欠佳：0.05 K factor
在Casual Connect大会欧洲专场上，Wooga的Stephanie Kaiser发表了一个很棒的演讲。尽管演讲的名称上说的是“病毒性”，但演讲中大部分讨论的都是用户参与度问题。玩家间的消息确实是玩家在替你做推广，但事实上他们是在现有玩家间推广参与度，并提升日活跃用户。而预测模型中所使用的病毒性是将游戏推向那些并未玩游戏的玩家。
我在这篇文章中插入了某些预测电子数据表的截图。蓝色的单元格代表数据点，其数据值已经在之前的文章中表述过了。电子数据表的用户可以编辑的是黄色的单元格。需要特别指明的是，“USE THIS”列中的单元格所涉变量数值与电子数据表中的复杂运算有很大的关联。所以，你可以根据蓝色单元格中的变量范围选择某个数值输入到“USE THIS”黄色单元格中，或输入你自己的自定义变量数值。这个数值可以是蓝色单元格数值范围内，也可以位于该范围数值之外，因为有些人可能想看下年净盈利或用户曲线图。
在上篇文章中，我决定使用ARPU * DAU %的方法来计算玩家付费。这种类型的变量肯定需要在你自己进行预测时监测和计算。你可以将用户以周为单位来划分，随着玩家在游戏中耗费时间的增多，你可以不断追踪玩家付费行为的趋势。随着游戏逐渐成熟，你也会看到这些变量会发生变化。新玩家的付费或许会比那些在4至6月游戏成熟后付费的玩家要多。而且，你也必须注意游戏成熟阶段的新玩家，因为流失率有可能会高得多。当选择继续玩游戏的新玩家数减少时，你应当随之下调ARPU期望值。在这些图标的元素中，我并没有让模型自动进行调整，因为这部分调整很复杂，而且也是模型需要进行改良的地方。
Financial Forecasting for Social Games
Last week, I began digging into the key performance metrics for financial forecasting of social games. This week, I’ll finish the rest of the variables defined in week 1, along with additional supporting links where appropriate.
If you haven’t read Part 1 of this series, be sure to check out the section on Player Populations for the relationship between New Players, One Week Players, and Retained Players and how Attrition Rates connect them.
Mature Games vs. New Games
If you haven’t read Part 2 of this series, be sure to check out the section on Mature Games vs. New Games for how user behavior and metrics performance can shift over time depending on how far along you are in your game’s lifecycle.
As everyone knows, virality on Facebook has changed a lot over the past year. No longer can you depend on virality alone to spur growth. It’s still a very important driver, though, as it not only acts as a discount on your Cost of Customer Acquisition, but also creates opportunity for truly explosive population growth, without needing a huge marketing budget. So if you can generate viral growth, more power to ya, but don’t depend on it completely for success.
Kontagent’s GDC presentation quotes a K factor of only 0.05 to 0.20. Wow, that’s really far away from the magical 1.0!
Player Populations: An argument could be made that Virality is different for New Players, One Week Players and Retained Players.
For New Players, if they don’t even stick with the game after Day 1, it seems highly unlikely that they will send out any invites. Since the granularity of my model is at a Weekly level, however, I still assign some virality to the New Player pool, but at a lower level (since this metric is being applied to the larger pre-attrition total). It’s not perfect, but it’s better than applying zero Virality to this population.
The One Week Players represent the population of players who have decided to keep playing and are most likely to send out invites to build their initial base of “neighbors” for in-game interactions. This two week window (covering both populations) is a critical metric to watch, especially at launch.
For Retained Players, they may have done most of the inviting their likely to do when they initially started playing, not several months into playing. Therefore, their virality is going to be less than the One Week population.
Poor: 0.05 k factor
note: adjust this for New, One Week, and Retained Players, and for Game Maturity
There’s lots of data out there on Daily Active Users and Monthly Active Users, because of App Data (http://www.appdata.com/). I encourage you to go there and poke around on lots of different games.
While there are certainly some expected ranges if you look at all of the biggest games, it’s important to note that niche titles may have much higher DAU % but on a much smaller base of MAU. So it really depends on the kind of audience you’re attracting and the behavior that your game encourages.
At Casual Connect Europe, Stephanie Kaiser from Wooga gave a great presentation. While the title says it’s about “virality”, most of the talk is really about engagement. Yes, player-to-player messaging is ‘viral’ in the sense that the players are doing the promotional effort for you, but they’re promoting engagement between existing players and encouraging higher Daily Active Usage (see the 38:35 minute mark)– contrast this with promoting to people not currently playing, which would be the virality that the forecasting model uses.
