Sum-of-the-Parts Valuation

Why you’d reach for this method

GE spent decades as the textbook case for why a single multiple fails a conglomerate. It had an aircraft engine division, a power generation unit, a healthcare imaging business, and GE Capital \u2014 a financial arm that by 2007 held over $500 billion in assets. Analysts who applied an industrial sector P/E to the whole company were implicitly valuing a leveraged financial services operation at the same discount rate as jet turbines. When the 2008 credit crisis repriced financial assets, GE Capital dragged the whole enterprise down. The consolidated multiple had been hiding the embedded risk in plain sight.

SOTP exists because different businesses have different risk profiles, different growth trajectories, and different capital requirements. A payments processor typically trades at 20–30× EBITDA. An oil refinery trades at 5–7×. A software business might trade at 15–25× while a specialty retailer sits at 8–10×. Put two of those together in one set of financials, apply a blended average, and you get a number that accurately describes neither business. You will systematically overpay for the low-multiple operations and underpay for the high-multiple ones depending on which way the blend runs.

The right question is not “what multiple does the market put on this company” but “what would each piece be worth if it were public and independent?” SOTP forces that question into the open. It is the only valuation method that makes the internal cross-subsidies visible \u2014 which is exactly why management teams at diversified companies rarely run it in investor presentations and why activist investors run it first.

How the table actually works

Use Alphabet as the concrete walkthrough. Alphabet’s 2023 10-K reports three reportable segments: Google Services, Google Cloud, and Other Bets. Google Services \u2014 which includes Search, Gmail, Maps, YouTube ads, and Google Play \u2014 generated approximately $87.1 billion in operating income that year on about $272 billion in revenue. Google Cloud reported $1.7 billion in operating income on $33.1 billion in revenue (2022 was still a loss year; Cloud crossed into profitability in 2023). Other Bets, which includes Waymo, Verily, and related moonshots, generated roughly $3.8 billion in revenue but lost approximately $1.2 billion at the operating level.

Step 1: Identify and define each segment. Alphabet’s three reported segments are your starting point, but Google Services can reasonably be split further into Search/Advertising and YouTube for a more granular analysis. For most purposes the three reported segments are sufficient. The rule is: split segments when the underlying businesses have meaningfully different growth profiles, margin structures, or peer groups. Combining them hides exactly what SOTP is supposed to reveal.

Step 2: Assign a financial metric to each segment. For established, profitable businesses, forward EBITDA is the most common anchor. For high-growth segments with thin or negative EBITDA, forward revenue is often more defensible. For Alphabet, using operating income as an EBITDA proxy works reasonably well because the 10-K discloses depreciation at the segment level, letting you back into segment EBITDA. For Google Cloud specifically, applying a forward EV/Revenue multiple rather than a trailing EBITDA multiple is arguably cleaner given the margin trajectory.

Step 3: Find pure-play comps and apply a multiple. For Google Services (Search and digital advertising), the closest listed analog is Meta, though Meta lacks Search and has a different growth profile. For Google Cloud, the public comps are AWS (visible through Amazon’s segment disclosures) and Microsoft’s Intelligent Cloud segment. Illustratively: Google Services at 15× EBITDA (mature growth, regulatory risk, high free cash flow conversion); Google Cloud at 20× forward EBITDA or 6× forward revenue (improving margins, strong secular growth); Other Bets assigned near-zero or a small revenue multiple as option value.

Step 4: Sum the segment enterprise values. Add up the implied value of each segment. Then subtract net debt, or add net cash if the balance sheet is net-positive. Alphabet carries substantial net cash \u2014 around $100 billion as of late 2023. That gets added to equity value, not buried in a footnote. Divide by diluted shares outstanding to get an implied share price.

Getting the segment numbers

ASC 280 requires public companies to disclose revenue and operating profit by reportable segment. That’s almost everything useful that’s guaranteed. Everything else is discretionary. Segment-level capex appears in roughly 60% of 10-Ks \u2014 typically in the segment footnote or Note 16. Depreciation by segment is disclosed less often. Working capital by segment almost never appears. This means the key inputs for a rigorous SOTP \u2014 segment-level free cash flow \u2014 usually have to be estimated, not read directly.

Where to look beyond the income statement:

• Note 16 (or equivalent segment footnote). Search for “capital expenditures by segment” and “total assets by segment.” Asset disclosure is guaranteed; capex is common but not universal. If you find both, you can cross-check capex intensity (capex as a share of revenue) against comparably structured competitors.
• MD&A segment section. Companies frequently give qualitative capex intensity guidance here \u2014 language like “Cloud infrastructure investment is expected to be significantly higher as a percentage of Cloud revenue than our advertising segment” is enough to justify a differential assumption in your model.
• Investor day presentations. This is often where the most useful segment-level economics appear. Companies use investor days to make the case that the sum-of-parts value exceeds the stock price \u2014 which means they often disclose the numbers you need to check their math. Alphabet has an annual “Google Cloud Next” conference; Comcast runs a detailed investor day that breaks out Xfinity, Peacock, and Universal separately.
• Earnings call transcripts. Analysts frequently ask about segment-level free cash flow and margin trajectory on calls. Management answers are inconsistent but sometimes revealing. If a CFO has twice deflected questions about Cloud margin, that deflection is data.

