When history fails: how Starbucks forecast through COVID

On 30 October 2019 Starbucks published fiscal 2020 guidance: global comparable store sales up 3 to 4 per cent and revenue up 6 to 8 per cent (Starbucks Corporation, 2019). Less than six months later it withdrew that guidance. If a company with years of sales data says it cannot estimate, what should the owner of a few coffee shops do with a revenue spreadsheet when history no longer predicts next year? This post reads how Starbucks handled it, through the releases it filed with the SEC, and asks what a small business can take away. See also the Starbucks forecasting case.

What did Starbucks say when it withdrew guidance on 8 April 2020?

It said it could not yet estimate the full financial impact beyond the second quarter "with reasonable accuracy", so it withdrew the full-year guidance. In the same passage it expected the third-quarter impact to be significantly greater than the second's and to extend into the fourth, because the US impact arrived late in the quarter and the flow-through of lost sales to profit was materially greater in the US than in China (Starbucks Corporation, 2020).

The numbers show why there was nothing to extrapolate. Through 11 March, US comparable store sales were up 8 per cent and transactions up 4 per cent, the strongest results in four years. From 12 March sales began to fall; in the last week of the month the decline settled at roughly 60 to 70 per cent, with only 44 per cent of US company-operated locations open. For the whole second quarter, US comparable store sales fell only about 3 per cent, because the first ten weeks were very good (Starbucks Corporation, 2020). A quarterly figure hid that the last three weeks had been entirely different.

With the year impossible to forecast, what did the company publish?

What it knew and what it could measure. The 8 April letter has three notable parts (Starbucks Corporation, 2020):

• China's recovery path. Comparable store sales there fell 78 per cent in February, 64 per cent in March and 42 per cent in the last week of March, the seventh consecutive week of improvement, at roughly the midpoint of the climb from a weekly low of -90 per cent in mid-February. The share of mobile orders fell from about 80 per cent in the last week of February to 27 per cent in the last week of March, which the letter links to a return to in-store transactions. The company wrote that progress might not be linear.
• Liquidity: about $2.5 billion in cash at the end of the second quarter, plus short-term borrowing facilities totalling $3.5 billion.
• An estimate for the quarter just ended: GAAP earnings per share of about $0.28 and non-GAAP of about $0.32.

Two things to learn. The company shortened its horizon to what it could pin down, and tied its recovery view in China to indicators it disclosed, such as weekly sales and the share of mobile orders. It also said it was using its experience in China to inform its response in the US (Starbucks Corporation, 2020). I read that as inference from one market to another; the letter does not use the word "analogy".

What do you see if you split transactions from ticket size?

Two forces pulling in opposite directions. Comparable store sales are roughly the number of transactions multiplied by the average ticket, and Starbucks publishes both.

The Q2 and Q4 2020 releases report "Americas and US" together; those two rows use that figure.

Check the multiplication for Q3: 0.48 × 1.25 = 0.60, that is -40 per cent. Q4: 0.75 × 1.21 ≈ 0.91, that is -9 per cent. The releases do not say why the ticket rose, so "customers bought less often but more each time" is only my hypothesis. What you can verify: if you forecast one revenue line, you cannot see that footfall and spend per visit are taking two different shocks, each recovering at its own pace.

How did the published ranges compare with the results?

Inside the range or above it, in every case. The table sets eight metrics of my choosing against the guidance published on 28 April (China Q3) and 28 July 2020 (the rest), for fiscal 2020.

Two things to notice. First, the ranges narrowed: the 28 July US guidance was -12 to -17 per cent, tighter than the -10 to -20 per cent that the release gives as the earlier guidance (Starbucks Corporation, 2020). Second, nothing came in below its range. The CEO's statement in the fourth-quarter release cites a faster-than-expected recovery in the US and China (Starbucks Corporation, 2020). From the releases I cannot tell whether the ranges were set cautiously on purpose or reflected real uncertainty.

One detail deserves a careful read: China's comparable store sales, -19 per cent in Q3 and -3 per cent in Q4, both include about 4 percentage points from a temporary value-added tax exemption (Starbucks Corporation, 2020; Starbucks Corporation, 2020). The July guidance for Q4 and the full year says explicitly that it includes this. The 28 April release does not mention the tax, so the first row of the table may not be like for like. My reading: the underlying recovery is lower than the printed figure.

For fiscal 2021, guidance was published on 29 October 2020: global comparable store sales up 18 to 23 per cent (actual 20), US 17 to 22 per cent (21), revenue of $28.0 to $29.0 billion ($29.1 billion), GAAP operating margin of 14 to 15 per cent (16.8) (Starbucks Corporation, 2020; Starbucks Corporation, 2021). I compared against the original guidance; any in-year revisions I have not read.

What can a small business take from this?

Scenarios first, model second. That is my use of the case, not something Starbucks says it did. Wack (1985) notes that forecasting errors became more frequent, and sometimes dramatic, from the early 1970s (Wack, 1985). My inference: when history does not guide you, a single number easily becomes a wrong anchor.

A hypothetical example: a chain of 6 coffee shops in Can Tho with revenue of VND 1,000 million a month in normal times. Every figure below is my own invention.

The method:

1. Split revenue into transactions times average ticket, as Starbucks publishes it.
2. Build scenarios only for what history cannot predict, here whether the shops are open. Run your usual model inside each scenario. See Scenario planning and the post One number or several scenarios?.
3. Remember that fixed costs do not fall with revenue. The Q2 release says that in the International segment, staff pay and occupancy costs kept running while stores were closed or on modified hours, taking the operating margin from 13.2 per cent a year earlier to -1.4 per cent (Starbucks Corporation, 2020). Scenario C must be counted in remaining cash, not only in revenue.
4. Tie each scenario to an early signal, like the share of mobile orders in China, and agree in advance what you will do when it appears.
5. Refresh on a fixed cycle, as in a rolling forecast, and give the bank or partners a range, not a point. Pick the method with Choosing a forecasting method.

When does this not fit, and where does the Starbucks example stop?

When it withdrew guidance, Starbucks had about $2.5 billion in cash and $3.5 billion of short-term facilities (Starbucks Corporation, 2020). In my view a small chain rarely has that cushion, so its constraint is liquidity, not forecasting technique, and for small, reversible decisions a rolling forecast is enough; scenarios earn their cost only when the decision is hard to undo.

The example is also one company, one shock and figures the company published itself. Fiscal 2022 shows a shock can return where you thought it had passed: comparable store sales in China fell 24 per cent, and the company's GAAP operating margin dropped to 14.3 per cent from 16.8, which the release attributes to investment in pay and inflation as well as sales deleverage from COVID restrictions in China (Starbucks Corporation, 2022).

Going deeper: what did I read, and what did I not?

I read in full the releases Starbucks filed with the SEC on 30 October 2019, 8 April, 28 April, 28 July and 29 October 2020, 28 October 2021 and 3 November 2022. I have not read the 10-K or 10-Q reports, the investor call transcripts, or the Form 8-K of 10 June 2020 that the July release refers to, so I do not know how the guidance moved in between.

For Wack (1985) I read only the public introduction, because the article sits behind a paywall. I have no independent analysis of how accurate Starbucks's guidance was, so the comparison table sets ranges against results and says nothing about forecast quality.