When free cash flow falls, conversion says if it matters

We have written this story about single companies more than once. A company reports higher profit, the cash flow statement goes the other way, and the gap turns out to sit in one working-capital line. monday.com showed it at a small scale, Fabrinet showed it at a large one — revenue up 35.7%, free cash flow down 98.0%, and NetApp showed it in the inventory line — profit up 60.9%, operating cash flow down 25.3%.

The obvious next question is how common that is. So we screened every company that files with the SEC.

The answer is that it is common, and that most of the time it means nothing at all. That second part is the reason this article exists.

The screen

Four figures, two consecutive years, from the SEC’s XBRL frames API: net income, operating cash flow, capital expenditure, and the same three for the prior year. Free cash flow is operating cash flow less capital expenditure.

The $1 billion floor is arbitrary and we say so. Without it the table runs to 484 rows and nobody reads it. Utilities and financials are removed because the ratio means something different there — a regulated utility funds its rate base with debt and recovers it over decades, so capex above operating cash flow is the business model, not an event. The same exclusion, for the same reason, applied to our screen of companies spending more on capex than they earn.

What the list actually shows

Here is the finding we did not expect. Of the 45 companies whose free cash flow fell while profit rose, fifteen still produced more free cash than they booked as net income.

A company that converts 182% of its profit into free cash flow does not have a cash problem because that figure was 200% a year ago. Bristol-Myers Squibb is on this list on the strict reading of the screen — profit up, free cash flow down — and its free cash flow was $12,845 million against $7,054 million of net income.

So the direction is not the signal. The level is. “Net income up, free cash flow down” is a sentence that fits a third of large American companies in a normal year, and used on its own it sorts nothing.

The full list

Sorted by cash conversion, lowest first. Every company here reported higher net income than the year before and lower free cash flow than the year before; the last two columns are what separate them. Figures are in millions of US dollars, from each company’s most recent annual filing.

Read the last two columns together. A company that fell from 382% to 119% — Cardinal Health — is not in the same situation as one that fell from 55% to 10%, and neither is in the same situation as Oracle, which was already negative and went further. The screen produced all three, which is what a screen does. Sorting them is what the ratio does.

The eight that convert least

Two of these we have written about before, and in both cases the reason is the same and is not hidden.

Operating cash flow          $31,977M
Capital expenditure         −$55,663M
───────────
Free cash flow              −$23,686M
Net income                   $17,087M   →  conversion −139%

Operating cash flow         $139,514M
Capital expenditure        −$131,819M
───────────
Free cash flow                $7,695M
Net income                   $77,670M   →  conversion 10%

Amazon’s net income rose 31.1% and its free cash flow fell 76.6%. Capital expenditure went from $83.0 billion to $131.8 billion. That is the whole of it, and it is the same distinction we drew when Amazon’s free cash flow changed sign depending on which definition you used: the arithmetic is not in dispute, the label is.

Oracle is the more extreme case, and we took it apart when its free cash flow first went negative. Note what the table above does not say: it does not say Oracle is in trouble, and it does not say the building will pay off. It says the company spent $55.7 billion on property while collecting $32.0 billion from operations, which is a fact, and that the market will settle the rest.

Spending that is not covered by operating cash has to be funded, and the funding shows up a few quarters later on a different line. Across 1,519 companies, aggregate interest expense rose 10.1% in the first half of 2026 — and Alphabet alone produced 16% of that increase.

Property is not the only place cash gets parked. Working capital absorbs it too, and inventory is the visible half of that: 66 companies ended fiscal 2025 selling less than the year before and holding more stock, adding $7.1bn of inventory while revenue fell $50.2bn.

The company that is not on this list

Forty-six companies cleared the screen. One of them was not real.

Vince Holding Corporation — a clothing retailer — appeared with net income of $6,378 million, which would place it above Freeport-McMoRan. Its actual net income for the year was $6.378 million, a thousand times smaller.

Form 10-K, fiscal year ended 31 Jan 2026        $6,378,000
Form DEF 14A, same period                   $6,378,000,000
──────────────
1,000× apart

The 10-K is right. The proxy statement is tagged wrong, and the frames API — which does not care which document a number came from — publishes the wrong one. Anyone screening on that endpoint puts a company with $6.4 million of profit into a list of billion-dollar earners and never knows.

We only found it because the number looked impossible next to the company’s own operating cash flow of $2.987 million, and we check figures against the filing before we publish them. The screen now re-reads every surviving company’s net income from its annual report and drops any name where the two disagree by more than 1%.

This is not a story about one small retailer. It is what a data source looks like from the inside. The SEC’s API is the best public financial dataset there is, and it is still only as good as what was filed into it.

We have written up the mechanism separately, including the tolerance that separates a units error from ordinary rounding: why the SEC’s own API will hand you a number 1,000 times too large, and the one extra request that catches it.

What would make this wrong

• Fiscal years are not aligned. The frames API groups annual periods that overlap calendar 2025, and 12 of the 45 companies do not close in December. Oracle’s year ends 31 May 2026 and Microsoft’s ends 30 June 2025 — both labelled CY2025, a full year apart. The list is a set of fiscal years, not a snapshot of one period.
• Free cash flow here is one definition. Operating cash flow less purchases of property, plant and equipment. It ignores acquisitions, capitalised software bought separately, and leases — three ways in which reported capex understates what a company actually spent.
• Four of the 45 rose from a loss, not from a smaller profit. JOYY, Marvell, Constellation Brands and Bristol-Myers Squibb had negative net income in the prior year, so “net income rose” is arithmetically true and narratively different.
• A falling conversion ratio is not a verdict. A company building capacity ahead of demand and a company whose collections are slipping produce the same row in this table. Distinguishing them requires reading the filing, which is what our company articles do and what a screen cannot.
• The $1 billion floor hides the small end. 484 companies passed the screen at any size. We looked at 75 of them and published 45. The other 409 are not claimed to be uninteresting; they are unexamined.
• The exclusions are by SIC code, which the company itself chooses. A company classified outside 6000–6799 that behaves like a financial will still be in the table.

Check it yourself

Every figure comes from the four API endpoints linked below, and the screen is a script in this site’s repository, so the table can be rebuilt from scratch:

node scripts/screen-income-up-cash-down.mjs

The one judgement we added is the $1 billion floor and the two sector exclusions. Everything else is arithmetic on filed numbers, and the Vince Holding discrepancy can be seen by opening the company concept link and comparing the 10-K and DEF 14A rows for the period ending 31 January 2026.

If a figure here does not match a filing, tell us and it will be corrected on the article and on the corrections log, with the date.