Occupation Manual 018 / Retail shift evidence

Let AI prepare the brief. Humans still run the store.

Use AI to organize approved, minimum-necessary shift notes into a draft brief and an exception list. Do not let it select or assess people, schedule work, set a price, change inventory, reconcile cash, answer a customer, or write to a system.

Short answer: treat AI as a source-bound briefing clerk, not a retail manager. Start with one approved shift purpose, a current procedure or priority, minimum necessary non-sensitive inputs, and named decision owners. Ask AI to make a neutral brief and a list of unknowns. If any source, policy date, person-data boundary, or authority is unclear, stop at a needs_review handoff.

Keep people, customer, and operational data out of unapproved tools. Do not enter schedules, attendance, performance notes, payroll, health/accommodation information, customer information, payment/cash details, inventory system data, account information, surveillance material, or anything the organization has not approved. This local worksheet only composes text in the browser; it cannot save, send, look up, or change anything.

The useful boundary

Safe for a bounded assistant

Organize supplied approved notes, state an approved shift objective, list source/date/owner, separate an exception from a fact, turn an approved procedure into a checklist, identify missing evidence, and prepare a neutral reviewer handoff.

Always human-owned

Any decision about people; schedules; price or promotion; refund, cash, inventory, vendor, customer, safety, security, privacy, or public action; source truth; and every system record or final decision.

O*NET describes this occupation as supervising retail sales workers, including customer service and work involving inventory, cash receipts, prices, and displays. That combination makes visible sources, limited inputs, and named decision owners essential—not optional.

Build the five-part Retail Shift Brief & Exception Desk

01

Bound one shift before you draft

Name the fictional or approved shift label, operational goal, shift owner, reviewer, and out-of-scope work. Stop immediately for safety, security, privacy, cash, customer threat, suspected theft, emergency, or employment/HR concerns and use the approved human procedure.

Output: shift card
02

Bind priorities and exceptions to current approved sources

Record the procedure, merchandising priority, price/promotion rule, safety rule, or escalation rule with its version/date and owner. A model response, old note, unverified count, chat, or individual assertion is not a source of truth.

Output: source check
03

Set data and authority boundaries

Use the least information needed for a draft. Explicitly list prohibited people, customer, payment, cash, schedule, and inventory inputs. Name the human who owns each action. Unknown authority means stop.

Output: job card & gate
04

Draft for clarity, not control

Ask AI for a neutral shift brief, a source list, exception labels, missing-information flags, and the next review route. It may not select staff, propose a schedule, assess a worker, set a price, resolve an inventory discrepancy, or make a customer commitment.

Output: unsent draft brief
05

Hand off exceptions to named humans

Package sources, dates, unknowns, data boundary, and review owner. A human may approve, revise, reject, or escalate in the approved process. The worksheet status remains needs_review; it is never an instruction, record, or decision.

Output: review-only handoff

Give AI jobs, not invisible authority

Bounded roleApproved inputAI outputHuman gate
Brief clerkApproved non-sensitive shift objective and procedure extractOrdered draft brief with source labelsShift owner verifies accuracy and approves any use
Source checkerSupplied source/date/owner fieldsMissing, stale, or conflicting-evidence listNamed policy/process owner confirms current source
Gap finderShift card and exception notesQuestions, unknowns, and stop triggersManager or specialist resolves/escalates gaps
Escalation routerApproved escalation categories and human rolesSuggested route and reasonHuman confirms the category and opens approved process

Failure modes that turn a helpful brief into an unsafe control

AI treats a staffing pattern as a people decision

Scheduling, performance, discipline, pay, promotion, hiring, termination, leave, accommodation, and protected-activity questions need the organization’s qualified human process. Do not use a prompt as a scoring or selection system.

An exception becomes a silent operational change

An out-of-stock note, cash mismatch, price issue, or refund request is not permission to change a system or promise an outcome. Keep the exception visible, source-bound, and owned by a named reviewer.

Too much data goes into a prompt

Briefing convenience must not become a data export. Use approved environments, minimum necessary inputs, least privilege, and the organization’s retention/security rules.

Fluency is mistaken for a current procedure

Generative systems can produce plausible but incorrect text. Preserve the source/date/owner, label unknowns, and stop rather than smoothing uncertainty into a confident instruction.

Run the worksheet: create a review-only handoff

Try a fictional scenario first. This form composes text only in your browser. It does not save, send, call an API, access a schedule, POS, inventory, payroll, HR, customer, or security system, or control any action.

Primary source desk

Next route: use the AI Agent Workflow Builder to formalize a least-privilege job card, the Customer System Manual for customer-process boundaries, and the AI Marketing Operating System only for approved promotional-claim workflows. A tool connection never authorizes a retail or people action.