Progamero - making games more satisfying
There is a problem of the automated user behavior analysis aiming to optimize the game or application. What do we mean? Users can take actions in the system, be satisfied with the result, get upset, come back to the system after the long absence, or, vice versa, enter the system for the umpteenth time that very day. The sequence of these events contain the invaluable information about the person and the system’s cooperation. It might give a cue whether a person is not satisfied with something and there is some need to support him, it may hint that there are some bugs or whether the person is prone to various decisions. There are numerous examples for each application.
Modern BI by definition very subtly serves the aim of user behavior description. It can only display the main KPI or known patterns. This is one side of the coin. On the flip side, you might have millions of users, each of whom behaves differently. Naturally, you fail to pay each of your analysts essential attention to each user.
Working on different projects in the gaming area, we faced this problem especially critically. Users take thousands of actions, each action has many parameters and applying standard methods how to optimize based on this data is very difficult, almost impossible.
Do you use this information to the full? If the answer is positive, most probably, you don’t need us. However, the answer is more likely to be negative. The issue is not about the lack of BI and analysts working with it. It's more delicate, it's about the nature of this information.
We suggest two parts essential for the transition of the manual adjustment of the application behavior for all users to the automated and personalized one. Using the technical language, reaching this aim we need two parts:
- Complex mathematical models of the user behavior, capable to predict his further actions and reaction to your actions and suggestions. Currently, this process implies recurrent neural networks.
- The infrastructure capable to support the neural network for every user with all concurrent bells and whistles essential for the serious production.
We offer the platform for modeling a user behavior in real time. Besides, we have professional services with the experience in the similar tasks who will be happy to develop the models adjusted to your application or game. In any case, whether we help you develop the models or you create them by yourselves - it’s your IP we don’t have any claims for.
A: Our system will send you the recommendations what to do with a specific user this very moment. For example, it’s worth awarding this very person with a bonus, otherwise he will leave.Q: How complicated is the integration?
A: The integration demands your system to send us the application’s events and accept out recommendations. It takes around one week.Q: What does the onboarding process look like?Каков процесс onboarding
A: In case you build/create models, onboarding demands the minimum adaptation of your models, after that the process becomes completely self-serviced. If we help you create the models, the preliminary prediction is about three months for that. This is the period to understand the data, build and perfect the models.Q: What is the expected revenue?Какова ожидаемая прибыль
A: The answer directly depends on the application and the stage of its optimization. 10-20% of improvement can be treated as positive evaluation.Q: Which KPI may be improved using your technology and methodology
A: 1st Day retention, 7th Day retention, LTV.Q: What’s the risk of aggravating the situation and losing my users?
A: The risk is minimal - our platform provides the opportunities of A/B testing and offline simulation which are bound to minimize risk.Q: How do I benefit from personalization?
A: You will definitely benefit selecting the system’s actions more precisely for each userQ: What’s the benefit of real-time?
A: There are some cases when the window of opportunity is pretty small. For example, a person couldn’t pass some game level three times in a row. If you don’t support him right now, you will definitely lose him as a customer tomorrow.Q: In which areas is your solution optimized?
A: Our solution is largely adjusted to the games, but every case with the difficult user behavior and the need of the prompt decision-making is taken into consideration. For example, we might mention security.
Q: Why will one need neural networks?
A: Standard methods of machine learning require manual extracting of the main influence factors from the data before actually applying the learning algorithms. This feature engineering is very complicated in case with the events flow.
Q: What is the your value proposition about?В чем ваше value proposition.
A: We help you improve the main KPI by means of personalized behavior. Our technology saves you from paying the army of DevOps. Besides, our professional service make up for the difficulties of hiring deep learning professionals capable of solving relevant tasks
Q: How difficult is it to build the whole infrastructure by myself?
A: It is possible having a dedicated team of DevOps and Big Data engineers. We evaluate the minimal platform as ten total lifetime. However, we have invested much more.
Q: Which insights will I get while operating?
A: Usually our simulators help notice the system mistakes, users with the awkward behavior and many other details which the average BI might easily miss.
Q: What is my application falls under GDPR?
A: Our attitude to GDPR is very serious. We view ourselves as a Processor (as stated in the point ###GDPR) and implement all required capabilities such as data pseudonamisation and forget me.
Q: What infrastructure do you support? Are you attached to any specific cloud?
A: Our platform is completely built on Kubernetis and we can work either in any of the popular clouds or on bare metal.
Q: What is your business model?
A: There are some intricacies (such as the number of events per user and their size, complexity of the model and exploitation location). As a ballpark, it is possible to charge $0.05 per user a month.