Saturday, August 22, 2015

[FFXIV] Extrapolating the Subscriber Count

Recently Square Enix released their yearly Eorzea Census for 2015 and it has some really interesting numbers. They've also talked about how they have over 5 million registered accounts now. Syncaine believes this directly correlates to active subscribers, but at first blush I'd say that assumption is unsubstantiated hogwash.

However, thinking on it further, I'm not actually completely sure off-hand. So between the census and some other "registered accounts" numbers I figured I'd perform some numerical analysis to see if he's really off-base or not.

First of all, let's talk about the numbers Square Enix has released. We know they were relatively flat at 2M accounts registered from April 2014 to 2.5M accounts at December 2014. Then they jumped like crazy, reaching 4M nearly 3 months later in February 2015. Finally, they've hit over 5M in August 2015.

I still think it's fallacious to directly correlate accounts registered with active subscriptions, however, if we keep in mind FFXIV is still early in its lifecycle, and the meteoric rise in accounts created relative to the previous year, I imagine that slightly more than doubling their registered accounts in a year probably means that they've more than doubled their active subscribers. For example, if you're not generating new accounts, you're extremely likely to be losing active subscribers over time. Natural attrition as folks decide they have played enough and drop their subscription. By that logic, it's likely that the more recently someone has registered an account, the more likely it is that someone is still subscribed.

That's not a perfect assumption--there are plenty of counter-examples--but it would support Syncaine's theory to a certain extent. So why don't we take a look at the Eorzea Census 2015 data under the assumption that registered accounts is equivalent to active subscribers and see if it makes sense.

Side note: I'm going to work under the assumption of one character per account. This isn't strictly true, but given FFXIV's ability to be whatever you like on a single character, removing the necessity for alts that other games generally have, I think this is an okay assumption to make to give us an upper bound of subscriber number estimates.

Total Playing Hours

The first metric I want to look at is total playing hours of all characters. Nearly 97B hours of play time since the inception of the game. Thankfully, they give us the total playtime a year prior, which reduces this number to just under 96B hours. Seriously, like two orders of magnitude difference, which is actually pretty nuts.

It's also interesting, because the prior year their registered accounts were relatively flat from launch through August 2014. There's no way those 2M registered accounts only accounted for 823M play hours if 2.5 times that number of registered accounts were actively playing for 118x that amount by the end of the next year. This by itself would clearly show that registered accounts does not equal active accounts.

96B hours played is about 8B a month, or 266.66M a day. That would require 11M characters playing 24H a day for the full year. There is literally no way that is correct. Square Enix must have misplaced a decimal point here because that would require every registered account to have 2 characters logged in constantly.

If we assume they did misplace a decimal point, that brings us to 1.1M characters playing 24/7, or 3.3M characters playing 8 hours a day. Or 6.6M playing 4 hours a day. The 6.6M is clearly impossible (as they only have 5M registered accounts), so either Square Enix has a really impressively addicted fanbase compared to the previous year, or something is seriously wrong with the scripts harvesting their numbers. These values don't make sense.

Last year's values make sense though, because 823M hours played would mean about 0.56M characters playing 4 hours a day, and that jives with their suggestion that they had over half a million players logging in daily. A number that would make far more sense for 2015 would be in the ballpark of 5.2B total hours played, as that would mean 3M character logging in daily for 4 hours a day plus the 823M from the previous year.

This just doesn't add up. The data is clearly flawed.

Dungeon and Raid Runs

217,817 characters have entered Alexander: Gordias, the LFR-esque version of the latest raid. If Square Enix truly had 5M active subscribers, this number would be an abysmal failure. 4.35% of the population, with 0.7% ever setting foot in the Savage version. For a raid that's been out for a month and a half, that's a pretty small number.

If we look at stats for WoW's Blackrock Foundry raid nearly 2 months in at the time, we're talking about 36% of players who completed the first wing on any difficulty.

