Why CPI is not that important for hyper-casual, or what metrics we’re looking at

Low CPI and high retention is always good, but even imperfect metrics don’t mean the project should be immediately shut down. We believe that it’s necessary to look at the combination of factors and there are several ways out.

We have many cases when other publishers turned games down, but they became hits after several iterations and brought in hundreds of thousands of dollars monthly. It’s not just about marketing, product metrics play a big part as well. That’s why you need to check all the theories first and only then decide what to do with the project.

Let’s start from the beginning.

Is CPI that important?

I can’t say that we don’t look at CPI at all. After all, this is one of the first metrics that helps determine the direction for the project’s development.

For example, if the CPI is low (currently, anything less than 20 cents for the US market is considered low), we try to start massively attracting users almost immediately. One of our benchmark games with low CPI is State.io, we’ve already talked about its development here.

CPI above 60 cents is usually considered high. The bottom line, however, is that if you attract a solvent audience, even $2.5-$3 is normal. That’s what happened to Chain Cube, which was abandoned by another publisher. The project had good retention, but low playtime, a rating below two stars, and a high CPI. We’ve had several iterations and as a result, the game began to bring in $500,000 in the first months of scaling and is still profitable. You can read more about it here.

Sometimes we see a $5 CPI and it still pays off. This happens when most of the audience is “burned out”, but paying users still come in. Even Chain Cube and State.io have high CPI from time to time, and it’s still possible to work with it.

In any case, CPI is not what we look at first.

What do we look at first

We look at retention. If it’s good, we’ll still try to acquire at any CPI.

Let’s say, Day 1 retention (R1) is 45%+, R3 is 25%+, and R7 is 13%+. In this case, CPI isn’t important at all. With a high degree of probability, we’ll be able to attract players and reach the break-even point.

There are several general kinds of retention:

1. Top games: R1 45%, R3 25%, R7 13%, R30 4%.

One of the examples is WormsZone.io. The LTV was low initially, but the game paid off due to the super high and long retention. We worked on LTV, added ad placements and the situation became even better. By the way, the game currently has 300 million downloads and it still brings good money.

2. Good: R1 40%, R3 15%, R7 9%, R30 2%.

Like Bottle Jump and Harvest. At first, they didn’t even break even (there was a rather high CPI), but we did work on the level funnel and ad placements. We tracked levels with a low win rate, fixed them, and built a funnel of levels with increasing difficulty. As a result, users began to spend more time in the game and LTV increased. At the same time, we worked hard on ad creatives, ASO and playable ads, which greatly increased the number of downloads.

3. Borderline: R1 35%, R3 12%, R7 7%, R30 1%.

Do Not Fall used to be like that. Although, working on the funnel of levels and balance also allowed us to fix the situation.

4. Below average.

If R1 is, say, 30%, then you need to look at how fast R3 and R7 fall. If they don’t fall very quickly, it makes sense to polish some things and try to acquire users on other ad networks. For example, in the game Bike Hop, the retention was rather low in the first days, but there’s no steep dropdown. The game pays off due to a fairly low CPI (up to 35 cents), which has been at about the same level for two years now.

If the game doesn’t have enough content to evaluate retention beyond the first day, then we look at the funnel of levels, at how many people have reached the end and seen all the content.

Let’s say there’s a game with ten levels, and 60-70% of players reach the end and abruptly start to drop off. The next step is clear: ask the developer to add another 10-20 levels and look at the way it changes the retention.

We’re also looking at other events like opening new skins. If we see that a large number of people opened all the skins, it’s safe to assume they finished the whole game and have nothing else to do. This is solvable by simply adding more content and affects R3 and beyond.

How to fix retention

Let’s imagine a situation: good CPI, but low retention. In this case, we should look at what could be the problem and try to fix it with several iterations.

Possible reasons for low retention:

• Critical bugs. Maybe the game crashes or doesn’t even start.

• Setting. Hyper-casual games are aimed at a wide audience, and users won’t accept skeletons in a cemetery as readily as they would stick figures. Therefore, we often test different visuals on the same project and the same traffic in order to choose the best one in terms of metrics. And we see how the same game can give radically different metrics with different visuals.

• Problems with win rate. The levels might be too difficult for a hyper-casual game. For example, a 50% win rate is often too low for this kind of audience. Making the first levels 90-100% passable is what we consider normal. In-house analytics tools are great for finding the tightest bottlenecks.

• You’re out of content. I already mentioned this, but I’ll leave it for the list.

• Players don’t get core mechanics. It’s difficult to do something here, but you can try to polish the gameplay, or look for other mechanics. You can come to a conclusion on this matter only by doing tests — compare the mechanics, fine tune the current one and observe.
  
• The mechanics are good, but the replayability is lacking. First of all, you need a mechanic that people will want to play for a long time. An example: in WormsZone.io, R30 was a high 20% at the start despite the fact that there wasn’t a lot of content. It happened because the game had an interesting gameplay people didn’t get bored of.

• We look at another important factor — the average time to complete the level. Sometimes people leave not because they’ve lost, but because they have run out of free time. They play for a minute and a half and leave, and it takes only 10-20 seconds to complete one level in a hyper-casual game on average.

