(This is an “Appendix” to a multi-part series answering the question: “Who will win the battle to make the next Game of Thrones?” Previous articles are here:
Part I: The Introduction and POCD Framework
Appendix: Licensed, Co-Productions and Wholly-Owned Television Shows…Explained!
Appendix: TV Series Business Models…Explained! Part 1
Appendix: TV Series Business Models…Explained Part 2
Appendix: Subscription Video Economics…Explained Part 1)
The best analogy for content libraries on streaming services, for me, is theme parks. When I tried to value the new Star Wars land Galaxy’s Edge at Disneyland and Disney World, I wrote about this future scenario:
Next year, I’ll walk into Disneyland in the off-season (probably September-ish). I’ll be wearing a Star Wars shirt. My brother will probably rock a Marvel shirt. That said, I’ll also have a four year old wearing, if current trends hold, either an Elsa (Frozen) or Belle (Beauty and the Beast) dress. Other family members will likely have Mickey shirts on.
So how much of that trip do you allocate to the opening of Galaxy’s Edge? My family already averages one trip to Disneyland every year, and my daughter knows that Mickey lives at Disneyland. So she’d go anyways. But what about me? I’ll definitely go to see the new park at some point.
Something about theme parks—maybe the permanence of the attractions—helps crystallize in my head the challenge of valuing content libraries. A theme park is a content library of rides, shows, shopping and food. Some of those attractions at Disneyland have been there since the 1960s. Those are the “library content” of Disneyland. Others are only one or two decades old. Those are the “recent library” of rides. Then there are the brand new attractions: Star Wars land, Cars land and a Guardians of the Galaxy ride. Those are the “new TV” of Disneyland rides.
The trouble is trying to value each of those pieces and disentangle them. At the end of the day, this both matters—because you need to make the best decisions possible to maximize revenue—and doesn’t—because at the end of the day the goal is to have revenues exceed costs on a total basis. Do the latter and how you get there doesn’t really matter.
My approach to valuing theme parks—calculating the money spent by both existing and new customers—gives us a good idea for how to value content libraries on streaming platforms. So let’s explain that. In today’s article…
– The rules guiding my approach to valuing content
– The “dream method”, which is what we’ll try to emulate
– The steps to the optimal method
– The HBO and Game of Thrones example explained
– Some other variations, caveats and thoughts
As I wrote these last two articles, I kept coming back to the “rules” that define good business models. A few stuck in my head for valuing streaming video. Thinking that way…
– First, no double counting. If a customer gets attributed once to a piece of content, they don’t get to count twice. (A good rule of thumb, you can’t attribute more than 100% of your customers!)
– Second, CLV trumps monthly revenue and other calculations. If you attract a new customer, CLV is the best way to capture their true value to your business.
– Third, be humble in attributing success. No single show or movie accounts for 100% of its viewers in a library model.
– Fourth, use real data as much as possible.
The Dream Method – The Probability of Resubscribing
The dream method for HBO would be, basically, to be God Almighty. Looking down omnipotently, reading the mind of every customer subscribed to HBO and knowing why they subscribed, and what percentage of that should be credited to Game of Thrones. Add all the percentages together and you have it. (Maybe our Google/Amazon/Apple AI overlords will be there soon…)
In the meantime, we have data. Especially streaming data if you’re Netflix, Amazon or (partially) CBS or HBO.
This data means you can track every customer. When their account starts. When it renews. When it lapses. And, crucially, what they watch the entire time. From the people who only watch movies to the people who complete every episode of Game of Thrones. In a big data sense, then you can compare their behavior to the customer who never watched Game of Thrones.
Say the results looked like this…
…GoT Viewers resubscribe after a year period at a 92% rate.
…non-GoT Viewers resubscribe after a year period at a 80% rate.
That means, of customers who started the year subscribed to HBO, by watching GoT, they were 12% more likely to stay subscribed to HBO. That’s the best number if you can find that, because it basically means that GoT increases the probability of staying subscribed by a huge, statistically significant margin. Now that GoT is cancelled, if those GoT watchers suddenly flee HBO, well we can also reverse engineer that to know that GoT had been keeping them subscribed.
