Chapter 6: The business of RecSys: how platforms make money
This introduction outlines platform revenue models and the ways recommendations can support them.
After this chapter you will be able to
- Subscription renewal and recommendations. Explain how recommendations support the subscription business model.
- Transaction volume and profit cuts. Distinguish increasing transaction volume from earning a profit cut on facilitated transactions.
- Attention economy revenue. Describe how platforms in the attention economy generate revenue, including services that users can access without a subscription.
- Subscriptions and user retention. Explain how subscription platforms earn money and why recommendations matter to retaining subscribers.
- Ratings and engagement targets. Distinguish optimizing recommendations for predicted ratings from optimizing them for engagement.
- Personalized presentation and overpersonalization. Explain how tailoring an item’s presentation differs from ordering recommendations, and describe the concern about misleading artwork.
- Transaction Based Electronic Upselling. Explain how purchase histories can generate recommendations for additional items and how their placement can help drive extra sales.
- Incremental Personalized Recommendations. Explain why a useful upsell suggests something the shopper might buy but had not thought to buy, and how online recommendations can be tailored to each shopper.
- Recommendation Placement Feedback Loops. Trace how limited recommendation slots can reinforce the prominence of popular products, and describe one way designers might reduce that tendency.
- Two-sided marketplace revenue. Explain how a two-sided marketplace matches buyers and sellers and makes money from their transactions.
- Recommendations Shape Market Opportunities. Explain how recommendations make a marketplace usable while shaping the distribution of work opportunities.
- Interdependent Marketplace Network Effects. Explain why a marketplace needs enough participants on both sides and how incentives and network effects affect its growth or decline.
- Platform Power Over Workers. Explain how platform control over matching and incentives can shift power away from gig workers.
- Seller Ranking Feedback Loops. Trace how ratings and rankings can reinforce established sellers’ advantage over newcomers.
- Scarcity of attention. Explain Simon’s view of attention scarcity and its implications for information systems.
- Monetizing users’ attention. Explain the attention economy business model and how online platforms can profit from users spending more time with them.
- Recommenders and attention. Describe the two functions recommender systems serve in the attention economy and an unintended consequence mentioned in the section.
- Advertising funded platform growth. Explain how advertising supports platforms that do not charge users and why scale makes this model powerful.
- Real time ad exchange. Trace how an ad exchange matches ads to a viewer while a page loads.
- Surveillance capitalism and profiling. Explain how platforms collect and infer user information for advertising, including how a product can provide new sources of data.
- Attention and data incentives. Explain the two reasons advertising-supported platforms value the time users spend on their services.
- Cookies and cross-site advertising. Explain how first-party and third-party cookies differ and how browsing across sites can inform ads.
- Location and inferred interests. Explain how location and social connections can suggest an advertising interest without a phone necessarily listening to a conversation.
- Data broker profile building. Describe how data brokers make personal information valuable for advertising by combining data from different sources.
- Paid promotion and ranking. Explain how paying to promote an organic post affects its ranking and why this can undermine the promise of recommendation.
- Political promotion transparency. Explain why promoted political posts can make campaign funding less transparent than political ads.
- Combined platform revenue models. Explain how an e-commerce platform can combine sales revenue, sponsored placements, and upselling, and distinguish organic from promoted results.
- Shopping data business uses. Describe how a platform can use shopping data for targeted recommendations and new business opportunities, as well as sell it to a data broker.
- Ranking conflicts of interest. Explain the conflict of interest when a platform ranks content while also offering its own products, and why outsiders cannot readily verify whether rankings have been manipulated.
What it builds on
- Users’ preferences and proxy metrics (chapter 4) · The idea that platforms use proxy metrics for users’ preferences.
- Heavy ranking (chapter 2) · The role of ranking in ordering items presented to a user.
- The rich-getting-richer effect (chapter 4) · The rich-getting-richer effect in recommender systems.
- Raw and derived scores (chapter 2) · Scores represent values used by the recommender system.
Section by section
Introduction
This introduction outlines platform revenue models and the ways recommendations can support them.
- Subscription renewal and recommendations: Explain how recommendations support the subscription business model.
Why it matters: Recommendations can help a platform retain paying subscribers. - Transaction volume and profit cuts: Distinguish increasing transaction volume from earning a profit cut on facilitated transactions.
