How to Read Segment Reporting: What Amazon Showed in 2015

Segment reporting is read by setting the company-wide totals aside and going straight to the note at the back of the filing that splits a business into its parts — then asking three things of each part: how fast is it growing, how much profit does it keep from each dollar of sales, and how is that profit defined. The headline figures are an average, and averages are where good businesses hide.

In April 2015, Amazon published results for the first three months of the year and, for the first time, broke out Amazon Web Services — its cloud computing arm, which rents computing power and storage to other companies — as a segment in its own right. The cloud business had taken in about $1.57 billion of revenue in the quarter¹ and earned $265 million of operating income,¹ meaning profit after the costs of running the business but before interest and tax. That is a margin near 17 percent,¹ sitting inside a company whose total operating income for the quarter was $255 million on $22.7 billion of sales — roughly 1 percent¹ — and which reported a small net loss for the period.¹ Revenue in the cloud arm was up about half from a year earlier.¹ Nothing about the business had changed between March and April. The resolution of the picture had.

What follows is a lesson in reading a disclosure, drawn from one company's filing history. It is not a recommendation to buy, sell or hold any company's shares.

Why Amazon read as a thin-margin bookseller for a decade

For most of its public life, Amazon reported in two geographic pieces: North America and International. The cloud revenue was in there, growing fast, but its costs and its profit were fused into the retail average. Anyone reading the consolidated statements saw enormous sales, razor-thin operating margins, and heavy spending on warehouses and servers. The honest reading of those totals was: a scale retailer that reinvests everything and earns very little.

The arithmetic of why that reading was wrong is worth spelling out, because it is the whole mechanism. A company-wide margin is a weighted blend. When the overwhelming majority of the sales dollars sit in a low-margin activity, the blend looks exactly like that activity — no matter what the smaller division is doing. Painting the company's 1 percent margin across $1.57 billion of cloud sales would have implied operating profit somewhere under $20 million. The segment line, once published, showed $265 million. That gap is not a rounding error; it is the natural consequence of pushing two unlike businesses through one number and then treating the number as a description of both.

Some analysts had tried to pull the cloud business out by inference. They counted data centres, estimated server fleets, surveyed customers, and modelled what a business of that shape ought to earn. The published estimates ranged from loss-making to highly profitable — and the range itself was the real finding. When the honest answer to "what does this part earn?" is "somewhere between a loss and a fortune," no conclusion resting on it can be trusted in either direction.

Owners who have run a multi-line company know this in their bones. A mature division earning steady margins, sitting beside a young one burning cash, produces a blended profit that describes neither. And anyone who has been through a sale process will remember the buyer's advisers asking, within the first week, for the profit and loss by product line — because they know the blended figure is a disguise, intentional or not. Segment reporting is the public-market version of that request: the one place in the filings where a company is made to answer it.

The note at the back, and the rule that decides what goes in it

Segment information lives in the notes to the financial statements, usually toward the back of the annual report — the 10-K in the United States — and of the quarterly report. It is among the least-read sections of the document and among the most useful. The filing, rather than the earnings call, is where this picture is drawn in full, because the call is management's narrative and the note is a disclosure the auditors have examined.

The rule governing that note explains both why it is powerful and why it can frustrate. Under US accounting standards and their international equivalent, a company must report segments the way its own most senior decision-maker looks at the business internally — what the standards call the management approach. If the chief executive reviews results by geography, the segments are geographies. If by product line, they are product lines. This was deliberate: the drafters wanted outsiders to see the business through the same lens as the people running it, rather than through categories imposed from outside. It is a very different logic from the industry buckets used to sort companies for convenience, where the label can mislead about what is actually inside.

The consequence is that the segments are only as revealing as management's internal view — or as management is willing to say its internal view is. There are size thresholds in the rules: broadly, a piece of the business becomes separately reportable once it reaches about a tenth of revenue, profit or assets. But the thresholds apply only to pieces management already reviews on their own. A division reviewed only as part of a wider geography never reaches the test, however large it has grown and however different its economics. No total was falsified in the years the cloud arm sat inside a geography. The company simply was not, on its own account, running it as a separate thing.

Which is why a new segment appearing is itself information. It usually means a business has grown large enough, or distinct enough, that the people at the top have begun managing it as a standalone enterprise. The resolution of the reporting changed because the internal governance changed. When Amazon made the split, it said plainly that the consolidated totals were unchanged, and it published the cloud line for earlier years as well — so readers could see that the business had already been meaningfully profitable the year before. The disclosure did not create a profitable business. It stopped burying one.

Three questions per segment, and the table that confesses what was left out

Three things are required of each reported segment, and these are the three to read.

First, revenue — including any sales made between segments, which are then stripped out in a line usually called eliminations so that the parts add back to the whole. Second, a measure of profit. The standard requires a measure, not a particular one, and that is where care is needed. Third, a reconciliation: a table that walks from the sum of the segments down to the consolidated figure on the income statement, listing everything the segments left out along the way.