This could also be applied to new customers. If you take all the new subscribers for a given time period, you can look at the ones who watch GoT versus the ones who don’t and model their behavior. You can also tell which are the customers signing up to watch GoT right away, and which ones don’t. Add those up and you can attribute all the best approximation for value we have. (With heaping doses of regression analysis and machine learning.)
Yet, we don’t have the big data to do this. I mean me, as a commentator on the strategy of entertainment. If I were managing content strategy at a streaming company, I would set a team of data scientists working on. But I don’t have that team or that data here. As an outside observer, well, we need to make some assumptions, but we can try to replicate that method.
My Method – Attributing New and Remaining Customers by CLV
So what I call the “optimal method” for valuing content starts by dividing customers into regions. We’re being fairly disingenuous when we pretend like each part of the world values content equally. Sometimes they do; often they don’t.
After that, I divide customers into new customers and current customers. This is key because some content will help attract new customers and some will help retain customers, meaning making them more likely to resubscribe. (But don’t go too crazy with this assumption. The best content tends to score well on every metric, including attracting and retaining customers.)
It’s also easier because most companies release subscriber numbers, so you can assume the growth is the new subscribers, and the rest are retained. (Though even this isn’t really accurate.
Next, calculate the “CLV” or “customer lifetime value” of a given customer. This is difficult to get right, and nearly impossible without data. (Learning point for the flixes: you have ZERO idea what your customer lifetime value is. Your models will have assumption but only the true data of customers will tell you for sure.) My inputs for the cable channels are fairly simple:
Then comes the big assumption: how many of those new and retained customers are attributed to a piece of content. For a good deal of shows—especially those going one season and getting cancelled—it could easily be zero percent. Or less. Anyways, try to use ratings data to triangulate, and then, you basically have this equation:
Total Value = (% of Existing Customers * CLV) + (% of New Customers * CLV)
Repeat by year. So it is actually pretty simple, but let’s do an example just to see it in action, and to explain one of the more complicated parts of my Game of Thrones article.
Game of Thrones Subscriber Estimates (as an example)
I wish I could have spent more time digging into this number, and explaining myself, but I had a (self-imposed) word limit for Decider. Fortunately, today I have all the space in the world to explain myself.
So let’s use my method above. First, I divided the world up into the three known parts. That’s domestic, international and digital. I’d have gone smaller by territory, but that’s what HBO gave me in their 10K statements. So that gave me this subscriber table:
Now, these subscribers numbers have some qualifications and caveats. Since it’s in HBO’s interest to obscure numbers as much as SEC-abettedly possible, as of 2015, when they launched HBO Now over-the-top, they just added digital only subscribers to the domestic total. Then they strategically leaked how many digital subscribers they have over the years at various conferences or investor calls. So I subtracted digital from domestic to get domestic growth. Oh, and HBO combines Cinemax and HBO subscribers, so I assumed most of the growth was for HBO, not Cinemax; that’s an assumption and I like to call them out when I make them.
To to put that table together required grabbing all the 10Ks, reviewing multiple news stories and clarifying that everyone was talking about the same time periods. Then, I made the growth numbers. These are key, because they’ll drive the “new subscribers” number.
Next came the time for “customer lifetime values”. The prices were fairly easy. HBO has been about $15 in the US for a long time, and the same digitally. However, it only keeps half (roughly) of each dollar form MVPDs (cable, telco and satellite companies) but it keeps up to 80% of digital sales. This is because—as Apple showed—HBO is the must have premium TV service. (That isn’t Netflix.)
Then came some qualifications. International revenue is much lower than domestic since it is in territories with much smaller purchasing power. Like Latin America or Eastern Europe. Most of the big and rich European countries are covered by licensing deals. So as I was looking for prices, I found a lot of international OTT prices for 6 Euros and under. So | put that price pretty low. There is also the digital side of it. Sure, HBO keeps a lot more of each month’s pay, but they also lose customers at a much higher rate. Check out this table from Second Measure:
Yikes. That’s some churn. So I dropped the number of months for the average digital subscriber way down from my domestic assumption. This also allowed me to attribute way more subscriber growth to GoT for digital than I would for traditional pay TV, which also makes sense, because again, look at those spikes. Add all those up and some genuine guesses for acquisition costs got to:
(BTW, if you think I did some part of my math wrong, send me your assumptions and I can update the model. I’m happy to take input. Crowdsource his baby!)