Player Populations: Just as with Virality, Daily Active Usage is likely different for New Players, One Week Players and Retained Players. New Players may have much higher visitation as they are initially learning the game, but as they master the game, behavior can change. Many games that rely on ‘scheduling’ can shift focus into longer periods between visits as the in-game rewards (coins, XP) for shorter turn-around tasks becomes less meaningful for game progress.
Poor: 10% DAU/MAU
note: adjust this for New, One Week, and Retained Players, and for Game Maturity
As you might guess, accurate data on player spending is fairly sparse out there on the interwebs or in conference presentations, at least anything recent. Game companies wisely keep this kind of data to themselves, or obfuscate it enough so folks can’t pin down anything too specific about what’s going on behind the curtain.
In Kontagent’s GDC presentation, they say that typical Average Revenue Per User (ARPU) is just $0.01 to $0.05 per day. Another way to measure it is looking at the percentage of paying users (1% to 3%) and then the Average Revenue Per Paying User (ARPPU), which they say is typically $2 to $10+.
Whales: That “+” at the end of Kontagent’s ARPPU estimate is worth emphasizing. A small portion of paying users spend a LOT on these games. The beauty of freemium games is that there’s no upper limit on spending, and Whales are what happens when you open up that possibility. We’re talking about over $1,000 a month as a possibility. Clearly this kind of anomalous behavior can play havoc with statistics — when your Average revenue and your Median revenue are so far apart, some of these ARPU and ARPPU numbers can be misleading as to what’s really going on behind the scenes.
Balancing ARPU and DAU: As a generalization, the more niche your title, the smaller your potential population but the better your ARPU is likely to be. In contrast, a mass-appeal title may have a huge player population, but a smaller % of them are likely to pay, which is going to drag down on your ARPU metric.
ARPU vs. ARPPU: you need to pick one of these two approaches, either a) calculate your DAU and apply an ARPU to that pool of players, or b) calculate a % of players who are paying, apply that to your DAU, and then assign an ARPPU to that pool of players. Either one is workable, but for my model I go with the ARPU * DAU approach.
New Players: as I commented earlier about Virality, it’s tricky assigning purchasing behavior to the New Player population when many of them won’t stick around even past Day 1 (which will get worse as the game matures). Some players, however, are going to jump in with both feet and start spending money, so it’s likewise not fair to assign zero to this population. Just be cautious, and also adjust your expectations downward for when the game is mature, since the higher attrition rate should be reflected in smaller ARPU.
Poor: $0.01 ARPU
note: adjust this for New, One Week, and Retained Players, and for Game Maturity.
Ongoing Operations Costs
This is a hard category to apply any standardized variables to, because it depends on your own team, the scope of the game, and how much formality you’re looking to put in place. There’s a big difference between a 5 person garage team and all of the infrastructure that Zynga has in place for a new title.
The two things you need to figure out are Fixed costs and Variable costs.
For Fixed costs, these are mostly going to be staffing allocated to the project, and perhaps some contractual relationships (like paying Kontagent for their services!). And if you want to get fancy, you can calculate whatever portion of company overhead (HR, executives, lights and rent) that your project is saddled with.
For Variable costs, these are going to scale based on some other metric of the game. For instance, associating your IT costs to the population of the game will make sure that you’re taking into account more servers for more players.
You could also make content creation a variable cost, associated with overall revenue of the game. This would reflect an increased allocation of staffing to a successful game, trying to keep that hoop spinning down the street as fast as possible — while for a less successful game, you may cut down on content staff to recognize the lower ROI of those efforts. It really depends on your approach to resource allocation and where you set the bar on ‘success’.(Source: PlotLuck Games)
Financial Forecasting for Social Games
Over the past several weeks, I’ve been developing a systematic approach to forecasting key financial metrics for social games. After outlining all of the variables, then working through the plausible ranges for these variables, I’ll now walk through the forecasting tool itself and how to use it before launch and after launch.
How to read the Spreadsheet Examples
I’ve inserted some screenshots of my forecasting spreadsheet in this blog post. The Blue cells represent data points. Their values have already been discussed in previous posts. The yellow cells are meant to be edited by you, the user of the spreadsheet. In particular, the “USE THIS” column are all of the variables that are actually grabbed by the parts of the spreadsheet that are doing all of the heavy lifting calculations. So the idea is to look over the range of variables in the blue cells, then either pick one value to input into the “USE THIS” yellow cell, or input your own custom variable, perhaps somewhere in between the blue variables or even something way outside the range if you just want to ‘see what happens’ to Annual Net Revenue or the Population curves.
