// recommendation algorithms

Source: Lewis C. Lin’s Newsletter

Netflix's algorithmic recommendations drive four-fifths of viewing behavior, making the choice of success metric existentially important—but the company struggles to distinguish between competing measurement approaches that perform almost identically. Which metric Netflix optimizes for determines whether it prioritizes engagement depth, retention, or discovery breadth, each with different implications for content strategy and subscriber lifetime value. The inability to decisively measure what's working reveals a deeper tension in recommendation systems: the metrics that are easiest to quantify (clicks, time spent) often conflict with the business outcomes that matter most (sustainable satisfaction, reduced churn).

Source: Lewis C. Lin’s Newsletter

Netflix's recommendation engine faces a choice: prioritize completion rates (a metric that shows user stickiness) or prioritize discovery and catalog diversity (which can reduce churn and boost long-tail content). Optimizing for one actively suppresses the other. A completion-focused algorithm narrows suggestions toward similar content; a diversity-focused one risks surfacing content users abandon, which tanks completion metrics that Wall Street monitors. Most platforms default to completion because it's measurable and immediate. But studios, creators, and subscribers have incentives tied to the underutilized 90% of Netflix's catalog.

Source: Rich on Tech Newsletter

Roku is adding algorithmic recommendations directly to its TV interface, moving beyond passive content discovery into predictive viewing. The approach mirrors Netflix's model but reaches viewers across 70+ million devices regardless of subscription service. Roku is positioning itself as a discovery layer above fragmented streaming—where the real money lies as linear TV advertising collapses and platforms compete for attention and ad inventory.