How Netflix Uses AI to Reinvent Its Business Model
Executive Insight | November 20, 2025
Netflix began life in 1997 as a DVD-by-mail start-up founded by Reed Hastings and Marc Randolph, a direct challenge to the late-fee economics of Blockbuster and the limitations of physical retail. The oft-repeated origin story, that a hefty “Apollo 13” late fee inspired the company, has since been debunked by Randolph, whose account highlights a more systematic search for an online retail category and a proof-of-concept mailing of a disc to test the model. Netflix launched its website in 1998, then spent the next decade iterating toward a subscription service with no due dates, before making its defining pivot to streaming in 2007, initially a modest “Watch Now” add-on for PC users. The red-envelope era officially ended on 29 September 2023, when Netflix mailed its final DVDs, closing a 25-year chapter that helped finance and de-risk the streaming pivot. In the early days, Jeff Bezos explored acquiring Netflix for roughly $14–$16 million, Hastings declined, and the company doubled down on building a direct-to-consumer platform that would later reinvent TV with binge releases such as House of Cards (2013).
Today, Netflix is a diversified entertainment platform with 302 million paid memberships at year end 2024, and from 2025 it no longer reports quarterly subscriber counts, now monetising through a mix of subscriptions, an ad-supported tier, live events, licensing, games, and physical experiences. The ads plan reaches more than 94 million monthly active users. Recent moves include a multi-year deal to stream the NFL’s Christmas Day games and becoming the home of WWE Raw in multiple markets from January 2025, alongside a growing portfolio of mobile games and the launch of “Netflix House” venues, with Philadelphia opening on 12 November 2025 and Dallas due to open on 11 December 2025. Financially, momentum remains strong with Q2 2025 revenue up 16 per cent year on year and an operating margin of 34 per cent, evidence that Netflix’s business model innovation is now as much about monetisation breadth and operating leverage as it is about content.
An AI-ready, data-native platform
When leaders say they want to be “the Netflix of our industry”, they are rarely talking about films. They are pointing, often loosely, at a model in which algorithms mediate the relationship with customers, business decisions are tested rather than guessed, and the product changes continuously in response to data.
Netflix is interesting not because it is digital, but because it is AI-native. Personalisation, pricing, content strategy, and even engineering decisions are shaped by models and experiments at scale. For executives trying to move beyond pilots and proofs of concept, it offers a concrete example of what it means to embed AI and analytics into the architecture of a business model, rather than bolting them on at the edge.
At present, Netflix operates less as a catalogue and more as a personalised discovery system. Netflix has long indicated that the vast majority of viewing on its service is driven by its recommendation system rather than by manual search, which is why it invests so heavily in personalisation. That system is not a single algorithm but a layered set of models that decide which rows of titles to show, how to rank them, and what artwork to display for each member, on each device, at each moment.
Initially, Netflix’s recommendation engines used relatively simple techniques. If people who watched similar things to you liked a particular title, that title moved up your feed. Over time, the models have become much richer, taking into account the time of day, the device being used, who is likely to be watching in a given household, language, and even content traits such as pacing, tone, and cast. More recently, Netflix has started to replace many of these separate systems with a single, large “foundation model” for personalisation, trained on billions of viewing interactions so it can learn a shared picture of both member preferences and the attributes of each piece of content.
Netflix now reuses this foundation model across its main personalisation surfaces, so the homepage, search results, recommendations and notifications all draw on the same underlying representation of each member and each title. The model learns a shared “profile” for people and content, and different product teams plug into that profile in different ways: some use pre-computed scores to rank rows on the homepage, others embed the model as a component inside their own systems, and others fine-tune it for specific tasks such as search relevance or messaging. This turns the foundation model into a common intelligence layer rather than a standalone experiment, which reduces duplicated modelling work, keeps the experience more consistent across touchpoints, and shortens the path from a new personalisation idea to a tested and scaled feature.
A culture of experimentation: from data to decisions
Netflix relies on large-scale online experiments to decide which product and policy changes reach its members. New recommendation algorithms are A/B tested against incumbents; alternative home page layouts and sign-up flows are tried on live traffic; pricing tiers and account-sharing rules are piloted in selected markets; and different combinations of artwork, trailers, and notifications are run head-to-head to see which variants measurably shift viewing, retention, and revenue. All of this runs through an internal experimentation platform and a dedicated experimentation and causal inference team that treats controlled A/B testing as the default way to ship changes.
