Case Study 18.2 — Starbucks, May 29, 2018: The Limits of a Training Day, and What Actually Changed the Outcome

Type: real, public case, chosen deliberately as a contested one. Facts drawn from public reporting and the company's own public statements; magnitudes given as ranges where the precise figure is not something we can stand behind. The evaluative claims about training effectiveness are attributed as industry and research consensus in general terms, not as a specific study, and should be read that way.

This is the complementary angle to Case Study 18.1. That one was about a training program that faded. This one is about a training day that was widely questioned at the time — and about the quieter thing that happened alongside it, which is arguably what did the work.

Background

On April 12, 2018, two Black men sat down in a Starbucks in Philadelphia to wait for a business meeting. They asked to use the restroom without having bought anything and were told the restrooms were for paying customers. A store manager called the police. The two men were arrested and later released without charges. Video taken by other customers spread within hours.

The immediate operating question is worth isolating before the response, because it is a training question and it is one every restaurant faces in a smaller form. What was the employee supposed to do?

The honest answer, publicly reported in the aftermath, is that the company's policy on non-purchasing guests was substantially discretionary and varied by store and by manager. Which means the employee was executing a rule — just not one anybody could point to, apply consistently, or be trained against. That is precisely the failure mode of §18.1: a standard that exists only in someone's head produces eleven private interpretations, and the one that ends up on video is the one you own.

The response

Starbucks moved quickly and publicly. The chief executive apologized in person and publicly. The company reached settlements with the two men and with the city, including funding for a program supporting young entrepreneurs.

Then it did the thing everybody remembers. On May 29, 2018, Starbucks closed more than 8,000 company-operated US stores for an afternoon and delivered racial-bias training to roughly 175,000 employees. Forgone sales for the afternoon were widely estimated in the low tens of millions of dollars, and the closure was covered on every national news outlet in the country.

And it did one other thing, much less covered. In May 2018 the company announced a changed policy: anyone may use Starbucks stores and restrooms, whether or not they make a purchase.

The operating issue: which of those two actions was the intervention?

This is the case, and it is genuinely contested.

The training day was criticized from two directions at once. Critics on one side argued it was corporate theater — a costly gesture calibrated for the news cycle. Critics on another argued something more substantive and more useful to an operator: the general weight of research on one-off implicit-bias training does not support the expectation of durable behavior change from a single session. Attitudes measured immediately after such sessions often move; behavior measured months later frequently does not. That is a fair summary of the broad consensus rather than a citation to any single study, and it should be treated as such.

The policy change is a different kind of instrument entirely. It did not attempt to change what an employee believed. It removed the discretion in which the incident happened. After May 2018 there was no judgment call available at the counter about who may sit down, because the answer was written and it was the same in every store.

Set the two side by side in this book's vocabulary:

By the four properties in §18.1, the policy is a standard and the training day is not.

What it shows

1. Training cannot substitute for a missing rule. If your people are improvising because nothing is written down, more training on judgment produces better-informed improvisation. Write the rule first. Then train it. The order in §18.2 — standards before everything — is this lesson stated as a curriculum.

2. Discretion is where inconsistency lives, and inconsistency is where your worst day comes from. Every restaurant has discretionary moments: who gets comped, who gets seated at the good table, who gets asked to leave, whose ID gets checked. Each one is a place where eleven employees will do eleven things. §18.5's ID threshold of 35 is the same instrument as Starbucks' open-restroom rule — a bright line that removes a judgment call precisely because judgment calls are where bias, favoritism, and inconsistency enter unannounced.

3. A single session is a signal, not a system. This is the same finding as Case Study 18.1 from the opposite direction. The signal has real value — it tells 175,000 people that the organization is serious, which is not nothing — but it sits at week 1 on the decay curve. What changes week 24 is recurrence, testing, and a rule.

4. The gesture is priced, and the price is legible. Closing stores for an afternoon cost real money and was intended to be seen to cost real money. That is a legitimate management move; visible costliness is how organizations signal that something is not routine. Just be clear with yourself about which budget it comes out of. A costly signal is a marketing expense that looks like training. It may be worth buying. It is not a program.

The lesson

Ask of every training initiative: am I teaching a behavior, or am I substituting for a rule I have not written?

If the answer is the second, the cheapest and most durable fix is not a session. It is one sentence in the standards document, owned by a role, with a moment when somebody checks.

For an independent operator the transferable version is small and immediate. Before you build a training module on "using good judgment with difficult guests," write down the five discretionary decisions your staff actually face — the ID threshold, the comp authority, the refusal of service, the guest who wants to sit without ordering, the dog on the patio — and make four of the five into rules. Then train the fifth one properly, which you can now afford to do, because it is the only one left.

Discussion questions

1. Which action — the training day or the policy change — do you believe did more to prevent a recurrence? What evidence would settle it, and is that evidence obtainable?
2. The training day cost tens of millions of dollars and the policy change cost almost nothing. Does that mean the training day was a mistake? Argue both sides, and identify what the expensive action bought that the cheap one could not.
3. List the discretionary decisions your staff make every shift. For each, decide whether it should become a rule or remain judgment, and say what makes the difference.
4. §18.5 sets Bellwether's ID-check threshold at 35 — a deliberately high bright line that produces eye-rolling and zero ambiguity. Is that the same instrument as the open-restroom policy? Where else in a restaurant would you use a deliberately over-inclusive rule to eliminate a judgment call, and what does the over-inclusion cost?
5. Case Study 18.1 showed a program that faded; this one shows a session that many doubted from the start. What do the two cases together imply about how an independent with a \$54,708 annual training budget should allocate between events and recurrence?
6. Write the one-sentence Bellwether policy that would have removed the discretion in the Philadelphia incident, in the format of Figure 18.1 — observable, timed or counted, and owned.