Why firms move into new businesses, when it pays, and whether Amazon has spread itself too thin
Diversification is the second question of corporate strategy: what range of products and services should the firm offer (Rothaermel, 2024, p. 319)? The evidence is unkind to both extremes. Firms that stay in one business and conglomerates that own unrelated ones tend to do worse than firms with moderate, related diversification (p. 328). So the test of a new business is never whether the firm can enter it. It is whether owning it makes the business, and the firm, worth more.
Amazon started in 1995 selling books online. By 2024 it also ran a cloud business with over $90 billion of revenue, an ad network with $47 billion, a film studio, a grocery chain bought for $13.4 billion and a primary-care company bought for $3.9 billion (case, pp. 1–3). The case opens on Andy Jassy deciding whether to keep entering “seemingly unrelated businesses” or to focus on a few (p. 1). Put bluntly: is Amazon spreading itself too thin, or are the businesses more related than they look?
What this session answers
Facts from the Gupta and Rodriguez case (rev. July 2024) are marked Case; where this page reads the case through a framework, the card is marked Analysis. Textbook pages refer to Rothaermel (2024), 6th edition.
Carried forward · Sessions 10 to 12
Corporate strategy sets the firm’s boundaries along three dimensions: vertical integration, horizontal diversification and geographic scope (Rothaermel, 2024, p. 298). The last three sessions built the tools and walked two of the axes. Today is the horizontal one, and it pulls the other two together.
Session 10 · the tests
Session 10 gave you Porter’s three tests: attractiveness, cost of entry, and better-off, where either the new unit gains advantage from its link with the corporation or the corporation gains from the unit (1987, pp. 7–8). Executives ignore the third test most (p. 11). Today it is the one that decides each Amazon business.
Session 11 · vertical
Session 11 read Amazon’s vertical moves: its own delivery, devices as a sales channel, and AWS as the infrastructure the store runs on. Those moves change how much of retailing’s value chain Amazon owns. Today’s question is different: which other businesses Amazon should be in.
Session 12 · geographic
Session 12 took the third axis: which countries, and how much to adapt to each. The textbook files geographic expansion under diversification too, and a firm doing both kinds follows a product–market diversification strategy (Rothaermel, 2024, p. 320).
One move, two axes
The same move can sit on more than one axis. The textbook uses AWS as its example of both horizontal diversification and vertical integration (p. 297). Session 11 read AWS as infrastructure Amazon makes for itself; today it is a business that sells to other companies, with over $90 billion of revenue and over $24 billion of operating income in 2023 (case, p. 1). Keep the two readings apart: one asks whether to make an input, the other whether to be in a market.
The case’s closing line
With so many businesses, the case says, Amazon was defying the basic principle of focus. It ends by asking whether the fundamentals of strategy have changed in the digital era or whether Amazon is heading for failure (p. 4). Hold your answer until the scorer near the end.
Types of diversification · Rothaermel (2024), pp. 319–323
A firm diversifies when it widens the range of products and services it offers or the regions it serves. Several product markets is a product diversification strategy; several countries, a geographic one; both at once, a product–market diversification strategy (p. 320). To say how diversified a firm is, the textbook uses Richard Rumelt’s scheme, which asks two things: what share of revenue comes from the main business, and how far the units share core competencies (p. 321).
Type 1
More than 95% from one business
The rest is not yet significant. Birkenstock makes one product; Meta earns almost all its revenue from online ads (p. 321).
Competencies: used in the one business
Type 2
70–95% from one business
At least one other activity earns the rest. About 80% of Harley-Davidson’s revenue is motorcycles; parts, merchandise and logo licensing make up the remainder (p. 321).
Competencies: shared by the main and minor businesses
Type 3a · related
Under 70% from one business
A new business is entered only if it can use existing resources and competencies. ExxonMobil took its oil exploration skills into shale gas by buying XTO Energy (p. 322).
Competencies: generally shared across the businesses
Type 3b · related
Under 70% from one business
Some businesses link to the main one, others do not. The textbook’s own example is Amazon (p. 322), listed with Disney and GE (p. 323).