Now comes the last set of “magic numbers”—which I’ll explain in a future article. There are four the percentages. That’s the number of retained customers and new customers by territory and/or digital. Here:
That could be six different numbers, but I settled on 2% for all retained customers. (Again, do you disagree? Send me your percentages. I’ll update the model and send the results out via Twitter.)
Let’s pause on this numbers. That’s saying that of HBO’s 49 million or so domestic customers (which again, includes a lot of Cinemax subscribers) only about million are due to Game of Thrones. Doesn’t that feel…really low?
It is. But using my experience—having seen the data version of this run above—it isn’t outrageous. At all. Especially when you’re attributing their entire lifetime value to this measurement. A person subscribes for all sorts of reasons: library movies, new movies, new TV shows, watching reruns \and, don’t forget, just plain old inertia. That’s right: some people will subscribe to HBO and it almost doesn’t matter what is actually airing; they’re rich and will just add it to the bill. That inertia could be 50% of returning customers. Add in 20% of first run movies, and well you “only” have 30% to play with, and with a two year CLV, you only have 15% to play with!
Yeah, 2% is small, but it adds up to a lot. Using the 2017 number as an example, that’s still $177 million, which more than pays for the season that aired, and that’s just the United States. Also, since 24 months is the CLV, some of the new customers are still “paying” for GoT in the second year on the service. That means about 4% of a year’s customers are due to GoT.
The new customers were harder, but I wanted to keep them a bit higher, given that this series got so buzzy. So it went up to 10% for domestic and 5% internationally (which could even be too low) and way up to 50% for digital. Again, that Second Measure data up above told me that 50% was imminently reasonable for digital. Especially for the churners, hopping in and out just for Game of Thrones.
So here’s my final table, and I pulled in the columns in to make it a bit easier to read then the Decider article.
As I was figuring out the attribution percentages and magic numbers above, I stared at these viewership percentages. The growth for Game of Thrones is just incredible, going to 43 million total viewers by this season. So as a sanity check, I saw what percentage of the growth both digital and pay TV in the US I attributed to Game of Thrones:
That seems pretty reasonable and sound. Basically, under 10% of GoT’s US total viewership at anytime is attributed to their bottom line. If anything, I could be low here.
Some Thoughts, Learning Points and Caveats
In real life, a lot of shows will get “0%” value.
This is my twice weekly reminder that value is not distributed equally or normally in the entertainment business. It is winner take all, and yes Game of Thrones was taking it all. Here’s my explainer on “logarithmic distribution of returns” and some examples here.
Yeah, I called this “sub-optimal” back in January.
I had two reasons for this. First, I thought it would be too hard to calculate for future series. I’ve changed my mind on that. This method done well requires lots and lots of data, but for streaming video it is our best appraoch. When it came to the Star Wars and Marvel films, we didn’t have that for Netflix. But more importantly, for things in second windows, Netflix is paying real money, so that’s the best approximation of value. At least, for that analysis.
“Cohort Groups” are the next big thing.
The next improvement to this model is to compare groups of customers who are like other groups of customers. So people who signed up in the same year; people who watch mostly the same shows; people who watch the same amount of content; nearly identical demographics because the big tech companies know exactly who you are. That’s cohort analysis and it’s buzzy because it’s more accurate than treating everyone the same.
But it takes even big data to another more complicated step. At this higher level, though, that’s false precision, not accuracy.
Conclusion: Phew, We’re Done with the Background
Now that I’ve explained how to make money both as a streamer and as a TV producer—on the individual series level—we can now dive back into our candidates for “the next Game of Thrones”, understanding their individual business models and, crucially, the opportunity of each series.