Industry background
Many sport betting companies are investing in multiple traffic sources such as: YouTube, Telegram, Instagram, Twitch, TikTok, Twitter (X), Kick, and faced complexities with deal evaluation (audit) and difficulties with profit prediction for concrete deals.
Business value
Our work has short-term and mid-term business impact because it brings the following business values:
-
1
AI model that allows to accurately predict a deal profitability
Model learned from the historical data сut low-value collaborations and prioritize high-value ones.
-
2
Scalable infrastructure
The infrastructure is already connected to APIs of services that provide statistics for YouTube, Telegram, Instagram, Twitch, TikTok, Twitter (X), and Kick profiles.
Since the data structures differ across platforms, we developed a unification layer that can be reused for connecting to other marketing sources.
The model outputs are also integrated into two CRM systems via API.
Process of how project was done
- Data analysis:
- analyzed technical requirements.
- analyzed data and data sources.
- prepared project estimate, roadmap and timeline.
- Data engineering:
- write authorization pipelines.
- parse, process, store data to PostgreSQL.
- join data between CRM and marketing platforms.
- develop database architecture for extracting patterns from data.
- schedule cron jobs.
- API development (endpoints)
- develop automated tests for ETL processes.
- set up Telegram alerts.
- Data science:
- configure data science environment at GCP and local
- prepare data.
- label data.
- learn models and test multiple approaches for comparing.
- evaluate model accuracy (true positive, false positive, true negative, false negative).
- pack ai model for production.
AI model mathematics
-
1
Data itself has patterns inside and these patterns were extracted by using powerful AI-driven math techniques. AI model utilize mean, median, dispersion and variance metrics splitted into natural business groups. Also we tested intervals approaches along with XGboost, random forest, applied a few neural models such as perceptron and multi-layered neural net.
AI model performance
We developed a solution that functions as an internal product fully integrated into 2 client’s CRM systems.
-
+ 0 %
correct results
3081 deals we audited by ai model already. This simplify the work and save time for personal who made decisions.
