Case Study 2 — Starbucks, 2007–2008: The Standards Nobody Wrote Down, and the Three Hours That Admitted It

Sourcing note. This case uses publicly documented material: a widely published internal memo from February 2007, the company's public announcements and reported results during fiscal 2007–2008, and Howard Schultz's own published account in Onward (2011). Approximate figures are flagged. Interpretation marked [our reading] is ours. This is a contested-decision case: reasonable operators disagreed at the time about whether the central action was leadership or theater, and the disagreement is the point.

Background: growth as an anesthetic

By the mid-2000s Starbucks was among the fastest-expanding restaurant businesses in the world. It had gone from under a thousand stores in the mid-1990s to well over thirteen thousand by 2007, opening new locations at a rate of well over a thousand a year at its peak (approximate; the company's public filings carry the precise counts).

Every headline number was excellent. Total revenue rose every year. Unit count rose every year. New markets opened. On any consolidated report, the business was working.

This is precisely the condition §37.6 warns about. Growth is an extraordinarily effective anesthetic. A company opening a thousand stores a year can absorb a great deal of unit-level decay in its consolidated line and never see it, because the top line goes up every single period regardless. It stops working on the day you stop opening.

The operating issue: a memo nobody was supposed to read

On February 14, 2007, Howard Schultz — then chairman, having stepped back from the chief executive role — sent an internal memo to the company's leadership. It was titled "The Commoditization of the Starbucks Experience." It leaked almost immediately and was published widely.

The memo's argument, paraphrased, was that a series of decisions made over the previous decade to enable growth and improve efficiency had each been individually correct and had cumulatively hollowed out the product. Among the examples he raised publicly:

• Automated espresso machines had replaced the older manual machines. They were faster, more consistent, and easier to train on — every virtue this chapter's §37.3 recommends. They also removed the visible craft of the drink being made and, being taller, blocked the sightline between the barista and the customer.
• Flavor-locked packaging had replaced scooping and grinding whole beans in the store. It preserved freshness and reduced labor and waste. It also removed the smell of coffee from a coffee shop.
• Standardized store design had made new units faster and cheaper to build and more consistent across markets. It also made them, in his words, less distinctive — closer to a chain and further from a neighborhood.

Read that list carefully, because it is the most uncomfortable thing in this chapter.

Every single one of those decisions is a decision this chapter tells you to make. Reduce variance. Automate the step where two competent people would do it differently. Specify the output. Make the unit easier to open, easier to staff, and easier to audit. Those are the correct answers to the questions §37.2 and §37.3 pose.

And in aggregate, according to the person who built the company, they optimized away the thing the guest was actually buying.

Why the systems did not catch it

[our reading] Here is the diagnosis in this book's vocabulary, and it is the reason this case is worth more to an independent operator than the Chipotle case.

The things that were lost had never been written down. Nowhere in any operations manual did it say: the store smells of ground coffee. Nowhere did it say: the guest can see the barista's hands. Nowhere did it say: the person making your drink appears to be exercising a craft. Those were not standards. They were by-products of how the work had happened to be done, invisible until they were gone.

This is the blind spot in the write-it-down test, and it is worth naming plainly. The test asks whether two competent people would do something differently. It does not ask what disappears when a thing stops being done at all. You cannot audit a standard you never wrote, and the standards you never write are the ones that seem too obvious to state.

Nor did the reporting package catch it. Comparable sales in the U.S. had been positive for years. Store counts were rising. There was no line on any dashboard called romance or theatre — and if there had been, no threshold could have been set on it. The signal arrived as a memo from a chairman walking into stores, which is to say: it arrived through §37.7's visit, and it arrived only because the person visiting had built the original and could feel the difference.

The memo is, in that sense, a district-manager visit report from the founder. It is what §37.7 means by "eat in the dining room, as a guest, at least once a quarter."

The contested decision: closing every store for three hours

Schultz returned as chief executive in January 2008. On February 26, 2008, Starbucks closed approximately 7,100 U.S. company-operated stores for about three hours to retrain baristas on espresso preparation.

It was, at the time, widely mocked. Competitors ran promotions during the closure window. Commentators called it a stunt. The direct cost was several million dollars of forgone sales plus the payroll of tens of thousands of employees sitting in a training session (the precise figure was estimated publicly at the time; treat any specific number you see as an estimate).

The case for it. A written standard that has drifted across seven thousand buildings cannot be restored by a memo, a video, or a line in a manual. It has to be re-transmitted by demonstration, simultaneously, in a way that is impossible to ignore or defer. Closing every store on the same afternoon is the only version of that which cannot be scheduled around. It also communicated something no document could: this matters enough that we will close. Every employee in the company received that message on the same day.