Granted, FFXIV moves a lot slower than WoW does, and there are plenty of folks who aren't yet max level, even 2 months in, but if we use that value as a basis then we're looking at a lower bound of 605,000 subscribers.

Dungeons run similarly shows a low number for potential subscribers unless an extremely significant portion of the population has no combat jobs above 50 period. Sohm Al is required for the main story past 54, and yet, it's only been attempted 450,000 times. This means, at a maximum, ~450,000 characters are through that section of the story. The actual number is probably quite a bit lower than that as multiple attempts would be made while leveling, especially through the duty roulette.

But again, FFXIV is a slower game, and I know a lot of folks who came in for the expansion and hadn't finished the 2.55 main story yet, so it's quite possible this number is low compared to the general populace of active subscribers. But even if I'm generous and say, only 25% of players have a combat job at over level 54, that's still only an upper bound of 1.8M active subscribers.

Edit: It's possible that the dungeon attempts isn't per character, but per instance spun up, which would basically multiply that 450k number by 4, bringing us to 1.8M characters through that section of the story maximum. That doesn't seem unreasonable. In fact, that kind of makes more sense given the 217k character number for entering Alexander: Gordias. With 1.4M runs of Neverreap plus Fractal Continuum combined for enough Tomestones of Law to hit the 170 ilvl requirement for Alexander, that would be under 7 runs per character if the dungeon runs were per character, which isn't quite enough to hit 170 if memory serves. That would bring the Expert Roulette dungeons to 5.6M or so runs on a character basis, which seems to be a little more reasonable as it'll be more than the Alexander populace which probably does make up more than half of those runs.

Given that, I doubt only 25% of the populace is over 54, so if we bump that to 50% that would mean an upper bound of 3.6M active subscribers.

Putting it Together

My guess for the active subscriber base is somewhere between 0.6M and 1.8M, probably closer to 1M in my opinion. If the actual number was 1M, that would mean about 21.7% of the populace have seen Alexander: Gordias, but still only puts a maximum of 45% of the populace higher than level 54 on one combat job. There's no way their total played time is remotely accurate. Frankly, I really don't think "registered accounts" is a good correlation to active subscriptions.  

Edit: Given my edit above, the guess expands to between 0.6M and 3.6M, but frankly the lower bound is bound to be much higher than 0.6M given the Expert roulette dungeon runs. If the dungeon runs is per instance rather than per character, I'd double my estimate to 2M.

#FFXIV, #Statistics


  1. Pause for a second, the Savage data isn't exactly as you think. See how many people they say entered Savage A4? Zero. However, there are attempts at A4 since a couple of weeks ago.

    That means, it's counting CLEARS rather than just entrances. In other words, all of that data does not show how many people have entered Alexander, just how many have beaten it.

    That leads me to believe that, that also counts for dungeons. It's counting clears rather than just entering.

    Then you have to take into consideration that this data does not show clears of 2.0 and 2.X dungeons/raids since the release of 3.0. Due to the nature of 3.0, this data essentially excludes new players and players who stopped at 2.0 thinking they didn't have to do the patches.

    And, here's the last bit that blows the whole thing to Hell: total play hours does NOT say it doesn't count free trials.

    1. Hmm, your point about Savage A4 is a good one. However, I had inadvertently already been counting it as clears as the data I was comparing it to in WoW was also clears.

      Even if the dungeon run is about clears, and not entrances, that would mean there are even fewer subscriptions than I suggested, not more.

      I also took into account the idea that not everyone is at level 54+ as Sohm Al would require. In my text I explicitly give it a generous 50%-75% of the population NOT being 50+ for the upper bound calculation.

      The total play hours could include free trials. That doesn't preclude the fact that the number is ludicrous. Also, free trials still require an account registration.

    2. Ah, but the registered accounts only account for subscribers. That means, free trials are a huge variable and also Chinese players (that means there could be 10 million registered accounts, 5 million subscribed at one point which would mean only they are counted in SE's "5 million registered accounts). Not active subs, but people who have subbed.

