counts idle gaps exceeding the configured minimum for that resource's shift, summed across all routes. Example — one resource with 2 qualifying gaps (w=50): 2 × 50,000 = 100,000 pts. Across a schedule, 3 resources with 1/2/1 gaps: (1+2+1) × 50,000 = 200,000 pts.
• Continuous approximation: exact for integer weights (round(1000×w) = 1000×w): gaps_penalty = clump_count × 1000 × weight_gaps.
• Precision notes: zero rounding error for integer weights. Floor negligible below weight 0.001. Penalty is per gap per route — 3 qualifying gaps on one resource contribute 3× the rate. The minimum-gap-duration threshold is a config parameter, not part of the weight calculation.

Conversation Flow

Two global rules (stated once, govern everything below)

• Round every derived weight up to the next whole number — ceiling, not nearest. Salesforce Field Service scheduling policies do not accept decimal weight values. This isn't repeated after each formula below; exceptions (low-weight floor warnings for ASAP and Minimize Gaps) are called out explicitly where they apply.
• Never compare raw weight values across objectives to infer priority — always compute and compare penalty rates per unit instead (see Step 3), since each objective has its own scale.

Step 1: Objective Selection

First explain the concept to the user: service objectives are soft scoring criteria that grade candidates who already survived the work rules — unlike work rules, objectives never reject anyone, they just influence which eligible candidate the optimizer prefers. As part of this same explanation, tell the user that Minimize Travel is always included automatically as the anchor objective, with a fixed weight of 1000 that every other objective's weight gets derived against through trade-off math — they don't need to select it.

Then present and ask about the remaining nine objectives one category at a time, in this fixed order, waiting for the user's selection before moving to the next category — never present all nine at once:

1. Customer-Experience Objectives — present these three and ask which (if any) the user wants:

• ASAP — Serve customers as early as possible
• Same Site — If two jobs are at the same place, do them back-to-back
• Group Nearby — Cluster jobs that are geographically close

2. Cost / Efficiency Objectives — present these three (note that Minimize Travel is already included automatically) and ask which of the remaining two the user wants:

• Minimize Travel (already included automatically — anchor, fixed weight 1000)
• Minimize Gaps — Keep technicians continuously busy
• Minimize Overtime — Avoid paying overtime

3. Workforce / Assignment-Quality Objectives — present these four and ask which (if any) the user wants:

• Preferred Resource — Use the preferred/named technician when possible
• Resource Priority — Prefer higher-priority resources (e.g. staff over contractors)
• Skill Level — Match the right level of expertise to the job
• Skill Preference — Honor preference rankings within a skill type (e.g. language preference)

Per-objective follow-up questions — ask these right after the category they belong to is answered, before moving to the next category. These capture configuration facts (not weight math), so they belong at selection time rather than being deferred to Step 2:

• Minimize Travel (always — ask once, at the end of the Cost/Efficiency category question since that's where Travel is called out as already included): "Should Minimize Travel also count the legs to and from a resource's home base — the drive from home to the first job, and from the last job back home — or should those legs be excluded from scoring?" Capture as two independent flags, excludeTravelFromHome and excludeTravelToHome (true = that leg is excluded). Default both to false (both legs included) if the user has no preference.
• Same Site (only if selected, ask at the end of the Customer-Experience category question): "Should Same Site treat two appointments as the same site only when they share the exact same latitude/longitude — useful for campuses or farms with no connecting roads between buildings — or use the default grouping (appointments within about one second of travel time of each other)?" Capture as useExactLocation (true = exact lat/long only). Default false (standard proximity grouping) if the user has no preference.
• Minimize Gaps (only if selected, ask at the end of the Cost/Efficiency category question): ask the user for the minimum idle duration their company counts as a gap (between 30 minutes and 24 hours; anything shorter isn't penalized) — see Minimize Gaps Trade-Off below for the full framing/example. Capture as params.minGapMinutes; default to 30 if the user gives no value. Asking it here, before the trade-off question, keeps "what counts as a gap" separate from "how hard should the optimizer work to close one."

Once all three categories have been asked and their follow-ups captured, move to Step 2 for the weight-deriving trade-off questions.

Step 2: Trade-Off Interview

For each selected objective (other than Minimize Travel), ask a trade-off question. The goal is to find the crossover point where the user considers the two options equally acceptable — that equivalence is what lets you calculate the weight.

After the user answers, solve for the unknown weight by setting the two penalty expressions equal, then round up per the global rules above. Always show the math.

Use the question templates below. Adapt the specific numbers in the questions to feel natural — you can adjust the example values if the user seems confused, but keep the math consistent.

Offer concrete preset answers alongside the open question. After asking the question template, give the user a small set of ready-made crossover points to choose from — plain numeric examples spanning a light/medium/strong preference — so they can just pick one instead of having to invent numbers from scratch. Always also invite them to describe their own exact trade-off in their own words if none of the presets fit; either path feeds the same math below. Presets are a convenience for answering, not a separate calculation path.

The sanity check, once: after deriving each weight, the trivial verification is to plug it back into both sides of the equality you just solved — they should match. That check is implied throughout and is not re-shown per objective below unless it surfaces something non-obvious, like SFS's integer-rounding drift (flagged explicitly for ASAP and Minimize Gaps) or a low-weight floor warning.

ASAP Trade-Off

Question template:

"If you could schedule an appointment right now but it would add [X] minutes of travel, versus scheduling it [Y] hours from now with no extra travel — at what point would those feel roughly equivalent to you?"

Example: "Would you prefer scheduling now with 30 extra minutes of travel, or scheduling 24 hours from now with no extra travel?"

Preset options to offer (pick the crossover that feels closest, or describe your own): "15 min travel ≈ 12 hr delay" (mild preference for promptness), "30 min travel ≈ 24 hr delay" (moderate — the example above), "60 min travel ≈ 48 hr delay" (strong preference for promptness).

Math (once user gives their crossover X minutes travel, Y hours delay):

travel_penalty(X) = asap_penalty(Y × 60)

X × (weight_travel / 120) = (Y × 60) × (weight_asap / 43200)

Solve for weight_asap:
weight_asap = X × (weight_travel / 120) × (43200 / (Y × 60))
            = X × weight_travel × 6 / Y

Note: with weight_travel = 1000 this simplifies to weight_asap = X × 6000 / Y.

Integer-formula validation (verify after deriving):

After calculating weight_asap, confirm the actual integer penalty rate SFS will use:

penaltyPerViolation = max( 1, round( 1000 × weight_asap / 43200 ) )
effective_weight    = penaltyPerViolation × 43200 / 1000

If effective_weight differs meaningfully from the derived weight_asap (typically >5% drif

Truncated - read the full file at https://github.com/forcedotcom/sf-skills/blob/HEAD/skills/field-service-scheduling-policy-designer-query/SKILL.md.