Why it matters: More purchases and a share of each transaction are distinct ways platforms can earn money. - Attention economy revenue: Describe how platforms in the attention economy generate revenue, including services that users can access without a subscription.
Why it matters: A service can be free to use while earning money from paid visibility.
Key terms: subscription model, recommender systems, subscription renewal, transaction volume, profit cut, attention economy, advertising, ad space, increased visibility
6.1 Subscription
This section explains how subscription platforms use recommendations to retain users and explores the implications of optimizing and personalizing content.
- Subscriptions and user retention: Explain how subscription platforms earn money and why recommendations matter to retaining subscribers.
Why it matters: Recommendations can help a subscription service keep customers by making its content easier to discover. - Ratings and engagement targets: Distinguish optimizing recommendations for predicted ratings from optimizing them for engagement.
Why it matters: What a platform chooses to optimize changes what it treats as a successful recommendation. - Data driven content production: Describe how Netflix used audience data in relation to House of Cards and the question this raises about creativity.
Why it matters: Audience data can inform not just what a platform recommends but also the content it produces. - Personalized presentation and overpersonalization: Explain how tailoring an item’s presentation differs from ordering recommendations, and describe the concern about misleading artwork.
Why it matters: Personalized images can influence expectations about content, not just which content people encounter.
Key terms: subscription fee, recommendations, long-term user retention, optimization target, user’s rating, engagement, proxy metrics, House of Cards, searches, viewership
6.2 Increased volume
This section explains how shopping recommendations increase sales, help shoppers navigate large inventories, and can concentrate attention through feedback loops.
- Transaction Based Electronic Upselling: Explain how purchase histories can generate recommendations for additional items and how their placement can help drive extra sales.
Why it matters: This shows how recommendations can turn shopping behavior into additional sales. - Social Proof From Purchases: Explain the indirect social proof in a statement that other customers bought a recommended item.
Why it matters: A recommendation can draw some of its persuasive power from other people’s choices. - Incremental Personalized Recommendations: Explain why a useful upsell suggests something the shopper might buy but had not thought to buy, and how online recommendations can be tailored to each shopper.
Why it matters: A recommendation is most useful for extra sales when it changes what someone considers buying. - Reduced Shopping Cognitive Load: Describe how recommendations help shoppers deal with inventories too large to browse fully.
Why it matters: Recommendations can save shoppers effort as well as encourage sales. - Recommendation Placement Feedback Loops: Trace how limited recommendation slots can reinforce the prominence of popular products, and describe one way designers might reduce that tendency.
Why it matters: The way scarce recommendation slots are assigned can concentrate attention on a few products.
Key terms: electronic upselling, bought together, placement, social proof, recommendations, tailored, upsell, cognitive load, inventories, feedback loops
6.3 Profit cut
This section explains how transaction-based marketplaces make money and how recommendations, network effects, and feedback loops shape their markets and participants’ opportunities.
- Two-sided marketplace revenue: Explain how a two-sided marketplace matches buyers and sellers and makes money from their transactions.
Why it matters: The transaction cut connects the platform’s revenue to its role as an intermediary. - Recommendations Shape Market Opportunities: Explain how recommendations make a marketplace usable while shaping the distribution of work opportunities.
Why it matters: Marketplace recommendations do more than order choices; they affect who gets business. - Interdependent Marketplace Network Effects: Explain why a marketplace needs enough participants on both sides and how incentives and network effects affect its growth or decline.
Why it matters: A marketplace’s success depends on keeping both sides engaged, not just attracting one group. - Platform Power Over Workers: Explain how platform control over matching and incentives can shift power away from gig workers.
Why it matters: The system that connects workers to opportunities can also leave them dependent on decisions they cannot see or control. - Seller Ranking Feedback Loops: Trace how ratings and rankings can reinforce established sellers’ advantage over newcomers.
Why it matters: Rankings can perpetuate an advantage that is difficult for equally capable newcomers to overcome.
Key terms: two-sided marketplace, transaction value, cut, recommender systems, liquidity, work opportunities, two-sided market, incentives, network effect, gig workers
6.4 Attention economy
This section explains attention scarcity, how platforms monetize users’ time, and the role of recommender systems in the attention economy.
- Scarcity of attention: Explain Simon’s view of attention scarcity and its implications for information systems.