Paid Customer Acquisition
Over there to the right is what the forecasting model looks like for Paid Customer Acquisition. The columns for Poor, Middling and Good, reflect the values discussed in Week 2′s blog post.
The section on Customers Purchased per Month has several yellow cells in the Low, Medium, High areas where you should provide your own budgetary ranges. There’s really no standardized ranges here, as it depends on what your budgetary capabilities are. Note: while the model calculates at a weekly level, I organized this by Monthly, since that’s likely how you’ve allocated your marketing budget.
The rows for Number of Customers take these budget values and apply them to the CPA values currently defined in the “USE THIS” column for that phase of game maturity.
“Free” Customer Acquisition
This section of the forecast covers your access to “Free” customers. As discussed in Week 2, if you’ve got multiple games in your portfolio, you can cross-promote to your existing audiences for free. If not, you may want to work with a service like Applifer , in order to get additional exposure without buying Facebook ads.
The numbers in the yellow cells are taken from the estimated impact from Applifer, but you should put in your own numbers here if you’ve got special access to customers on your own.
Attrition of Player Populations
This section of the forecast covers attrition rates of the different Player Populations. As discussed in Week 1, Attrition behavior will be very different soon after launch compared to several months into the maturity of your player base.
Note: I don’t apply a separate attrition rate to the One Week population. There’s one rate that takes New Players and transforms them into One Week Players, and then I start applying the same Retained Players attrition going forward. I could create more granularity here, but I don’t think it’s necessary. In earlier versions of the model, I had much more elaborate population curves, spread out over several weeks, but based on other research, I think populations settle into a Retained attrition behavior rather quickly.
Viral Customer Acquisition
This section of the model covers virality behavior of the different population pools. Notice that in the blue data cells, I’m using the final K factor and some assumptions on click-through rates, in order to back into the number of Invites. In the end, the K factor is all that’s used in the model (the yellow USE THIS cells), but you will want to track Invites and Conversion Rates so that you can improve those metrics in your game.
As discussed in Week 3, I decided to go with the ARPU * DAU % method of calculating spending. This category of variables is definitely one to track and update immediately in your forecast. By breaking your populations into individual weekly cohorts, you can start tracking trends in spending behavior as players spend more time with your game. You should also look for shifts in these variables as the game matures. New Player spending at launch may be more aggressive than New Spending four to six months into the game maturity. Also be careful with the New Player population at Maturity, as the Attrition rate is likely to be much higher, so you should downshift your ARPU expectations later on, when fewer of your New Players are likely to stick with the game. I don’t automate that adjustment here in these dashboard elements, because it was getting complicated enough as it was, but it’s certainly another level of refinement that I may look into.
Once you’ve established which variable values are appropriate for your game, you can chart out sensitivity analyses for each variable, to see where the tipping point is between failure, ho-hum success, and rocket-to-the-moon success.
CPA at Launch:
Here’s an example of the sensitivity curve for the price of acquiring customers at launch. As you can see, there’s a point right around $0.25 that the Annual Net Revenue really starts to climb. This reflects a lot more customers being acquired at launch for the same defined budget.
Clearly, a lesson here is if you have a successful game at launch where other key metrics (retention, spending, etc.) are performing well, you want to take advantage of that window of opportunity by acquiring as many cheap customers as possible, until the ‘low hanging fruit’ isn’t so low anymore.
Attrition of Retained Players
This graph shows the effect of varying levels of attrition of the Retained Players population on Annual Net Revenue. As with the CPA chart, this is not a linear relationship. The tipping point looks like it’s happening around -10% attrition. Better attrition than that, and you have a very stable pool of retained players to monetize.
While Day 1 and Week 1 attrition are important early indicators of future success, the stability of your retained players tells you how leaky the bucket is long term.
Managing Forecasting over Time
Forecasting isn’t just for pre-launch planning. Once you’ve got real data, start plugging those into the spreadsheet into the specific weeks where that data applies, overwriting the formulas that had been calculating the forecast data. Now, your Past Weeks are feeding off of actual data, while your Future Weeks are still using your forecasting variables — and now your end of year goals, like total Net Revenue or end of year Player Population will be better informed and more accurate.
And, of course, take this opportunity to change your forecasting variables so that they are using the numbers that reflect where your game currently falls in the spectrum of Poor, Middling, and Good. It will be hard to be Good in all categories, but the forecast should reflect reality, not unrealistic optimism. Then you’ll be laser focussed on the real areas that need improvement. (Source: PlotLuck Games)