- Project: Data science algorithm development to help an ad network increase revenue for customers
- Duration: 3 months
- Technologies: Python, Pandas, NumPy, Scikit-learn, Keras, PyTorch, Django
- Target audience: Online advertisers
Our client is an online advertising network providing customers with services that enhance online presence and boost ROIs. Their platform helps connect advertisers to publishers to present quality content in the right place and at the right time.
To help its clients get the most benefit from advertising campaigns, the company uses an RPM approach, which means calculating revenue per thousand ad impressions. Their web-based platform assists in building the robust ad strategy, including choosing the right ad format, publishers, devices, etc. However, a lot of that relies on manual data processing and lacks technical tools to facilitate revenue optimization.
The Client’s Request
The client contacted Velvetech to fill in the resource gap in their internal software development team. They wanted to come up with a solution based on a custom-built algorithm that would collect and process available data to provide insights on the ways to optimize ad revenue. That would help the company inform its decisions and serve the customers better. That’s why they asked our data engineers and data scientists to lend a hand with this project.
In order to provide high-level services and drive profits for customers, online ad networks have to handle large volumes of data sets. Our client’s platform is designed to help advertisers monitor different metrics, decide what ad makes the most money and why, and what ad units to choose for better ad performance. If done right, it results in extended reach.
However, there’s some data that can be missing at times. So you need to determine what sets of available data to use in order to see how to score an ad in a given situation. As ad units vary greatly, the options to monetize the traffic are numerous.
For example, there are articles, banners, right-rail ads, interstitial ads, and more. You also need to consider different devices your target audience use and how it impacts the revenue.
So having an algorithm that would make the process intuitive and data-driven would help automate analysis and support informed decisions. Being experienced in both data engineering and data science, our team was excited to tackle the task.
To foster implementation of the data science solution and make the process transparent, we divided it into three major parts. Thus, the project delivery included the following iterative stages:
- Creating the Proof of Concept and implementing initial data analysis
- Scaling the PoC to a production-grade
- Expanding the solution’s capabilities for revenue optimization
Each of the mentioned above phases would provide the company with a viable tech solution.
Data Science PoC Implementation
By researching available data sets and using statistical modeling, our data scientists first formulated the mathematically-based approach to optimize ad units, time intervals, impressions, and revenue.
Following this step, we built a robust strategy designed to generalize to real-world scenarios. It allowed the model to adapt properly to previously unseen data and avoid overfitting. The latter often occurs in data modeling when a certain function corresponds too closely to a set of data points. Therefore, leading to inaccurate predictions.
Lastly, our data scientists implemented the algorithm that relied on Python and aimed to boost the profits based on the analysis of data slices. As a result, the client acquired a proof of concept that incorporated real data and provided the opportunities for revenue growth.
Delivery of a Full-Scale Data Science Solution
While moving to the next part of project implementation, we had to include a data lake solution to organize the storage of the ad network data. This also required building CI/CD pipelines that would allow for integration of previously delivered Python codebase to process and manage data.
Ultimately, after completion of this stage, the company could benefit from the production-grade data science solution and apply it to large-scale datasets to maximize advertising performance.
Enhancing the Algorithm for Better Ad Performance
Finally, to help our client gain more insights into ad campaign performance and increase revenue per visitor, Velvetech’s team incorporated data analytics tools. We expanded the solution’s architecture so it could support conducting A/B testing, multivariate analysis, and some other analytical techniques.
Our collaboration on the project equipped the company with an algorithm that helps efficiently process and analyze key advertising data sets. The solution based on data science technologies allows the ad network to reinforce its platform and better cater to the clients’ needs.
Now, using a data-driven approach for various campaign types helps the client enhance click-through rates and better monitor ad performance. It’s feasible to compare multiple ad campaigns, forecast ad revenues, and help advertisers make informed decisions on how to build a profitable ad strategy.
As the project goes on, we continue to work on improving the algorithm and making it more sophisticated. Following the client’s requirements and adjusting the solution to their unique needs, our team plans to implement AI-based tools to further optimize the ad network’s platform, reduce possible errors, and increase ad performance.
How to Get Started
Our specialist will contact you to schedule a personalized consultation within one business day.
Velvetech offers complimentary consultations; after which, we will provide you with a proof of concept in just 3 days, an accurate outlook of the cost and timeline of your project and a competitive estimation, and an assembled team – ready to start your project within 7 days.
Proof of Concept in
Start Project within
Apply For A Consultation Today!