When you cannot get the number you need, make the limitation visible in your model. If you are using operating income as a proxy for EBITDA because depreciation is not separately disclosed, write it down. If you are assuming capex is proportional to revenue because the filing does not break it out, note the assumption and the sensitivity range. A model that hides its data gaps is a model that will produce false confidence. The discipline of labeling your proxies keeps the output honest.

Choosing your comparables

The multiple you assign to each segment determines the output more than any other input in the model. A one-turn difference on a segment generating $2 billion of EBITDA is a $2 billion change in enterprise value. On a company with 1 billion diluted shares, that is $2 per share \u2014 often 5–10% of the implied price from a single choice that can look like a rounding decision. This is why the argument in an SOTP is not in the arithmetic. It is in the comp selection.

The right process: identify the 3–5 most structurally similar publicly traded pure-plays for each segment. Calculate their median forward EV/EBITDA. Then ask whether the growth rate and margin profile of your segment justifies trading at that median. A cloud division growing revenue at 40% per year with a clear path to 25%+ EBITDA margins should trade at a premium to a peer growing 10% with mature margins. A cloud division growing 8% with 5% margins should probably not use the sector median at all.

For a concrete example: if you were building a Comcast SOTP in 2024, your cable/broadband segment comps would be Charter Communications and WideOpenWest (WOW). Your streaming segment (Peacock) comps would be weighted toward EV/subscriber or EV/revenue since Peacock is not yet profitable. Your theme parks (Universal) would use Six Flags, Cedar Fair, or SeaWorld as the operator comps. Each trades at a structurally different multiple. Charter trades at roughly 7–8× EBITDA as a mature cable operator. Theme park operators trade closer to 10–12× on an EBITDA basis given their asset intensity and attendance-driven economics. Using a single cable sector multiple on Comcast gives you a wrong answer by construction.

Two common failures to watch for in your own models. First, the mixed peer group: using Microsoft as a comp for a segment that has nothing structurally in common with Microsoft\u2019s margins or growth rate, simply because Microsoft commands a high multiple and a high multiple is useful for a bullish target. Second, the trailing/forward inconsistency: applying a forward multiple to the high-growth segment (because trailing looks expensive) and a trailing multiple to the mature segment (because trailing is lower and shows the business as cheap). The two moves work in opposite directions and both inflate the output. Pick forward EV/EBITDA on next-twelve-months estimates, apply it uniformly across all segments, and only deviate with an explicit, disclosed rationale.

The conglomerate discount

Build the SOTP table for most diversified companies, compare the implied price to where the stock trades, and you will typically find the stock is 10–20% below the sum. This is the conglomerate discount, and it is not irrational. Investors are pricing real risks that a sum-of-parts table does not capture.

The primary cause is capital misallocation risk. If a conglomerate has one excellent business and two mediocre ones, the market assumes management will route the cash flows from the excellent business toward the mediocre ones \u2014 because that is what conglomerate managements tend to do. GE did it for a decade before the reckoning. Investors who owned GE in 2005 were implicitly funding aviation and healthcare capital investment at financial-services leverage ratios. The second cause is complexity: a company with three distinct businesses requires three separate analyses, which produces a wider range of possible valuations, which justifies a price closer to the low end of that range.

Reading a sell-side SOTP without getting played

Sell-side analysts reach for SOTP because it gives them more inputs to tune. A single EV/EBITDA multiple is hard to reverse-engineer to a target \u2014 there is only one variable to adjust. With five segments and five multiples, each defensible on its own, you can produce almost any output while appearing rigorous. The two telltale signs that the model was built backward from the target are specific and checkable.

First, check the implied upside. Analyst buy targets cluster in the 15–25% upside range. That range is not coincidental \u2014 it is actionable enough to justify a client call without being embarrassing if the stock stays flat. When a five-segment SOTP model lands at exactly 20% above the current price, interrogate every multiple in the table. Look at the comps used for the largest segment. Check whether forward or trailing multiples are applied and whether the convention is consistent. A 1× change in the multiple on the dominant segment can easily swing the output by 8–12%.

Second, check whether every segment appears in the table. A common manipulation is excluding the money-losing division or folding it into “corporate / other” at zero value. An honest SOTP assigns a negative value to unprofitable segments \u2014 either a burn-rate multiple, a restructuring cost estimate, or an explicit note that the segment is excluded and why. If the division that has dragged operating results for three consecutive years does not appear as a negative line in the SOTP, the analyst made a choice that benefited the target.

• Are the comps pure-play or does the table use diversified companies as proxies?
• Are forward and trailing multiples applied consistently across all segments?
• Does every segment appear \u2014 including unprofitable ones?
• Does the implied price land in the 18–25% upside zone, suggesting the target came first?
• Is a conglomerate discount applied, and does the report explain why that specific haircut was chosen?
• Is net debt current, or drawn from the prior annual filing when the debt load may have changed?