That reconciliation is where the definition of segment profit becomes visible, and it is where trust in an analysis usually runs ahead of the reading. Companies commonly report segment profit before costs they regard as company-wide: head-office expense, stock-based pay, restructuring charges, sometimes interest and tax. In its 2015 annual report, Amazon's full-year segment profit was presented before stock-based compensation, with that cost — more than $2 billion for the year — reconciled separately underneath.² On that basis the cloud arm showed close to $7.9 billion of revenue,³ about 7 percent of the company's roughly $107 billion of sales,³ and around $1.9 billion of segment operating income³ — roughly two-fifths of the profit the segments reported between them, though the North American retail segment still earned more in absolute terms.³

The objection writes itself: if corporate overhead and stock pay are excluded, the segment margin is flattered. It is. The analytical check is not whether a margin was flattered but whether the conclusion survives putting the excluded costs back. In a later annual report, Amazon re-presented those same 2015 figures with stock-based pay inside the segment lines, and the cloud arm's operating income came down to about $1.5 billion.⁴ The exclusion changed the number. It did not create the gap. A business earning that on under a tenth of group revenue is a different animal from a retail operation earning a few cents on the dollar. The same discipline that applies to reading a company's adjusted profit figure like a buyer applies here: find out what was taken out before admiring what was left.

A short comparison makes the two layers concrete.

The habit worth building is unglamorous. Open the segment note before the income statement. Work out the margin per segment from the company's own figures. Read the reconciliation to see what each of those margins excludes. Then go back to the front of the report, where the totals now read as what they always were — a weighted average of businesses that may have almost nothing in common.

The margin that flatters, and the capital that never appears in it

There is a fourth question, and it is the one most often skipped. A margin tells you what a business keeps from a dollar of sales. It says nothing about what the business had to spend to be in a position to make the sale.

Amazon's 2015 note also disclosed assets by segment, and the cloud business — on a small fraction of group sales — carried more property and equipment than the North American retail business,² and absorbed a large share of the year's additions to property and equipment. The slice with the pristine operating margin was also the slice swallowing the capital. That is not a criticism of the business; infrastructure businesses are built that way. But it changes the reading. Segment operating profit is an operating measure. It is not cash left over once the servers are bought and the buildings are up — a distinction this publication has drawn at length before.

So where a company gives the asset and capital-spending lines by segment, read them beside the margin. A division showing a 25 percent margin while consuming most of the company's capital spending is a different proposition from one showing the same margin on rented premises and a few laptops. And where a company gives no segment asset line at all — which is common, because the rules only compel it when the senior decision-maker actually reviews it — the absence is itself a fact. It tells the reader how much of the business they are being allowed to see.

Resolution can fall as well as rise: Meta and Microsoft

The Amazon episode is a story of resolution improving, and it would be a mistake to leave the impression that the arrow only points one way.

Sometimes a split reveals a loss rather than a profit. When Meta began reporting its virtual and augmented reality efforts as a separate segment, Reality Labs, alongside its advertising businesses grouped as Family of Apps,⁵ the consolidated margin compression that had been read as decay in the core business resolved into something quite different: an advertising operation with margins that had barely moved, sitting beside an experiment losing many billions of dollars a year.⁶ Whether the experiment is wise is a separate question, and not one segment data can answer. The point is that until the split, the two could not be told apart from outside. A deteriorating core business and a healthy one funding a costly bet look identical in a blended number, and they are not remotely the same thing to own.

Microsoft is the clearest example of the hazard in the other direction. The company moved from segments organised around its historic product divisions to a split between consumer-facing and commercial groupings, and then moved again to the three segments it has used in broad form since — productivity software, cloud infrastructure, and personal computing.⁷ Each change reflected a real shift in how the company was being run; the cloud business had become the organising principle. Each change also broke the chain of comparable history. The Windows division, once visible on its own, became one component of a larger personal-computing segment. Companies must restate prior periods when they redraw segments, but only for the periods they present — which means a decade of history can be reorganised into something that cannot be lined up against itself. A ten-year segment trend is often not available, and a reader who assumes otherwise is comparing two differently drawn maps.

What it takes to trust an analysis built on segments

Most people stewarding significant capital will not be the ones reading the note. They will be handed a conclusion — by a manager, an adviser, a friend with a view — and asked, in effect, to trust it. Segment reporting is one of the few places where that trust can be tested without becoming an analyst, because the questions are short and the answers are either in the filing or they are not.

Which segment is the claim actually about? A thesis that rests on "the company's margins" rather than a named part of the business is a thesis about an average, and averages, as this episode shows, can be wrong about every component at once. What definition of segment profit is being quoted, and what sits in the reconciliation underneath it? If the answer is a shrug, the margin is being used as a decoration rather than evidence. Does the conclusion survive putting the excluded costs back in? Has the company redrawn its segments during the period being compared? And where is the capital going, as against where the profit is being shown?

None of that requires technical skill. It requires knowing that the answers exist in a specific place, and that a careful analysis will have been there. This is the same test that applies to anyone presenting work for a decision: not whether the conclusion is impressive, but whether the person can show the path they took to it — which is why the questions worth asking of a manager are mostly about process rather than opinions.

The portable principle is this: a consolidated number is a claim about an average, and nobody owns an average. Businesses are owned one part at a time, and the parts are disclosed — imperfectly, on management's own terms, in a note most readers skip. Nothing about Amazon's economics changed in April 2015. What changed was how much of them could be seen. Before acting on any analysis of a multi-business company, it is worth knowing whether whoever wrote it went and looked.