Clew’s Deep Dives offer a closer look at some of the most loved search-related products.
A brief-ish history of the company
Every startup has a crisp, neat origin story. The one Reed Hastings tells of Netflix goes like this: He gets slapped with a $40 fine by Blockbuster for his late Apollo 13 rental. Annoyed by the crappy service, he thinks as every founder struck with an idea does, there’s got to be a better way. Netflix, a content behemoth, employs the power of narrative to draw its audience of talent, partners, and investors. But in truth, it was never clear in the beginning that Netflix was a good idea. The history of the video-streaming platform holds unfamiliar characters, complicated storylines, and a bit more mess than the edited final cut.
A key character to introduce is Marc Randolph, Netflix’s initial CEO in the late-nineties when the company was born. Back then, He and Hastings lived in Santa Cruz and belonged to the same ride-share group. By chance, the two wound up sharing a lot of car rides to work, during which they’d pitch and kill startup ideas. Randolph understood Hastings to be sharp, intense, and respected for the impressive career successes under his belt. The pair tended toward businesses that sold things on the internet. Often, it’d be Randolph suggesting things like custom sporting goods, personalized surfboards, or individually formulated dog food. Hastings would then tear into the business model and outline point for point why the idea wasn’t up to snuff. The idea to rent movies online actually came up in a car ride, but was thrown out. At the time, VHS tapes were the dominant format and the economics weren’t attractive because shipping costs were too high for the bulky packages. Netflix was not a genius epiphany and in fact, took weeks and months to emerge as something that looked even vaguely promising.
It wasn’t until the pair noticed the quiet rise of DVDs, which were a fraction in size and weight of VHS tapes, that they reconsidered the idea of a movie-rental business done online. The idea appeared wholly underwhelming to anyone else looking from the outside, with many people telling Randolph, “that will never work.” When they pitched it to a VC firm, one of the partners who happened to be French, said to Hastings and Randolph quite matter-of-factly: “This is sheet.” From his point of view, the big jump in the movie business would be from analog to digital. DVDs were just a stepping stone toward the inevitable age of video-streaming.
Investor: “Why would I invest in a company that won’t be around five years from now?”
Randolph and Hastings agreed on that point, but understood something the investor missed; studios were the gatekeepers and they were betting on the DVD format. Building a working business model and industry relationships preemptively would be a huge advantage come a time when the world shifted.
Hollywood’s problem with VHS was that video stores like Blockbuster had established themselves as the middlemen. Since tapes were being rented out dozens of times to consumers, the studios felt compelled to hike up the prices of tapes to capture their share of the value created in the home-viewing market. Studios bet big on the cheaper DVD format because it encouraged audiences to buy and build an at-home library. But one thing Hollywood execs were not bullish on was the idea of putting content online. They’d witnessed the mess of what happened to the music industry and were understandably skeptical about enabling consumers to digitally access films. Another hurdle slowing the transition to streaming included premature infrastructure since high-speed internet wasn’t reliable nor widely accessible. For as long as studios could muster and internet speeds remained slow, it was going to be the golden age of DVDs.
So despite the neat origin myth told of Netflix, it wasn’t an obvious business to Randolph and Hastings, nor was it terribly exciting to investors. In fact, the first $2M in funding came from Hastings drawing $1.9M from his own pocket and Randolph having many uncomfortable conversations with friends and family asking for money.
The days leading up to the launch of Netflix
When the funding finally came together, the work began. Netflix needed a comprehensive library of all DVD titles, a site for handling orders, and a process for fulfillment. If they wanted to be the single best place to rent DVDs, customers had to be able to find what they wanted, when they wanted it. Netflix would build an inventory of all films ever released and hold multiple copies of the popular titles. Since it wasn’t a physical store customers could browse, there had to be a way to explore titles on the website. For customers to be able to find something they wanted to watch, Netflix needed to collect heaps of data on each individual movie; objective facts like who the director was, which actors were involved, the date of release as well as subjective data like the film’s mood and genre. To do this from scratch would require manually collecting and inputting all this data for each title they carried in inventory.
Luckily, there was a guy named Michael Erlewine. He was, as described by Randolph, a hoarder who found a way to monetize his obsession. Erlewine collected information on movies. He’d enlisted a team of people to find movies, watch them, and annotate every detail. The result was an absurdly large and thorough database. Netflix needed this database if it had any hope of launching on time so Randolph cut a deal with Erlewine. Netflix would satisfy Erlewine’s cinephilic obsession by providing the man with a copy of every DVD title ever released in return for access to the data he collected. The team at Netflix manually scanned the cover images of all 926 DVD titles they had in stock and used Erlewine’s database to fill in the details. On April 14, 1998, they launched Netflix.