Competencies: some businesses share them
Type 4 · unrelated
Under 70% from one business
Few if any links. Samsung, Berkshire Hathaway, Yamaha and the Tata group are the textbook’s conglomerates (pp. 322–323).
Competencies: few if any shared
The bars only illustrate the revenue thresholds; like the textbook’s own schematics, they are not drawn to any firm’s numbers. The first variable separates types 1, 2 and the rest; the second separates related from unrelated (p. 321). Text version: Rothaermel’s Exhibit 8.10, p. 323.
The textbook’s verdict on Amazon
Rothaermel traces Amazon from one product, books, to a wide range of goods built on its online retailing capabilities, then a platform open to outside sellers, cloud computing, devices and streamed content for Prime members, and files it under related-linked (p. 322). The label claims that the links exist but are uneven. The Amazon section below tests that claim one business at a time.
Case
The case does not report revenue by business, so Rumelt’s 70% line cannot be applied exactly. It does give three anchors for 2023: total revenue of $575 billion, AWS above $90 billion and advertising above $47 billion (pp. 1–2). That puts AWS at roughly 16% of revenue and advertising at roughly 8% (this page’s arithmetic).
Analysis
Rumelt’s second variable is not “same industry”. It is whether units share competencies in products, services, technology or distribution (p. 323). A cloud service and an online store look unrelated by industry code. Whether they are related depends on what they actually share, which is a question about Amazon’s resources from Session 5.
Motives · Rothaermel (2024), pp. 299–300, 334–335; Porter (1987)
The textbook gives five reasons firms grow: to increase profitability, lower costs, increase market power, reduce risk, and motivate managers and employees (p. 299). It then grades them, and they are not equal (pp. 334–335). Sorting a motive is the first step in judging a diversification move, because the same deal can be defended with a good reason and pursued for a bad one.
Value-creating
Increasing profits and lowering costs are clearly tied to competitive advantage (p. 334). For diversification the route is economies of scope: two or more outputs produced together for less than each would cost alone, with the same resources and technology. The textbook’s example is Amazon offering many product categories off one retailing expertise and one network of fulfilment centres (p. 302).
Only under conditions
More market power can add to advantage but can bring antitrust action (p. 334). Growing to reduce risk has fallen out of favour with investors (p. 334): shareholders can diversify their own portfolios more cheaply, without paying acquisition premiums, so risk reduction should be a by-product, not the motive (Porter, 1987, pp. 7, 11). An internal capital market adds value only if headquarters allocates capital better than outside markets would (Rothaermel, p. 332).
Managerial
Some growth benefits managers more than shareholders: empire building, job security, corporate jets and retreats at expensive resorts. CEO pay often tracks firm size more closely than firm performance (p. 300), and the textbook calls perks and job security illegitimate reasons to grow (p. 335). Porter’s version: executives confuse company size with shareholder value (1987, p. 11).
Activity · sort the motive
Pick a category for each statement. The feedback gives the textbook’s or Porter’s reasoning. Two of them describe moves Amazon made.
Diversification and performance · Rothaermel (2024), pp. 328–333
The question to ask of any diversified firm is whether its businesses are worth more under its management than they would be if each were managed on its own (p. 328). That is Porter’s better-off test in other words. The research the textbook summarises answers it by type of diversification.
Low and high diversification are generally linked to lower performance, moderate levels to higher performance (p. 328). The textbook adapts the curve from Palich, Cardinal and Miller’s review of three decades of studies.
Diversification discount · pp. 328–329
Unrelated diversifiers are often valued at less than the sum of their units. GE Capital earned half of GE’s profits on a third of its revenue before the financial crisis and nearly sank the group when credit froze. When GE announced the spin-out in 2015, its stock rose 11%, adding some $28 billion of market value: a rough measure of the discount it had carried.
Diversification premium · p. 329
Related diversifiers are, on average, valued at more than the sum of their units. To earn it, diversification must lower costs through economies of scale, raise value through economies of scope, or do both (p. 329).