• Also, recently I noticed that the polishedness of the prototype and the overall look & feel of the game greatly affect the overall retention. This includes sound and special effects, even vibration — that is, everything that creates an overall impression. Currently, this has a strong impact on all metrics, as the players have become more demanding and the market got more and more oversaturated.

And if we had a good CPI, but low retention, we would most likely start making another game in the same genre at the same time and test out new theories. We talked about such a case in this article: we released five Hidden Object games with one studio. As a result, the profit from these games became comparable to the release of a super hit. Our position is simple — it makes no sense to completely redo the game that already attracted a certain audience and generates income. It’s easier to make a new project and test a new hypothesis on it.

What if R1 is low and R3 and R7 are high? And vice versa.

In the first case, this means that it’s very difficult to find an audience for such a project. That is, it gains a large audience on the first day that quickly leaves the same day, but then it leaves much less. This means that it’s difficult to isolate the desired audience, you have to narrow it down all the time, which means that CPI will be expensive. This is typical of the puzzle genre, but can still pay off.

There are many more reverse situations when R1 is high and then it drops sharply. This means that the game is most likely underdeveloped in terms of depth and content. In other words, there’s a core mechanic accompanied by some content, and then the player has nothing to do and leaves in a couple of days. This is quite typical for hyper-casual games, especially now — there’s a lot of competition and there are many similar games. What you need to do is fix it quickly, before the audience burns out.

What to do if all metrics are average

Look at the game’s rate of return. With average CPI and retention, it can go both ways.

If there’s none, we’ll try to fix the things that are obvious to us with the developers — increase monetization, polish the retention. We’ll also try to reduce the CPI a little with ad creatives and ASO.

Here we talked about cases like that in ASO. In some cases, simply changing the icon can increase the number of downloads by 30%.

If these measures help, then the game is very likely to start making money, even if it wasn’t paying off at the beginning.

The same scenario with the projects that barely break even — this is actually a common situation when you start to launch traffic, it happens to about 1 out of 3 games. This too can often be corrected with iterations, since we’ve already launched more than a hundred projects and know what works and what doesn’t. Also, we constantly test hypotheses in practice — no project is getting simply abandoned.

Another thing is if you have a project that has been iterating for half a year and is just starting to pay off — in our experience, it will most likely fail. The audience will start to burn out (and retention will follow), CPI will grow and payback will still start to drop down. In this case, it’s better to focus on more promising theories.

Other metrics and nuances

In addition to the funnel of levels and the amount of content, we look at the playtime. For hyper-casual games, good playtime is 10+ minutes on day 1 and day 0. If the playtime is lower, then either the game isn’t very good, or it has real problems, and then we give it to QA to check the errors, operability of all systems, even for playability itself. Maybe the game just doesn’t start for half of the people.

Nonetheless, we have cases when the playtime increases over time. This mainly happens because of constant tests and project improvements. During hypothesis testing, all changes are rolled out only to a limited audience and if successful, the update goes to all users. Thus, 100% of our updates benefit the project, the rest simply don’t reach a full-scale release.

The easiest way to increase playtime is to add more content. At launch, State.io only had one map — a map of the United States. As soon as we started adding maps of South America, Europe, Asia, Africa, all metrics immediately increased, including playtime.

There’s also CPM — this metric characterizes the quality of the audience. For example, white-collar workers in New York City will have a CPM 100 times higher than young people in India. CPM in the US is $100, in India it’s $1-$2. In other words, CPM shows how much advertisers are willing to pay for 1000 impressions for this audience, therefore, a paying audience will have a higher CPM.

Quite often, you can evaluate the quality of the incoming audience by running tests. For example, Facebook shows a certain visual mainly to women who spend money on marketplaces. It shows another, more “childish” visual to a young audience. According to this principle, you can indirectly assess whether you hit the audience and what kind of acquisition it’s going to be. With an expensive CPM, an expensive CPI is normal, but if a non-paying audience has an expensive CPI, it’s bad news.

At the same time, different networks have different CPI. And speaking of tests, we run them on Facebook and TikTok. But in reality, the main part of the acquisition in hyper-casual is ad networks. The situation may be the opposite there. If an audience from TikTok came to you at an expensive CPI, this doesn’t mean that it’ll be the same on ad networks.

On the other hand, retention on ad networks most often correlates very strongly with Facebook and TikTok, so we can most likely trust it.

When to abandon a project

If we’ve reached the point when the project starts large-scale acquisition on ad networks, the main metric here is the rate of return. It’s easy to calculate: divide LTV by CPI. LTV, in turn, is the ratio of revenue to the number of active users. The main goal for the rate of return is to be greater than zero.

We shut projects down only if we don’t have any hypotheses to test. If there are any bottlenecks or other problems, we first try to solve them.

I already talked about Chain Cube, which got dropped by another publisher. The situation with Sword Play was the same. We spoke about the publisher abandoning the game after several iterations. We rolled them back and made the project top 1 on iOS in the USA.