Why it matters: This explains why managing attention, not just providing information, is a central design problem. - Monetizing users’ attention: Explain the attention economy business model and how online platforms can profit from users spending more time with them.
Why it matters: It connects a platform’s incentive to hold attention with the ways it earns money. - Recommenders and attention: Describe the two functions recommender systems serve in the attention economy and an unintended consequence mentioned in the section.
Why it matters: This shows how recommendations can both direct attention and expand the time available to monetize.
Key terms: attention economy, scarcity of attention, users’ profiling, ads, promoted posts, recommender systems, users’ attention, addiction, social anxiety
6.4.1 Ads and users’ surveillance
This section explains how advertising-funded platforms profit from targeted ads, user surveillance, and sustained attention, and examines the risks of those incentives.
- Advertising funded platform growth: Explain how advertising supports platforms that do not charge users and why scale makes this model powerful.
Why it matters: A platform can be free to use while still making money from its users. - Evolution of ad targeting: Distinguish earlier audience targeting from the web's ability to microtarget ads at scale, and explain why reaching relevant buyers is valuable.
Why it matters: The web changes the precision and scale of targeting, not the basic desire to reach likely customers. - Real time ad exchange: Trace how an ad exchange matches ads to a viewer while a page loads.
Why it matters: The exchange turns user information and advertisers' preferences into an ad shown in the moment. - Surveillance capitalism and profiling: Explain how platforms collect and infer user information for advertising, including how a product can provide new sources of data.
Why it matters: Ordinary use of a service can reveal information even when a user does not explicitly provide it. - Attention and data incentives: Explain the two reasons advertising-supported platforms value the time users spend on their services.
Why it matters: Time spent matters to platforms both as an immediate revenue opportunity and as a source of better targeting. - Harmful targeting and accountability: Explain how granular ad categories can be harmful and why their existence raises questions about platform transparency and accountability.
Why it matters: Without outside scrutiny, harmful targeting options may remain invisible to the public.
Key terms: ad-supported, scale, surveillance, microtarget, ad targeting, relevant buyers, ad exchange, real-time bidding, allocated budget, surveillance capitalism
6.4.2 Tracking across the internet
This section explains how cookies, location data, and data brokers enable advertising based on activity beyond a single platform.
- Cookies and cross-site advertising: Explain how first-party and third-party cookies differ and how browsing across sites can inform ads.
Why it matters: This helps explain why an ad on one site can reflect something you browsed elsewhere. - Location and inferred interests: Explain how location and social connections can suggest an advertising interest without a phone necessarily listening to a conversation.
Why it matters: A surprisingly relevant ad can result from combining information rather than necessarily recording a conversation. - Data broker profile building: Describe how data brokers make personal information valuable for advertising by combining data from different sources.
Why it matters: It shows why even a site selling a product may have an incentive to collect information unrelated to the sale.
Key terms: cookies, first-party cookies, third-party cookies, location, over-surveillance, data brokers, profiles
6.4.3 Promoted content
This section explains how paid promotion changes post rankings, how promoted posts differ from ads, and why their lighter scrutiny can reduce political transparency.
- Paid promotion and ranking: Explain how paying to promote an organic post affects its ranking and why this can undermine the promise of recommendation.
Why it matters: Paid promotion can put what someone paid to show above what the system would otherwise rank most valuable to you. - Promoted posts versus ads: Distinguish promoted posts from ads in their usual targeting, format, and campaign goals.
Why it matters: The distinction helps explain when a creator might use a promoted post rather than a more customized ad. - Political promotion transparency: Explain why promoted political posts can make campaign funding less transparent than political ads.
Why it matters: Similar political messages can face different scrutiny depending on how they are promoted, making campaign money harder to trace.
Key terms: promoted content, organic posts, ranking, scores, promoted posts, ads, engagement-based metrics, promoted post, political ads, transparency
6.5 Mixture of models
This section explains how platforms combine revenue models and use recommender systems and data in ways that can create conflicts of interest and competitive advantages.
- Combined platform revenue models: Explain how an e-commerce platform can combine sales revenue, sponsored placements, and upselling, and distinguish organic from promoted results.