The discount depends on institutions · p. 329
The discount holds where capital markets are developed, as in the United States. In emerging economies such as India, unrelated groups like Reliance can gain advantage by filling gaps in capital markets and courts. The business-groups deck under Go deeper takes this further.
Related diversification
Sources of value
Economies of scope; economies of scale; financial economies from restructuring and internal capital markets.
Sources of cost
Coordination costs, which rise with the number, size and types of linked businesses; influence costs.
Unrelated diversification
Sources of value
Financial economies only: restructuring and internal capital markets.
Sources of cost
Influence costs: managers manoeuvring for capital and resources, and the misallocation that follows.
Paraphrased from Rothaermel, Exhibit 8.14 (p. 330) and pp. 332–333. Related diversification can create more, but it is also the only one that pays coordination costs: the links have to be managed.
Question mark
Low share, high growth
Cash flow negative. Build share or harvest and divest.
Star
High share, high growth
Earnings high. Hold, or invest for growth.
Dog
Low share, low growth
Earnings low, unstable. Harvest or divest.
Cash cow
High share, low growth
Cash high and stable. Invest enough to hold.
Relative market share → (low to high)
The firm is treated like a portfolio of stocks. Stars tend to become cash cows as their market matures; question marks become stars or dogs. Manage the portfolio clockwise (p. 332).
The outline’s two articles
Goold and Luchs (1993) trace four decades of management thinking on why firms diversify; Laeven and Levine (2007) ask whether the diversification discount also applies to financial conglomerates. Neither text was in the course-pack folder this page was built from, so the page does not summarise them. Read both for class and test the textbook’s claims above against them; full references are under Readings.
Internal capital markets
Headquarters knows more about its units than outside investors do, so it can, in principle, send capital to the best returns, sometimes at a lower cost. That creates value only if it allocates better than external capital markets would. GE’s AAA-rated finance arm lowered its cost of capital until the crisis cost it the rating in 2009 (p. 332).
Restructuring and spin-outs
Restructuring refocuses a firm on its core competencies: spin out weak units, and free strong ones that were subsidising them under extra layers of bureaucracy (p. 330). eBay spun out PayPal; Johnson & Johnson and GE split themselves up in 2023–24 (p. 331).
Analysis
Do not fill the missing axis from memory or the news. A matrix with invented data looks rigorous and is not.
The case · Amazon in 2024 (Gupta & Rodriguez, 2024)
“An online retailer” does not cover a cloud provider, a film studio, a grocery chain and a primary-care company. The case calls the businesses seemingly unrelated (p. 1); the textbook’s own Amazon case asks you to classify them with Rumelt’s types and to say whether AWS is related to the core at all (Rothaermel, 2024, p. 334). Take them one at a time and ask four things: what was the motive, how related is it, which resource does it share, and does Amazon add value as its owner? The first three are below; the fourth is the scorer in Spreading too thin?
The core competence–market matrix · Rothaermel (2024), pp. 324–326
Hamel and Prahalad’s matrix, as the textbook adapts it, crosses existing or new core competencies with existing or new markets. Each quadrant is a different diversification move:
Session 10 introduced the matrix with NVIDIA. Here it does a second job: the further a business sits from the lower-left, the less it borrows from what Amazon already does well.
Your map so far
New · existing market
Build to protect
New · new market
Mega-opportunity
Existing · existing market
Leverage
Existing · new market
Redeploy
Market → (existing to new)
Place businesses with the cards below and they appear here.
Activity · place and justify
Step 1: choose a quadrant. Step 2: choose the shared resource that justifies it. Facts on each card come from the case; the placements in the feedback are this page’s reading of those facts. Several businesses have two defensible placements, and the feedback says so.
Is AWS related to retailing?
The textbook’s answer is yes: Amazon had to build huge data centres for its peak holiday demand and put the spare capacity into cloud computing, gaining economies of scope and scale (Rothaermel, p. 322). The case shows the other side: AWS sells elastic IT infrastructure to many companies, in a fast-growing market where large technology players compete hard, and Bezos kept investing while analysts questioned it (p. 1). The shared resource is real; the customers and the rivals are not retail’s.
