Mar 10, 2021
No image
Completed
Building performance-oriented mobile DSP with innovative user behavior prediction mechanism
$100,000+
4-6 months
United Kingdom, London
6-9
Service categories
Service Lines
Cloud Consulting
Software Development
IT Services
Domain focus
Advertising & Marketing
Business Services
Technology
Subcategories
IT Services
Business Analysis
Challenge
Dataseat was looking to quickly get a share of the performance ad market. The idea was to build a DSP (media buying system) to enable ad targeting at users who have demonstrated interest in specific game genres. Dataseat had 3rd party data on user behavior. They didn’t want to use white-label DSPs, due to the lack of features, flexibility, impact on capitalization, aggressively growing prices, and IP issues. From-scratch development wasn’t a good fit either because of time limitations. Another issue was that existing AI libraries were poorly integrable into AdTech solutions due to their real-time nature.
Dataseat was looking to quickly get a share of the performance ad market. The idea was to build a DSP (media buying system) to enable ad targeting at users who have demonstrated interest in specific game genres. Dataseat had 3rd party data on user behavior. They didn’t want to use white-label DSPs, due to the lack of features, flexibility, impact on capitalization, aggressively growing prices, and IP issues. From-scratch development wasn’t a good fit either because of time limitations. Another issue was that existing AI libraries were poorly integrable into AdTech solutions due to their real-time nature.
Solution
There wasn't much time for team integration into the domain, so Xenoss needed a team that could start working immediately. We leveraged our network inside the AdTech engineering community to build a team of experienced engineers from top AdTech firms. Xenoss designed a development plan that would entail simultaneous feature development. It allowed us to ship the early versions extremely fast. To speed up development even more, Xenoss used a library of ready-to-use blocks called the Xenoss Framework. Xenoss compiled the AI models into directly executable code using a sophisticated approach and adapted the AI toolset to be used in the high-performance environment.
There wasn't much time for team integration into the domain, so Xenoss needed a team that could start working immediately. We leveraged our network inside the AdTech engineering community to build a team of experienced engineers from top AdTech firms. Xenoss designed a development plan that would entail simultaneous feature development. It allowed us to ship the early versions extremely fast. To speed up development even more, Xenoss used a library of ready-to-use blocks called the Xenoss Framework. Xenoss compiled the AI models into directly executable code using a sophisticated approach and adapted the AI toolset to be used in the high-performance environment.
Results
The MVP version was delivered 14 weeks after the project kickoff. It’s one of the best results on the market for complex DSP projects. Optimally designed solution architecture ensured record low expenses per one QPS. The monthly cost for the whole system is below $20k. Eight c5.2x large servers handle 400k queries per second. Due to the smart adaptation of the existing AI toolset, the real-time prediction mechanism was included in the MVP, ensuring an above-market conversion rate. In just 3 months after the start of the development, the 4 major SSPs were successfully integrated.
The MVP version was delivered 14 weeks after the project kickoff. It’s one of the best results on the market for complex DSP projects. Optimally designed solution architecture ensured record low expenses per one QPS. The monthly cost for the whole system is below $20k. Eight c5.2x large servers handle 400k queries per second. Due to the smart adaptation of the existing AI toolset, the real-time prediction mechanism was included in the MVP, ensuring an above-market conversion rate. In just 3 months after the start of the development, the 4 major SSPs were successfully integrated.