A divested segment corrupts multi-year CAGRs, and the error always understates growth

Claim

fin.py reported "Revenue CAGR 24.8% (3y)" for AppLovin. The real growth rate of the business that exists today is roughly 70-75% per year.

The vendor's annual revenue series:

FY2022	FY2023	FY2024	FY2025
$2,817M	$1,842M	$3,224M	$5,481M

Read naively, revenue fell 35% in 2023 and then compounded 24.8% over three years. Read correctly, FY2022 includes the Apps (mobile games) business that AppLovin subsequently divested, while FY2023-25 are the continuing Software Platform only. On a like-for-like basis the series is $1,842M → $3,224M (+75.0%) → $5,481M (+70.0%).

The vendor CAGR understated growth by roughly 3x.

Why it is dangerous rather than merely wrong

The error has a direction. A divested segment is nearly always shrinking, lower-margin, or strategically unwanted — that is why it was sold. So the old total-company base is larger than the restated continuing base, and dividing a continuing endpoint by a total-company start always produces a CAGR that is too low.

Compare this to [[pitfall-vendor-forward-eps-is-the-wrong-fiscal-year]], which errs toward cheap and manufactures false bargains. This one errs toward mediocre and hides real compounders — it causes a name to be screened out, which is a cost that never appears in a portfolio and therefore never gets audited.

It also fabricates a phantom decline year, which is worse than a wrong average. A screen looking for "revenue growth positive every year" rejects the name outright, and a human reading the table sees a business that shrank 35% and stops reading.

The tell

A single year that breaks the trend violently, with no matching break in margins or cash flow. In the APP case, revenue supposedly fell 35% in FY2023 while:

gross margin rose 80.8% → 83.7%
operating margin rose 41.9% → 59.3%
FCF rose $998M → $2,072M

A business that loses a third of its revenue does not simultaneously expand margins and double free cash flow. When the revenue line and the profitability lines tell opposite stories across one boundary year, the boundary is an accounting change, not a business event.

Other tells: - The break year coincides with a known divestiture, spin-off, or discontinued-operations classification — check the 10-K's segment note before computing anything. - The oldest year in the vendor's series is the only one that looks out of family. - Two vendors disagree about a single historical year while agreeing on the rest.

How to apply

Before quoting any multi-year CAGR, look at the earliest year for a discontinuity. If one year breaks the trend by more than the others combined, do not compute across it.
Compute the CAGR only across the comparable basis, and say so explicitly — "FCF CAGR available only from FY2023 because the Apps divestiture breaks the series" is a finding, not a gap. The framework asks for 5-8 year CAGRs; when the data cannot support one, flag it rather than fabricating it (§1 requires flagging incomplete data points).
Cross-check against margins and cash flow. They are usually restated consistently even when revenue is not, so they reveal the true shape of the business.

What would falsify this

A vendor feed that restates the entire historical series to continuing operations, making the computed CAGR correct without adjustment. Some do; the point is that you cannot tell which without looking, and the failure is silent when they don't.

[[pitfall-vendor-forward-eps-is-the-wrong-fiscal-year]] — the same family (vendor period boundaries applied inconsistently) but the opposite direction: that one makes stocks look cheap, this one makes them look mediocre.

[[principle-primary-source-beats-vendor]] — the segment note in the 10-K settles this in one minute; no amount of vendor cross-checking does.

History