The case against it. Three hours cannot install a craft, and the drift was not caused by a training gap — it was caused by equipment and format decisions made years earlier, none of which a training session reverses. [our reading] The closure treated a structural problem with a cultural gesture, and the gesture was expensive, easy to satirize, and, critically, not repeatable. You get to do it once.

Both readings are defensible. What is not in dispute is what came next.

The outcome

Over 2008 the company:

• announced the closure of roughly 600 underperforming U.S. company-operated stores (later expanded), and stated publicly that a large majority of them had been opened in the immediately preceding years;
• sharply slowed new-unit growth, abandoning previously stated expansion targets;
• reported negative comparable-store sales in the U.S. — the first such declines in its history as a public company — and a quarterly loss during fiscal 2008 (publicly reported; approximate);
• launched a multi-year turnaround program that Schultz later documented in Onward.

That second bullet is the one an independent operator should sit with. A large share of the stores being closed were stores the company had recently opened. In this chapter's terms, the group had been opening units that cannibalized units it already had, and the consolidated revenue line — which went up each time — could not tell it so. Comparable sales eventually could, but by the time a comp decline is large enough to be unambiguous, the leases are signed and the build-outs are paid for.

This is Figure 37.5's central question at national scale: did Unit 3 take Unit 1's guests? Starbucks answered it in 2008, by closing Unit 3.

What this shows, in this chapter's terms

1. Growth hides operating decay for exactly as long as you keep growing. Total revenue and unit count are measures of your construction schedule. Comparable sales and comparable traffic are measures of your business. A company can report record revenue every quarter while every existing store gets slightly worse, and the arithmetic will not object.

2. Standardization can optimize away the product. This is the honest limit of everything else in Chapter 37. Reducing variance is correct. Reducing variance by removing the step where the craft was visible is also, sometimes, removing the reason people came. There is no formula that distinguishes these two in advance. The only defense is a leader who is regularly in the room as a guest and who is allowed to say "this is worse" without a number to support it.

3. Cannibalization is invisible on a consolidated report and obvious on a comp report. Before you sign a second location, you need a written answer to: where will its guests come from, and how will we know if they came from us? Chapter 2's market analysis is where that answer belongs; §37.6 is where you find out whether it was right.

4. Culture does not travel by manual, and it does not travel by a three-hour meeting either. What a mass retraining can do is signal priority. What it cannot do is rebuild a norm that was dismantled by equipment and format decisions. If the structure produces the behavior, changing the structure is the only remedy — which is the same lesson as the sick-employee policy in Case Study 1, arriving from the opposite direction.

5. The founder's presence was the only functioning sensor. A memo from a chairman walking into stores was the company's early-warning system. That is a working sensor and a catastrophically unscalable one — it exists in exactly one person. Building an instrument that detects the same thing without that person is the entire project of §37.3 and §37.7, and this case shows how hard it is, because the thing being detected had never been specified.

Discussion questions

1. Every efficiency decision named in the 2007 memo is a decision this chapter would endorse. Write the one-page decision framework you would have wanted the company to apply before each change — a set of questions that would have surfaced "what disappears if we do this?" without blocking every improvement. Then test your framework against a real decision in a restaurant you know.

2. "You cannot audit a standard you never wrote, and the standards you never write are the ones that seem too obvious to state." Working from a restaurant you have worked in or eaten in regularly, list five such unwritten standards. For each, decide whether it could be written as an auditable obligation (§37.3's Hospitality callout) or genuinely could not.

3. Was closing 7,100 stores for three hours the right call? Argue both sides in a paragraph each, then state your own answer and the single piece of evidence that would change it. Note explicitly what the action could and could not accomplish.

4. A large share of the stores closed in 2008 had opened recently. Design the pre-opening analysis a three-unit independent group should run before signing a fourth lease, specifically to test for cannibalization. What data would you need, where would it come from (Chapter 26), and what result would make you walk away?

5. Compare this case with Case Study 1. Chipotle's failure was in a system that was documented and not controlled tightly enough; Starbucks' was in a system nobody thought to document at all. Which failure mode is more dangerous for a two-unit independent, and why? Does your answer change for a ten-unit group?

6. The company's early-warning system was one person's ability to walk into a store and feel that it was wrong. Design the closest scalable substitute you can for a five-unit independent group: who does it, how often, what they record, and how what they record reaches a decision. Be honest in your write-up about what your substitute will miss.