The two sources date AWS differently: 2006 in the textbook (p. 297), 2002 in the case (p. 1). This page follows the case.
Case
Case, pp. 2–3. Exits are part of a diversification strategy, not proof that it failed.
Analysis
The three businesses with the largest sales figures in the case, AWS, advertising and Amazon Business, all redeploy competencies Amazon already had into new markets. The top row, where Amazon had to build new competencies, holds the long bets and the write-offs alike: Echo, Whole Foods, One Medical, the Fire Phone, the salon. That is the textbook’s related-linked pattern: a dense core of links and a periphery with few.
Competitive advantage from diversification · the case; Rothaermel (2024), chs. 7, 8 and 12
Each business puts Amazon against a different specialist. In cloud computing, large technology players compete hard (case, p. 1). In video, Amazon spent over $19 billion on TV and music content in 2023 against Netflix’s $13 billion, and its app store rivalled Google’s and Apple’s (p. 2). In business supplies it met W.W. Grainger and regional family-run wholesalers; in grocery, Walmart (p. 3). A generalist beats specialists only if something it owns travels from market to market. Read the case for what travels and three candidates stand out.
01
AnalysisThe textbook’s own example of economies of scope is Amazon: one network of fulfilment centres lets it offer many product categories at lower cost than each would cost alone (p. 302). The same promise travels in the case: free two-day shipping launched Prime (p. 1), and 24-hour delivery set Amazon Business apart from regional wholesalers (p. 3).
02
AnalysisThe textbook names Amazon’s proprietary AI-based recommendation algorithms as a core competency (p. 301). In the case, recommendations go back to Amazon’s early years, the same shopping data sold over $47 billion of advertising in 2023 (p. 2), and AWS sells the infrastructure itself (p. 1). The textbook’s point: a search engine knows what people search for; Amazon knows what each person views and buys (p. 334).
03
AnalysisMore than 200 million Prime members spend over four times what other customers spend, and they kept joining as the fee rose from $79 to $139 (case, pp. 1–2). New businesses reach them through it: Prime Music in 2014 (p. 1), streamed content free to Prime subscribers (Rothaermel, p. 322), One Medical at $9 a month for Prime members (case, p. 3).
Network effects · Rothaermel, ch. 7
Network effects arise when a product becomes more valuable as more people use it (p. 259). Platforms that match producers with consumers live on them, and markets with strong network effects can tip to a single winner (pp. 276, 281–286); Session 9 covered the chapter. Amazon opened its store to outside sellers in 2000; by 2023 independent sellers made more than 60% of its sales (case, p. 1). More sellers widen the selection that draws shoppers, and more shoppers draw sellers.
Complements · Rothaermel, ch. 12
In the razor–razor blade business model, the first product is sold cheaply, even at a loss, to drive demand for the complementary goods that earn the money (p. 479). Kindle launched at $399 and fell to about $100 after the iPad, as a convenient way to buy Amazon’s digital books and films; Echo takes orders by voice; the Fire Phone’s Firefly button identified objects to buy on Amazon (case, p. 2). Each device is a way into the store.
Analysis
The textbook defines economies of scope as cost savings from producing several outputs together (p. 302). Network effects and complements work on the value side instead: a business is worth more to the same customer because of the others. That changes the test for each link. Ask whether a specialist rival in that one market could copy it. A rival can build a cheaper e-reader; it cannot easily build a store, a delivery network and a membership to go with it.
Case
Launched in 2005 as two-day shipping for $79 a year, Prime rose to $99, $119 and then $139 in 2022, and membership kept growing. By 2023 it had more than 200 million members, who spent more than four times what other customers spent (case, pp. 1–2).