In the first 15 minutes of the site going live, orders were flooding in. There were only two servers in the office, and though a good problem to have, they both crashed. The team rushed out and bought eight more to quintuple the capacity. With new servers set up and running, it wasn’t 45 minutes before they all crashed again. By then, the team wasn’t just running out of servers, but also boxes, tape, paper, and ink. The first day proved that people wanted Netflix, but it also revealed a ton of problems. In terms of product, the team needed a way to automate more of the process, smooth the users’ online ordering experience, and steer people to lower-demand titles. More importantly, since Netflix was giving customers the option to purchase DVDs outright, few people actually took the rental option, resulting in a huge revenue disparity between DVD rentals and sales.
The Subscription model and Cinematch
Randolph and Hastings knew that selling DVDs online was a commodities business and a company like Amazon could and would take over soon enough. Though Netflix was enjoying revenues from the DVD sales segment, it was in the company’s long-term interest to focus on rentals because it was less competitive, more operationally challenging, and defensible. The decision to double-down on the rental segment set the course for Netflix’s future.
Problem was, the rental program at the time relied on customers having the foresight to plan what they wanted to watch, order in advance, and wait to receive the title in the mail. The team needed to think of something fast because customers who rented once didn’t rent from Netflix again. The team came up with three ideas. First, Netflix would allow customers to rent four DVDs at a time, keep the rentals as long as they wanted, and swap a return for a new title. Second, customers could add movies on the site to their queue. Netflix would then automatically mail queued titles following the return of watched rentals. Third, they’d make all this possible with a subscription model, which was uncommon at the time, where users would pay $15.99 a month to use the service. When Netflix posted a banner on the site to promote the new rental model, 90 percent of people who clicked through and provided their credit card information on the first day.
Slowly, Netflix started taking shape as the streaming service we know and love today. The missing part of course was the rough strokes of how personalization worked in the early days. When the company first launched, there were 926 titles available, but by the end of 1999, the DVD library grew to 5,000 titles. It became increasingly difficult for users to browse options. There wasn’t enough resources to personalize for each user manually so Hastings suggested listing a limited number of recommended titles on Netflix’s homepage. Faced with an abundance of choice, users needed an efficient way to find movies. Hasting’s solution also meant that users could be directed to interesting albeit less popular titles which would optimize Netflix’s inventory. In order to show users what they would be most interested in, Netflix used a process called “collaborative filtering”; similar to how Amazon makes suggestions based on customer’s common purchase patterns. But to make these recommendations, it wasn’t enough to know which titles customers ordered. Netflix introduced a 5-star rating system to find out if customers actually enjoyed the movie. All this accumulated to the launch of Cinematch, Netflix’s personalization algorithm, in February of 2000.
Survival of the fittest
By September 2000, Netflix raised a $50M Series E, bringing its total funds raised to over $100M. The company held a content library of over 5,800 titles, had 200,000 paying customers, and made shipments of over 800,000 discs each month. When the tech bubble burst, the startup graveyard was littered with direct-to-consumer businesses like Dr.Koop, Boo.com, and Webvan who’d operated in adjacent industries. Unlike them however, Netflix had a working business model and positive unit economics. Still, an unfortunate side effect of growth was that Netflix found itself perpetually cash-strapped as it offered free trials as part of its user acquisition strategy. This was terrible timing because VC dollars were drying up. During this dire period, Hastings and Randolph seriously considered selling the business to John Antioco, the CEO of Blockbuster. During a meeting to discuss the potential acquisition, Hastings proposed a figure of $50M. Hearing that, Antioco apparently tried to hide his laugh. The deal obviously never went through, and Blockbuster is nowhere to be seen today.
By late 2001, Netflix reached 500,000 users, while still bleeding cash. In order to survive, the company made mass layoffs; a sort of Darwarnian culling that would leave only the superstars who’d work harder knowing they were alongside the best. Indeed, the company, who’s published a 100+ slide deck detailing the nature of its workplace, has been forthcoming about its high performance culture. On slide 24, it reads: “We’re a team, not a family.” Early hires were given the autonomy to solve hard problems and to exercise their own judgement. When an engineering manager asked Randolph for permission to leave early on Fridays to visit his long-distance girlfriend on weekends, Randolph simply said to him:
“Look, where and when you work is entirely up to you. If you can run your group effectively on three and a half days a week in the office, all power to you. Go ahead—I’m envious. Wish I was smart enough to do that. Just remember, you’re a manager. Part of your job is making sure that your team knows what you want them to accomplish and why it’s important. Do you think you can do that without being around?”