Why it matters: Knowing how revenue streams overlap helps explain why a highly ranked product is not necessarily the best one. - Investor pressure and product quality: Explain how investor demands can encourage platforms to seek new revenue while cutting costs and product quality, and describe the resulting strain on sellers.
Why it matters: This connects platform business decisions to the experience of both users and dependent sellers. - Shopping data business uses: Describe how a platform can use shopping data for targeted recommendations and new business opportunities, as well as sell it to a data broker.
Why it matters: Data can support several parts of the same business model rather than serving only one purpose. - Ranking conflicts of interest: Explain the conflict of interest when a platform ranks content while also offering its own products, and why outsiders cannot readily verify whether rankings have been manipulated.
Why it matters: The platform's control over visibility can benefit its own products even when customers cannot see how rankings were determined. - Data advantage and regulation: Describe how access to marketplace data can advantage a platform's own product development and state the regulatory separation discussed in the section.
Why it matters: The data a platform gathers as an intermediary can also give its own products an advantage over sellers.
Key terms: organic content, promoted content, sponsored content, upsell, investors, revenue streams, enshittification, data flywheel, shopping behavior, targeted recommendations
6.6 Summary
Where this is used later
- (chapter intro) (chapter 8) · Provides the distinction between engagement and predicted liking for the chapter’s discussion of engagement.
- Enshittification (chapter 9) · The section develops the discussion of platforms reducing incentives after becoming established.
- Power over users, and lack of contestability (chapter 10) · The discussion of platform power over users builds on examples of dependence and limited control.
- Meme wars (chapter 7) · Platforms' attention-economy business model
- Rabbit holing (chapter 8) · The attention economy model
- The power of incentives (chapter 10) · How platforms make money in the attention economy
- Feedback loops and extreme content (chapter 8) · Its discussion of feedback loops and extreme content builds on the role of recommendations in engaging users.
- News (chapter 7) · Advertising and the attention economy
Concepts explained in this chapter
- advertisementPaid promotional communication distributed to an audience to advance a product, service, cause, or other objective.
- user profilingThe construction of inferred attributes, interests, and characteristics about users from collected data and interactions.
- ad exchangeAn automated marketplace that matches available digital advertising placements with advertiser bids and audience criteria.
- ad-supported platformAn online service financed primarily by displaying advertisements rather than charging users directly.
- microtargetingThe practice of directing advertisements to narrowly defined audiences using detailed personal, demographic, or behavioral information.
- real-time biddingAn automated auction process that selects an advertisement for an available placement in response to a user's opportunity to view it.
- surveillance capitalismAn economic system in which behavioral data gathered through monitoring people is treated as a source of prediction, influence, and profit.
- recommendationA content or item suggestion generated by a recommender system for a particular user.
- subscription business modelA pricing model in which users pay a recurring fee for access to a service or its content.
- attention economyAn economic environment in which capturing and monetizing users' limited attention is a primary source of value and revenue.
- cookieA small data file stored by a web browser to preserve website-related state or record activity.
- cross-browsing data linkageThe linking of a person’s browsing records across different websites or devices.
- data brokerA business that acquires, aggregates, analyzes, or sells data about individuals or organizations.
- first-party cookieA browser cookie issued and accessed by the website that created it.
- third-party cookieA browser cookie associated with a domain other than the website currently being visited and capable of enabling data sharing across websites.
- recommender systemA computational system that selects, ranks, or presents items to users based on inferred preferences or other criteria.
- critical massA sufficient level of participants or activity at which a platform becomes self-sustaining or substantially more valuable.
- incentiveAn arrangement of rewards, penalties, and expected benefits that shapes participants' choices and behavior.
- liquidityThe ease and speed with which a marketplace can match participants and enable transactions.
- two-sided marketplaceA marketplace platform that connects two distinct participant groups whose interactions depend on each other, such as buyers and sellers.
- personalizationThe adaptation of item selection, ordering, or presentation to an individual user's inferred characteristics or preferences.
- retentionA measure of how successfully a platform keeps users returning or remaining active over a specified period.
- user engagementThe degree to which users interact with or remain attentive to content or a platform.
- promoted contentPaid content distributed by a platform to increase the reach or prominence of an otherwise regular post.
- online platformAn internet-based service that provides digital functionality and connects users, content, or transactions.
- behavioral dataData recording users’ actions and interactions, including their activity while browsing websites.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.
