Revenue Forecasting: a Precision Sport
I have always been fascinated by the precision sport of the revenue forecasting. Out of all the companies where I worked, eBay will forever have a special place in my heart when it comes to exercising this sport. eBay is where I have learned the trade, where my performance review target was “to hit the revenue forecast within the month to 1ppt and within the quarter to 3ppt,” and where many late nights were spent in the parking garage bonding with the rest of the finance team over the month close and figuring things out.
What’s my “so what” on the revenue forecasting?
If you do it, do it well. When a revenue guidance is issued by a public company, the confidence interval tends to be quite narrow (2ppt wide for a mature business such as eBay, 4ppt wide for a growing one such as Uber).
It’s ok to temporarily issue no numeric guidance or increase your confidence interval when the volatility is high – Airbnb, PayPal, Uber all did that when they needed to avoid issuing hugely incorrect forecasts.
Sometimes the volatility comes from the market (like COVID or the economic slowdown), but sometimes it comes from the company's own growth strategy (promotional actions of Uber). A less volatile proxy for the revenues (Gross Bookings for Uber, or GMV for other companies) can be a good solution in the latter case.
How do large, complex, multi-product line businesses forecast their revenues? How do public companies, whose stock price depends on predictability, deliver on their promises, quarter after quarter?
I asked my friend ChatGPT and its pdf plugins to analyze several hundred quarterly financial statements and create a dataset of revenue guidance confidence intervals versus realized revenues of major public companies in the Technology and Marketplaces space.
What are the amazing insights from this AI-driven research? Who am I kidding; I ran out of my 50 questions per 3 hours limit in an hour, called ChatGPT an idiot, and did a much more limited analysis by hand myself – but at least you can trust the results I share below.
I looked at eBay, PayPal, Airbnb, and Uber in detail. In the past four years, these companies:
have been selective with issuing revenue guidance at all,
kept the guidance at the highest level of aggregation,
withheld the guidance when the uncertainty increased,
“sandbagged” the forecast to land at the top of the guidance range, and
overcorrected when forecast went wrong.
Let's look at them in more detail below.
eBay tends to issue a revenue guidance for the next quarter with the average variance of about 2ppt. The only notable exception has been Q2 2020, where the interval was 4ppt. Does eBay always hit its stated goal of forecasting precision?
Well, the answer is a careful “actually, kind of yes.” If I remove two outlier quarters, the average error is indeed 1ppt! Very impressive; and requires a high quality of the database systems, reconciliations, and analytics that goes into building this forecast.
Airbnb is a newcomer to the revenue guidance business after their IPO in December 2020. Note that the comparison graph starts only a year after the IPO – the first time that Airbnb has issued a revenue guidance was in Q4 2021, when – not coincidentally – their business has returned to a predictable YoY growth pattern for the second quarter in a row.
Another observation of note is the direction of the forecast error: the actual revenues have exceeded the mid-point of the forecast every single time, on average exceeding the top range by 1ppt as well. Every public company CFO always tells their team: “there is no such thing as a good surprise.” The markets nevertheless do punish overly optimistic forecasts and companies do "sandbag".
While PayPal is not a marketplace, I include it here for several reasons. Number one – I admire the exemplary performance of PayPal over the years, with the strong consistent growth before, during, and after the pandemic.
The second reason is how PayPal translates forecasting uncertainty into a revenue guidance. Some companies, like Airbnb or Uber, simply did not issue guidance in the quarters with high volatility. Others, like eBay, churned out a consistent forecasting approach.
What PayPal did was different: it changed its stated precision of the revenue forecast depending on the structural uncertainty in that quarter. PayPal is the only company in my sample that issues one-point forecasts (“we expect the revenues to be $6.80B”) at all – and when it does, it hits them within 1ppt accuracy. But it’s also the only company in the sample that explicitly widened its confidence interval at times to reflect increased volatility.
In Q4 2020, PayPal has widened their confidence interval to 4ppt, while eBay overshot its forecast mid-point by 7ppt, and neither Airbnb nor Uber issued any guidance at all. PayPal hit the revenues within 1ppt precision in that quarter.
Finally, let’s look at Uber “playing it safe” with the forecasting precision. Uber does not issue a revenue guidance – only the Gross Bookings one, before marketing promotions introduce high volatility to their take rate. It’s interesting and kind of cool that Uber simply sums up their Gross Bookings across the three businesses – mobility, food delivery, and freight – and it seems to work quite well. What you can also see is that their quality of forecasting is getting better over time. The confidence interval has been 4ppt-wide in the first quarters that the guidance has been issued and is slowly narrowing towards 3ppt. The actuals are hitting the mid-point of the forecast quite well, with less than 1ppt deviation in 2023.
Would you want to see analyses like that for more companies? What are the insights you find the most practical? Please hit the comment section with your feedback.