More recently, Netflix has started to treat experimentation as a portfolio problem. Work on return-aware experimentation and optimising returns from experimentation programmes frames each test as an investment decision, where scarce traffic and engineering time are directed to ideas with the highest expected impact, and success is measured by the cumulative lift in engagement, retention and revenue generated across many launches, rather than by how many individual experiments happen to show a statistical “win.”
Three questions to shape an AI strategy that delivers business impact
Netflix’s use of AI matters not because it is novel, but because it shows how data, experimentation, and modelling can be organised around clear commercial objectives. To translate that into your own context, it helps to step back from individual tools and focus on three questions about decisions, data, and capabilities.
1. Which business decisions, challenges, and opportunities should define how we use advanced analytics and AI?
An AI strategy earns its place when it is clear which decisions it is meant to improve. The starting point is to set out the concrete challenges and opportunities in the current business strategy. Which parts of the customer journey can be enhanced and optimised? Where are new revenue pools and business models emerging? For each of these, identify the specific decisions that matter, such as which customers to prioritise, how to configure capacity and networks, and where to take or reduce risk.
The question then is where advanced analytics and AI can materially change the outcome of those decisions. The outcome should be a short, explicit list of decision areas, each tied to a strategic requirement and a quantified business effect, where it is worth investing in building and deploying models rather than treating AI as a broad capability in search of a problem.
2. What data, experimentation, and model design do those decisions require?
Once the priority decisions are clear, the next step is to work backwards and specify what is needed for those decisions to be made intelligently and repeatedly. For each decision area, define the key variables that must be understood. Examples include customer behaviour, content, or product characteristics. Then identify where, in existing products, channels, and processes, that information should be created and maintained. That often means redesigning how interactions are logged, how operations are tracked, and how commercial data is structured so that high quality, analysis ready data is produced as part of everyday business activity.
In parallel, be explicit about which approaches are required across the analytics toolkit. Descriptive analysis clarifies what is happening today. Predictive machine learning estimates what is likely to happen next. Prescriptive analytics and optimisation translate those estimates into concrete choices. Causal analytics and controlled experiments establish which actions genuinely change outcomes. Across all of them, there needs to be a clear plan for how models will be built, validated, monitored, and explained, including how explainable AI will be used so that product owners, legal, and risk teams can understand and challenge model behaviour when required.
3. What capabilities, structures, and controls do we need to run advanced analytics and AI as a core business capability?
The first two questions define what matters and what it technically requires. This one is about whether the organisation is set up to deliver it consistently. That means deciding which shared platforms will underpin data management, experimentation, and model operations, and how teams will work around them. In practice, it involves putting in place an environment where product owners, data scientists, engineers, and operational leaders can jointly frame decision problems, build and test models, and move successful approaches into production systems without rebuilding the basics each time. It also means clarifying roles. Who owns an AI enabled decision, who is accountable for model performance, and who maintains the underlying data and infrastructure.
At the same time, advanced analytics and AI need clear principles and controls. Governance has to cover how models are approved, monitored, retrained, and retired, how experiments and causal analytics are used to demonstrate real effect on business outcomes, and how explainable AI is applied so that regulators, risk functions, and senior leaders can understand and challenge model behaviour when required. Alongside this, there is an investment question, which skills are missing, what training decision makers need to interpret model outputs responsibly, and how incentives will reward evidence based decisions rather than intuition alone. Answering this third question establishes AI as an operating discipline that can be managed, audited, and improved over time.
From insight to implementation: leading with AI in your organisation
These questions are a useful filter for any AI conversation, whether it starts with Netflix or with a problem closer to home. They shift the focus from tools to business use cases, from abstract data issues to the way processes generate data every day, and from isolated pilots to a repeatable way of building and scaling solutions. The real work now is to apply them inside your own organisation, in the parts of the value chain where AI can lift performance and efficiency.
Our AI for Competitive Advantage Executive Masterclass is designed to turn these questions into action. Over the programme, executives work with concrete cases and structured frameworks to identify high-value use cases, and build a practical toolkit for leading AI initiatives that strengthen the business model and sharpen the organisation’s competitive position.