← AI in physical storesC2 · LESSON 6 OF 7

C2, Lesson 6: Store Labor and Scheduling: Matching People to Customers

Tags: Sourced / General / Our view / Illustrative. About 12 minutes.

1. The story: a barista's schedule, August 2014

On August 13, 2014 the New York Times published a report on Jannette Navarro, a 22-year-old single mother working at Starbucks, whose hours were set by scheduling software. The Times said she rarely learned her schedule more than three days before the workweek began. The next day, Starbucks executive Cliff Burrows told employees in an email that the company would upgrade its scheduling software to make shifts more consistent, would curb back-to-back closing and opening shifts (the NYT's word; the Seattle Times quotes the email as saying employees would "never" be required to), and would post schedules a week in advance ("no exceptions," per the Seattle Times). The NYT put the workforce at 130,000 baristas; Starbucks said most US employees were part-timers working 20 to 30 hours a week. [Sourced: New York Times, Aug 13-15, 2014; Seattle Times, Aug 14, 2014] The software did what it was told: fit labor to customer traffic as tightly as possible. It treated people's time as a fully flexible input. Cities and states later wrote rules about it. [Our view]

2. The one idea

AI scheduling does two things: it forecasts how many customers will arrive each hour, and it builds shifts to match. The forecast is a technical task. The schedule is a human one: people have children, classes and commutes. A good system optimizes both cost and predictability. [General]

3. The kitchen-table version

You run the stall on a market Saturday. If you know the lunch rush hits at noon, you want three helpers at noon and one at 3 pm. But those helpers need to know on Monday, not Friday night, or they quit. A forecast tells you the rush. Fairness decides how you use it. [General]

4. How it works

  • Forecast: past sales, traffic counters or camera counts, weather, holidays, promotions, local events. Output: expected transactions per hour. [General]
  • Task model: how many person-minutes each task needs (checkout, restock, online picking, cleaning).
  • Schedule builder: assigns shifts under rules (availability, minimum rest, labor law, budget).
  • Live adjustment: sends people home early or calls them in. This is the part that causes the most harm. [Our view]

5. Worked example: flat versus matched staffing

All numbers [Illustrative], computed by script. A 12-hour day. Expected transactions per hour: 20, 30, 45, 60, 50, 35, 30, 40, 70, 90, 60, 30. A cashier handles 15 an hour; wage $20.

  • Flat: 5 cashiers all day = 60 staff-hours = $1,200. It leaves 21 staff-hours idle in slow hours and still leaves 15 transactions unserved in the two peak hours.
  • Matched (rounded up each hour): 2, 2, 3, 4, 4, 3, 2, 3, 5, 6, 4, 2 = 40 staff-hours = $800. Saves $400 a day. But matched has no slack. If demand runs 20% above forecast in every hour, about 81 transactions a day go unserved. A forecast is a guess, so keep a buffer or a floating person. Add the human cost: with 30 staff, a 10-point rise in annual turnover at a replacement cost of $3,000 each is $9,000 a year, which can erase a small scheduling saving. [Our view] (Replacement cost is an assumption; measure your own.)

6. Rules to know

  • New York City's Fair Workweek Law for retail employers: give schedules 72 hours before the first shift, no on-call shifts, no cancelling a shift with under 72 hours' notice, and no requiring work with under 72 hours' notice unless the employee agrees. It took effect November 26, 2017 (for retail, per the city). [Sourced: NYC DCWP retail employer page; NYC DCWP Fair Workweek page] Other cities and states have their own laws; check where you operate. [General; not legal advice]
  • Put the rules into the scheduling tool as hard limits, not suggestions. [Our view]

7. What it costs and what it breaks

  • Software and data setup; poor forecasts waste the whole gain.
  • Staff trust: unpredictable hours drive turnover. Starbucks changed its policy within a day of public attention.
  • Service quality: too little buffer shows up as lines and empty shelves.
  • Fairness: auto-scheduling can quietly favor some people; review who gets the short or late shifts.
  • Skills: cross-train so one person can staff checkout and online picking.

8. Monday checklist

  1. Pull a month of hourly sales and plot them by hour and weekday.
  2. Compute staff-hours versus transactions; find your idle and overloaded hours.
  3. Set your schedule-notice rule (a week is a good target) and honor it.
  4. Write down min rest between shifts and ban "clopening."
  5. Track turnover and overtime alongside labor cost.

9. Try it yourself (no code, 30 minutes)

Use a day of your own hourly sales. For each hour, divide by the per-person rate and round up. Compare to your actual schedule. Add one extra person at your peak and see what it costs.

10. Self-check

  1. What did Starbucks promise on Aug 14, 2014? More consistent shifts, no back-to-back close/open, schedules a week ahead.
  2. In the example, how much does matched staffing save per day? $400.
  3. Why is a tight schedule risky? Forecasts are wrong; a 20% surprise leaves transactions unserved.
  4. What does NYC's law require for retail schedules? 72 hours' notice, no on-call shifts, limits on short-notice changes.
  5. Which cost might erase the saving? Turnover.

11. Where this is going

Scheduling tools will use more data and predict more finely. The pressure will be to use that precision responsibly: schedule early, keep slack, and let staff swap shifts easily. [Our view]

Sources

Review flags

  • I read the Seattle Times article in full and the NYC retail page; the NYT pieces only as excerpts (paywall). The "rarely learned her schedule more than three days before" line is quoted as the Seattle Times reports it.
  • Whether Starbucks fully followed through after 2014 is not examined; do not claim it did.
  • RESOLVED (Oct 7, 2026): NYC's current retail page (nyc.gov/site/dca/businesses/fair-workweek-retail-employers.page) and the DCWP retail FAQ (Apr 2022) were read: 72 hours' written schedule, no on-call shifts, consent needed for added time under 72 hours, limits on cancelling or shortening shifts. Amendments after April 2022 were not searched; the lesson states the rules as the city page lists them. The Nov 26, 2017 effective date remains a snippet from the city's page.
  • Turnover cost and all example numbers are illustrative. The only legal content is the NYC facts above, attributed, plus "not legal advice".

Claim ledger (C2 Lesson 6)

#ClaimSourceVerifiedStatusRe-check
1NYT report on Navarro, Aug 2014; hours set by softwareNYT; Seattle Times2026-10-06Seattle Times read; NYT excerptnone
2Starbucks email: software upgrade, curb clopening (NYT) / "never" (Seattle Times), schedules a week ahead ("no exceptions", Seattle Times); 130,000 baristas (NYT)Seattle Times; NYT2026-10-06Wording differs by source; verifier-confirmednone
3NYC retail rules: 72 hours, no on-call, notice limitsNYC DCWP2026-10-06Read2027-01
4NYC Fair Workweek retail protections took effect Nov 26, 2017; covered retail employers have 20+ NYC workersNYC DCWP2026-10-06Verifier-confirmednone
5Staffing and turnover arithmeticOur computation2026-10-06Illustrative, script-checkednone