Analysis
Amazon spent over $19 billion on TV and music content in 2023, more than Netflix’s $13 billion, and runs a studio that makes films (case, p. 2). Read as a stand-alone streaming business, that budget has to be earned back from viewers. Read as a complement to Prime, it is part of what members get for a fee that kept rising while they kept spending. The case gives you the facts; which reading you choose decides how the unit should be judged.
Analysis
Kindle sold out at $399 in 2007 and was cut to about $100 after the iPad (case, p. 2). The textbook adds that Amazon priced new bestsellers at $9.99 as e-books to promote the Kindle (Rothaermel, p. 480). At different times the device and the books were each priced to sell the other. Either way the profit, if there is one, sits in the store, not in either product alone.
Analysis
Porter’s better-off test runs both ways: the new unit gains from the corporation, or the corporation gains from the unit (1987, p. 8). Video and devices pass it, if at all, mainly the second way. Judged alone, each may look like a loss-maker; judged by what it does for the store and for Prime, it may be the point. Links like these have to be managed, and the textbook warns that coordination costs rise with the number and variety of linked businesses (p. 332). So how should the centre judge a unit whose job is to make another unit sell more?
In June 2017 Amazon paid $13.4 billion, a 27% premium over the chain’s trading value, for 460 stores and high-income customers. Experts asked whether a technology company with a soaring share price should enter a thin-margin industry at all (case, p. 3). Put the deal through Session 10’s tests.
Attractiveness
Grocery is large, about $800 billion of US spending, but highly competitive, with razor-thin margins and most grocers, Whole Foods included, struggling (case, p. 3). Porter asks for a structure that supports returns above the cost of capital, or an entrant able to change it (1987, p. 8).
Cost of entry
A 27% premium for a chain with 1.2% of the market against Walmart’s 14.5% (case, p. 3). Porter’s test is that the entry price must not capitalise all the future profits (p. 8); the premium has to be earned back by something Amazon adds.
Better-off
On its own, Amazon had won no more than 0.2% of grocery in ten years, and AmazonFresh had limited success (case, p. 3). Stores, customers, delivery and Prime could make each side better off; palm-scan payment in some stores from 2021 is one visible link (p. 3). The case gives no evidence yet of the payoff.
The verdict tags are this page’s reading of the case against Porter’s tests; argue with them in class.
Decision · the better-off test, business by business
Relatedness alone does not settle it. A business can share a resource with the rest of Amazon and still be worse off inside it. So score each business twice: how strong its link to the rest of Amazon is, and whether it, or Amazon, is better off because Amazon owns it. The professor’s business-groups script names four ways a parent can add value, drawing on Goold, Campbell and Alexander: influence on the unit as a stand-alone business, linkages between units, central functions and services, and corporate development such as acquisitions. Use them to judge the second score.
Activity · portfolio scorer
Link to Amazon · 0 nothing visible in the case · 1 thin, such as a price discount · 2 a real shared resource · 3 it runs on Amazon’s core resources
Better off inside Amazon · 0 worse off · 1 no visible gain · 2 a clear gain in one direction · 3 a strong gain for the unit and for Amazon
Parenting bet
Weak link, owner adds value
Core: keep and integrate
Strong link, owner adds value
Exit, sell or spin off
Weak link, no gain
Fix the economics or shrink
Strong link, payoff unproven
Link to Amazon → (2 or more is strong)
Portfolio verdict
Score at least five businesses to get a verdict.
“This page’s reading” is one defensible set of scores built from the case and the textbook; each row explains it. The verdict rule is this page’s heuristic, not a formula from the readings: a business needs 2 or more on both scores to count as core.
The case for “well positioned”
The case for “spreading too thin”
Course synthesis · Sessions 1 to 13
The course asked two questions. How does one business win in its market? And which businesses should a firm own? The second depends on the first: diversified companies do not compete, their business units do (Porter, 1987, p. 6).
Part 1 · business & competitive strategy
Part 2 · corporate strategy
The thread. Every corporate move is judged with business-level tools. A new business must sit in an attractive industry (Sessions 3–4), draw on a real resource (Session 5), and win in its own market (Sessions 6–9). Porter’s three tests from Session 10 are the course in one sentence.