Apparently the employee broke it off with his partner soon after. That kind of exchange demonstrated what the company referred to as radical honesty, and it wasn’t exclusive to junior employees. Randolph himself got a healthy dose of it when Hastings sat him down one day and presented a Powerpoint presentation outlining why Randolph was no longer fit to lead as CEO. Up until then, Hastings had taken a backseat in the company, offering the bulk of the initial investment and serving an advisory role. The company had reached a certain scale however, that Hastings found himself increasingly involved. Though certainly a hard pill to swallow, Randolph relented, knowing that Hastings was better suited to lead the fast-growing team.
With Hastings in command, an outcome-obsessed culture in place, and a masterplan to lead Hollywood through a digital transformation drawn up, Netflix as we know it had taken shape. Netflix officially launched its streaming service in 2007.
Financials and business model
Netflix was revenue-generating since Day 1, a rare feat for any technology company. On May 22, 2002, it filed its S-1 in preparation for an IPO later that month. The DVD player proved to be one of the fastest-selling consumer electronics devices in history; holding a place in 62 percent of American television households. At that point, the company was the largest entertainment subscription service in the US with over 600,000 subscribers and a library of more than 11,500 titles.
From the first day of launching the site in April of 1998 to January of 1999, Netflix’s revenues came primarily from DVD sales, with a fraction from rentals. In order to focus on the rentals, the company made the difficult decision to kill their DVD sales business. A few months later, in September 1999, Netflix launched the subscription model and began building out revenue-share agreements to scalably acquire content with lower up-front cash payments. . The secret sauce was its use of technology in improving user experience and business operations. Cinematch made it possible for Netflix to maximize the utilization of its inventory, putting to use 11,000 out of the 11,500 titles available. In fact, 73 percent of customer orders were from the back catalogue. As the number of users and user-generated ratings increased, the accuracy of personalized recommendations only improved. The result was a fly-wheel effect, drawing in more users, retaining them, and improving the experience further.
The risks outlined in the S-1 included excessive user churn, the business’s dependence on the US mail system, and anticipating demand for content. In the first business quarter ended March 31, 2001, the churn rate averaged 10 percent. By the first quarter of 2002, the average fell to 7 percent, which the company predicted would remain the average churn rate over the lifetime of a new subscriber. The concerns relating to the challenge of physical delivery and accurately ordering enough DVDs of a particular title were unique to that time. In the most recent annual report for the year ended December 31, 2018, much of the business has changed and the fear has since shifted to the cost of licensing and producing content in an increasingly competitive landscape.
By the end of 2018, Netflix grew to a team of 7,100 full-time employees and a reported 139M paid memberships in over 190 countries. By 2019, Netflix has spent more than 70 percent of its revenues on content; the total for the year estimated to reach $15B—more than any other media company. And it’s no surprise to hear that the business expects $3.5B in negative free cash flow. Since interest rates have been low and wealthy investors have been keen to park their money in high yield investments, Netflix has managed to tap into debt markets to fuel its spending. But accessing cash won’t remain easy forever and there’s no promise that marginal investments in content will yield the same return for the company. In Q2 of fiscal 2019, the company saw a drop in the total number of US paid memberships for the first time and half the expected number of additional global subscribers. The news wiped out $17B from Netflix’s market cap.
When Netflix first started streaming, it seemed nonsensical to spend what appeared to be a lot of money at the time on content-licensing. Now, the very studios who handed over their content thinking it a bargain, realize Netflix was the real winner from the exchange. Having smartened up, studios and content providers are developing their own streaming services and keeping their work products exclusive to their own platforms. As the competition stiffens, the price for programming is likely to increase. This hasn’t deterred Netflix, who happens to be the biggest spender of them all, taking on substantial debt to license and produce content. The company had $10.4B in senior notes outstanding and $8.4B in total content liabilities as of December 31, 2018. Netflix has no plans to cut back on spending, though it does recognize the risk that the long and fixed nature of its obligations will bring a world of hurt in the event of an economic downturn.
The one thing Netflix has over every one of its competitors is time, time spent collecting, learning and building features off of user data (much like Google Maps’s advantage over Apple).
