Segmentation @Playtika

Introduction

Since it was founded in 2010, Playtika develops online social casino-type games and it does it so successfully that many of its games, such as Bingo Blitz or Slotomania, are leaders in their field. With millions of monthly users, segmenting users into various groups is imperative to provide them the best possible experience. Adding AI to its analysts’ know-how, Playtika develops optimal personalized experiences.

Segmentation as Part of the Analytics World

The first lecture was given by Stav Zohar (Business Analyst Team Lead @Playtika). He started by defining segmentation: dividing your target market into approachable groups with similar characteristics. This can be set by demographic, geographic, psychographic, behavioral needs, etc.

Benefits of Segmentation

• Know your customer better
• Be cost-efficient
• Increase customer retention
• Create market opportunities
• And more…

Once you create segments, how do you define success? Look at the whole cake: overall growth + how KPIs are improved (it’s not just about increasing revenue but about the overall engagement/experience of players).

How to Create Segmentation

1. Set objectives and expectations
2. Define the method and tools you’ll use
3. Identify segments
4. Evaluate their potential
5. Create a strategy
6. Launch this strategy -> Monitor -> Optimize

For example, you could segment according to age or game mastery. You could end up with a young player who reached a high level in the game. His experience should be different from an older player who’s just starting the game.

In Playtika, games are free but the experience is better when you purchase. RMF segmentation helps identify groups of customers for special treatments thanks to its RMF metrics:

• Recency: Freshness of the customer activity, be it purchases or visits
• Frequency: How often customers make transactions or visits
• Monetary: If the customer has the intention to spend, or the purchasing power of the customer

For example, you could measure the time since the last order, the total number of transactions, the average time between engaged visits, the average transaction value…

If 1 is the highest score and you get the following table, you could form a strategy of contacting recent but infrequent users and create experiences or offers that will encourage them to play more.

Why Segmentation in Gaming

• No player is the same
• Personalization is key for fun – “how did they know I’d like that!”
• Enhances growth

In Bingo Blitz, for example, a newbie sees a screen that is less busy than a more experienced user. Also, there is an onboarding process with tips on what to do.

You can segment according to what you expect the player to do. For new players, you just want them to finish their games.

If players come back, if they haven’t played for more than 30 days, you don’t throw them straight into the game. As you would do offline, you’d first extend a hand: “So glad you came! Let’s play :)”

Take two users:

• Level 16, joined recently, 40 credits, never purchased
• Level 548, comes every 3 days, 900 credits, purchases in $10 and had 40 transactions in lifetime

They have different needs. The more recent one is limited by the limited content he’s come across.

7 Criteria for Segmentation

Word of caution: Don’t forget the WHY! What are you trying to achieve?

1. WHO are you trying to identify? What behavior you’re looking for? Payer/non-payer, FTD (First time deposit), STD (Second time deposit), TTD, High/low value…
2. TIME FRAME – What are you examining? If a user is inactive for 3 years, what should happen when he comes back? When does a player becomes a dormant player, 30 days, 90 days? In Playtika, it depends on the game.
3. METRICS – For example, measure the purchase engagement with LTV (Lifetime Value), ATV (Average transaction value), recency, product preference… Use a few of them.
4. CALCULATION TYPE – If you want to measure the users’ payment comfort zone, how do you calculate it? Do you use different calculations for different segments? How far back do you go to calculate the average?
5. NUMBER OF GROUPS – How many groups should be created? Usually, you’re limited by operations. But also, you should evaluate what is the value of adding a 4th, 5th, … additional sub-group. You need to identify how much difference one finds in players within the group and between groups.
6. WEIGHT – E.g., the behavior of last week counts more than that 2 weeks ago
7. MIN/MAX – What to do with outliers? Let’s say someone plays more than 10 hr/day, what would be the point of further segmentation? What do you want them to experience? But also, what more can you bring them? Hard to make a more relevant offer, so would it make sense to cap top player spending?

Let’s take an example, a Black Friday offer. If you know what a customer likes to pay (e.g. 10 shekels), you can push a bit so they pay an amount outside of their comfort zone (e.g. 13 shekels). They pay more, but then they get 50% more, it feels like a “once in a lifetime” offer.

If you push too much users out of their comfort zone, it can backfire. Also, you’ll get users who see the offers that their siblings get, and they could get upset if despite their loyalty, they don’t have access to good deals.

Do tests. Develop a strategy as you go. Make sure your segmentations are similar/coherent and that you can explain to a player why you made that segmentation.