      As for the raid, a lot of people have to gear up to enter even regular Alexander and a lot of people can't beat it. You call it LFR but it's really not. It's between CT and BCoB in terms of difficulty (or at least that's the general feeling). Then you have to take into account many players don't want to do 8 man raids because of BCoB or not interested - between crafters and people who just prefer to run dungeons or hunt. Then you have to take into account that there's a large population that won't try an 8 man raid if there's no echo, and those that refuse to PUG a raid. Then you have to take into account that there are still people having difficulty with the first floor.

      As for Sohm Al, there are a lot of people not at 50+ and many that have hit 50 but are levelling an alt class as well. Let me put it this way, I have a friend who started the game with Heavensward's release. He is levelling everything together. He plays nearly everyday for hours, and yet he has not even gotten past Ifrit Normal. There's a lot of players that are doing old content first.

      My point is simple: there are too many variables. Until there are less, you cannot make a population guess.

    3. "Registered accounts only account for subscribers." - We literally do not know this. That's the entire point of this exercise: SquareEnix is deliberately obfuscating their active subscribers. Also, you cannot use the free trial until you register an account, which would suggest that they would be included in the "registered accounts" number.

      Agreed on all the variables. There's a reason why the gap between my lower and upper ranges is so massive. However, that still doesn't preclude us from taking an educated guess.

    4. No, I'm not talking about ACTIVE subscribers. When they talk about registered accounts, they only talk about players who have subscribed.

      Here's the Nantes event slide that shows this:

      In other words, the amount of free trials is unknown but the playtime amount does NOT say they don't count them. Remember it says playtime for ALL characters - which includes alts as you noted above as well.

    5. Ah, I see, interesting. Thanks for linking that slide, I was unaware of that. That's...confusing how they word things, but oh well. That does clear up things a bit.

      Still, the playtime number is still ludicrous, even if one does take into account free trials. Well, unless gil spam bots are under free trials, in which case I might actually believe that number :P

      We're talking 11 million characters logged in 24/7 for the entire year to hit that number. Or 66 million characters at 4 hours a day for the entire year. Free trials won't make up that number.

    6. You can't forget some other things though - such as AFK also adds up and that usually the game doesn't have auto-kick most of the time. There's also a high possibility that bots ARE counted.

      Then you have to add in frequent content updates which does bring back a lot of players to play constantly. Then there was the massive amount of people from games like WoW looking to try out FFXIV due to expansion and unhappiness with their game.

      Subs also DON'T take Chinese players into account.

    7. Sure. I get all that, I really do. But what you're suggesting here is that FFXIV has more active players over the past year by an entire order of magnitude than WoW did at its prime.

      Sorry, but I don't believe that. If that were true, you would bet that they would announce that directly, because it would be an unprecedented coup.

    8. No, it's the equivalent of WoW adding up all 100 million people who have played the game since its inception. Think of how many play hours that is. Same thing.

    9. FTA:
      "The first metric I want to look at is total playing hours of all characters. Nearly 97B hours of play time since the inception of the game. Thankfully, they give us the total playtime a year prior, which reduces this number to just under 96B hours"

      Which is to say, 96B of that 97B was from this year alone according to SE's own slides.

  2. I don't think your comparison of Alex Normal to LFR is correct. Alex Normal actually has a fairly high item level requirement of 170. That's basically full Law gear. And even of the people who have that, I think a lot of people compare it to Binding Coil, which was basically fixed group content. Just as an example, I haven't done it yet, even though I meet the requirement.

    There is a reasonable case that Alex Normal is the top 10%, especially since the drop-off from 1 to 4 is not that steep (that itself is a bit of an oddity).

    1. Largely agreed, hence why I put my comparison as a lower bound. Frankly I think the actual value is fairly significantly above the lower bound, but without more data, I can't say for sure.