Next on the course calendar: Reflective Note 2 is due by this session, and the end-term examination follows it.
Go deeper
These interactive decks were built before the 2026-27 outline. Use them for practice; where a figure or example differs from the readings, the note on each card says so.
Deck · 1-year MBA
Built for an earlier course numbering, so its own title says Session 8. Five growth drivers, Rumelt’s four types, the inverted U, the competence–market matrix, a strategy simulator and a 15-question check.
It says Apple captures 95% of the “profits”; the textbook says about 95% of the value created (p. 314). Its matrix examples (Honda, Gallo) are not the textbook’s.
Deck · 2-year MBA
Built for the two-year course, where it is Session 14. Covers transaction costs, vertical integration, related and unrelated diversification with Indian examples, a discount-versus-premium chart and a three-decision CEO simulation.
Its Amazon figures for 2025–26 come from outside the readings, and it credits the performance curve to Goold and Luchs (1993), listed there under Goold and Campbell; the textbook credits Palich, Cardinal and Miller (p. 329).
Deck · 2-year MBA
Despite its file name, this is the business-groups deck. Why the conglomerate discount that is the default in developed markets often turns into an advantage in India, with the Tata group and a “Does this parent add value?” tester.
Pairs with the institutions point on p. 329 of the textbook and with the second score in today’s portfolio scorer.
For class
01
What business is Amazon in? Give one definition under which AWS, Prime Video and Whole Foods all belong, then name the business your definition cannot explain.
02
The textbook calls Amazon related-linked. Which of its businesses break the link? Would Amazon’s shareholders be better off without them, and how would you know?
03
If Prime Video’s job is to keep Prime members paying and shopping, how should Amazon measure it? Should Kindle be judged on its own profit, given how Amazon has priced the device and the e-books?
04
Put One Medical through Porter’s three tests, using only what the case reports. Does it fare better or worse than Whole Foods?
05
The textbook’s chapter case says some investors want AWS spun out as a stand-alone company, and asks whether it is worth more inside Amazon or alone (Rothaermel, p. 334). Argue both sides with the discount, the premium and the better-off test.
06
You are Jassy. Name one business to exit and one to double down on. Which motive would a sceptical board member attribute to each choice: value-creating, conditional or managerial?
Knowledge check
Answer before you look. Each answer explains itself, and the explanation matters more than the score. The full practice bank for this session is at /teaching/quiz/1yr-08.
Readings & preparation
Readings are obtained through the course pack. They are not linked or hosted here.
Gupta, S., & Rodriguez, M. L. (2024). Amazon in 2024 (HBS Case No. 9-514-025, rev. July 16, 2024). Harvard Business School Publishing.
Goold, M., & Luchs, K. (1993). Why diversify? Four decades of management thinking. Academy of Management Executive.
The course outline lists it under Goold and Campbell. Read for class; this page does not summarise it.
Laeven, L., & Levine, R. (2007). Is there a diversification discount in financial conglomerates? Journal of Financial Economics, 85(2), 331–367.
The course outline gives volume 82 (2006); the published article is volume 85 (2007). Read for class; this page does not summarise it.
Rothaermel, F. T. (2024). Strategic management (6th ed.). McGraw Hill.
Chapter 8, “Corporate strategy: Vertical integration and diversification”: pp. 298–302 and section 8.4, pp. 319–333. The page also uses ch. 7 (network effects and platforms) and ch. 12 (business models).
Porter, M. E. (1987). From competitive advantage to corporate strategy. Harvard Business Review, 65(3).
Today uses its three tests and premises. Page numbers follow the 32-page HBR.org print.
Before class
Read the case (four pages of text) and list every business it mentions with its year. Then read the textbook’s section 8.4. The chapter’s opening case is also about Amazon, with figures for 2021–22 that differ from the case; for numbers, use the case.
Also cited on this page
The professor’s Further study teaching script, for the session’s framing, and his business-groups script, for the four ways a parent can add value used in the portfolio scorer.