Cinematch, Netflix’s earliest recommendation engine, was created while the company was still in the physical DVD rental business. Relying on item popularity wasn’t sufficient and the intention wasn’t to find a universal indicator of what was relevant, but to determine which titles best satisfied a user’s unique tastes. When the company launched its instant streaming service in 2007, the shift didn’t just change the way users consumed content, but it also created all types of new data describing user behaviour. Some of the data sources used in the ranking algorithm included the following:
- Viewer ratings measured on a 5-star scale
- Item popularity
- Stream play context (duration, time of day, device type, etc.)
- Items added to queue
- Content metadata (actors, director, genre, parental rating, etc.)
- User interactions (scrolls, mouse-overs, clicks, etc.)
- Search terms
- External data (box office performance, critic reviews, etc.)
The best indicator of subscriber satisfaction was total viewing time and the more time people spent on Netflix, the better the month-to-month subscription retention. The goal of the ranking algorithm then, was to show users content they’d enjoy so as to maximize content consumption. Netflix doubled down on its data-driven approach, even ponying up a $1M cash prize to anyone who could improve the performance of its Cinematch algorithm by 10 percent.
As Netflix grew and scaled globally, personalization became more robust, evolving to support things like content localization and personalized artwork. Not only would Netflix be showing users the most relevant content, but it would showcase titles in the most compelling way possible. Given there’s a 90-second window to capture a user’s attention, it’s necessary to show artwork and previews that accommodate for factors like viewer’s preferred language, watch history, and personal taste. By managing the way titles are presented to viewers, Netflix didn’t just shift views from one film to another, but it actually increased aggregate viewing times and engagement. The early versions of artwork personalization relied on A/B testing and measuring engagement metrics like click-through-rate and play duration for each variant. Later, it became a matter of finding the single most compelling artwork for an individual user. Here’s an example of how Netflix changes the visuals for Good Will Hunting to suit two different users:
User 1: Jane’s search history shows she’s interested in standup specials.
User 2: Jack’s has a penchant for romantic comedies.
Making this happen at scale is a huge technical challenge. Even the seemingly simple task of curating static images from a film or show is complicated. A single season of a TV show has an average of 9M frames. Dedicating a team to manually select the most promising frames is tedious and difficult to scale. Instead, Netflix uses a special ranking algorithm to select the most promising subset of frames to be used as artwork. Every frame is annotated to capture data like the following:
- Visual metadata such as brightness, colour, contrast and motion blur.
- Contextual metadata such as using face detection for pose estimation and sentiment analysis.
- The type of camera shot which provides insight on the tone or mood of the frame.
- Object detection to determine if there are any meaningful non-human features in the frame.
- Composition metadata relating to standard principles used in film and photography like the rule-of-thirds, symmetry, and depth-of-field.
All this data is then parsed to surface frames that meet certain criteria, fit for promotional artwork. A single film streaming in 190 countries would require visual variants for every type of user and for over 30 languages. Uncovering a bit of what goes into artwork personalization reveals the scope and complexity of what happens at Netflix to deliver a viewing experience that keeps subscribers coming back.
What’s the future of Netflix?
Even with 7,500 high-performance, data-driven employees working around the clock, achieving member satisfaction, retention, and impressive user acquisition is becoming increasingly difficult. Aswath Damodaran, a finance professor at NYU’s Stern School of Business commented on the difficult times ahead for the streaming platform:
“For a decade, Netflix has spent more and more money on content to get users and increase market capitalization, and it worked. But the question is: how do you get off this treadmill? At some point, spending 75 percent of every dollar on content won’t be sustainable.”
Besides the risk of a credit crunch, Netflix has to fend of the likes of Amazon, Disney, and Apple who are none too small, deep-pocketed, and looking to play the streaming game of thrones. The company’s shift to creating original content may be more than a reaction to competitors reserving exclusive rights. Netflix has been heralded as the platform where quirky, off-beat stories that traditional studios would never have backed, are told. The company is in a strong position to leverage all the data it collects to personalize a viewer’s experience all the way through, not just post release of content, but from the very beginning of the film-making process. It’s toyed with the idea already, like in the Black Mirror episode Bandersnatch which offered an interactive viewing experience where viewers could make a choice to affect the storyline.
Netflix has undoubtedly changed the way people consume content. Maybe it’s time for the streaming giant to innovate on story-telling, the very product it delivers. Or maybe not. Maybe that’ll never work.