New Era of Analytics Powered by AI

The second lecture was given by Mendi Gold (AI Product Team Leader @Playtika). Games are packed with content. Analysis is a must. You want to control and gain a profound understanding of game success. In fact, at Playtika, you don’t move anything without the supervision of an analyst.

Within games, there are lots of behavior sub-segments. The best way to cater for this diversity is personalization. It could be about how fast a progress bar fills up or finding the sweet spot between an experience that is too easy (boring) or too challenging (give up).

Solution: create groups with minimum variance within the group and with still a balance between segment sizes. Once created, check that the group is homogenous. This is a very iterative process. It takes a bit of time and several iterations to nail down.

In the past, this process was 100% human. Nowadays, after defining the number of groups and min/max, you can run an AI model to minimize variance and balance segment size. For example, run K-means - a classic machine learning algorithm – to find clusters. This is better than what an analyst can do as analysts can deal with 1 or 2 parameters only, but AI can handle lots of parameters. Note, too many parameters = gibberish. About 5 parameters works fine.

Still, you can add business logic. For example, you can force separation between levels 0-60 and 60-120.

Also, once you have your groups, you need to make sure the segmentation is used within the product.

If a player changed behavior, the system must respond, for example, as you go from non-payer to payer.

Note that after a few months, the model might not work well anymore as the truth of then is not linked to the truth of today. When the user is moves to another segment as you update the model or the segments are automatically readjusted, you want to make sure there is no drastic change felt by the users. That’s different from if a user puts lots of effort to move from level 0 to 5 in 2 days. In that case, the user knows the effort he put in and he’ll be ok to experience change.

In Playtika, there are thousands of segments. But they are also central segments for the pillar features of the game. When a new segment is identified, there’s a question of whether/how it spreads over other segments and how to avoid conflict. Also, there’s no magic: garbage in, garbage out!

Coming back to AI:

• It’s a great time-saver! Some segmentation processes used to take 2 weeks and now take 2 hours.
• It helps quantify vague terms making them actionable
• It allows you to discover hidden personas – such as for Bingo Blitz. Looking at low-engaged users, checking their monthly login and game days, one sees a new group of users who like to login frequently but who don’t play much. These were labeled “hoarders” as their behavior allows them to accumulate the rewards offered when login.

By creating a better experience for these hoarders, the size of this low-engaged group could be reduced.

Q&A Panel

For the last part of the evening, the two speakers were joined by Paulina Arstein (Director of Analytics & Game Economy @Playtika), Gadi Ganon (VP Product Technologies @Playtika) and Amnon Calev (General Manager @Playtika) for a Q&A panel.

What’s the main pain of segmentation?

• Managing the player community – why did my grandma see something different than me? As mentioned before, if you can’t explain the logic behind different segment treatments to your users, don’t do it.
• Managing the segmentation knowledge as analysts or dev move on

How does segmentation prove itself?

The goal is personalization, not segmentation. You can create models that adapt and predict, but you must have people who decide how many segments to use, who check that the segments are ok, who make sure the influence on reality works as intended, which is a dynamic process.

Be very precise about what you want to achieve and be careful about what you want to improve. AI is great if the problem is precisely defined. Otherwise, you can’t do anything with your segments, it’s not effective.

Will AI replace humans?

It doesn’t… But it helps analysts start work at a point that is very different from before.

But don’t be afraid of personalization, taking the long-term view, where the system knows more and more what is good for the user. Still, check your system once in a while, and look for ways to improve. How to encourage a user to go from 8 spins to 10? From 30 to 40?

What’s coming up in 2023?

The market will be much more challenging as users are faced with inflation…

Data can help to reach new users, to go faster and be more efficient than before.

You can’t solve global problems with game segmentation… 2021 was a year of great growth due to Corona, companies grow reserves that they can now use as today it’s harder!

With AI, images become commodity.

Use less analysts but more data analysts. Methods become much more involved. You’ll need to know how data science works if you want to be relevant in 2 years’ time.

But then, there’ll probably be tools that make the experience much more straightforward such as tableau for business analytics :)

Thank you very much Playtika for this meetup as insightful as your previous meetup on gamification! This time I took a picture at the entrance of the conference room in front of this awesome mural which has an inspiring quote: “Infinite ways to play!”

?? If you liked these notes, come and join my recently created LinkedIn group?prodUXnotes?to read and share insights in product management & UX notes!

#meetupnotes #aiingaming #segmentation #playtika