Industry Consumer Goods and Retail, Media and Advertising, Hi-Tech, Social Sciences and Development
Specialization Or Business Function Media and Advertising (Clickrate Optimization), Pricing and Actuarial (Price Level Indications, Pricing Relativities, Claims Processing Optimization), R&D (Performance Analysis, A/B Testing, Design Optimization), Exploration and Production (Predictive Maintenance), Engineering and Design (Electrical/Electronics, Mechanical Engineering), Change Management
Technical Function Data Warehousing (Data Mapping, Data Integration), Data Engineering, Software and Web Development, Robotics
Technology & Tools Data Analysis and AI Tools, Web Analytics Platforms, Programming Languages and Frameworks, Machine Learning Frameworks
Control Theory Project- CPA Optimization
Expert consultation for a software PID Controller Implementation and Tuning
In a nutshell, this is Taboola’s main business model:
On each widget, we recommend content from multiple advertisers
Every time someone clicks on a piece of content, they’re directed to the advertiser’s page, and Taboola gets a referral fee called CPC (Cost Per Click)
The income generated is shared with the publisher, measured at RPM (Revenue Per Mille impressions) based. Mille stands for a thousand.
Taboola chooses which item to show to a user according to an algorithm that ranks items by their expected revenue [eCTR * CPC = eRPM]. On each impression the item with the highest expected revenue is shown:
eCTR - Estimated Click Through Rate (Clicks/Impressions) is predicted by our algorithm
CPC - Cost Per Click - given as an input by the advertiser set on the campaign level
While advertisers pay per click, their marketing goals are usually defined by cost per action (CPA) and scale of actions. Action can be - purchase, lead generation etc. This means that one of the clients’ challenges is to generate Actions at scale while keeping profitability (i.e. CPA that is lower than the worth of actions) which can be achieved by adjusting the CPC of the campaign.
Taboola’s advertisers invest a lot of time in manual optimization of their campaigns, trying to achieve the optimal balance of CPA goal and scale of actions while tuning the CPC on the campaign level and on some sub-campaign levels taboola offers. Setting the CPC too high means the advertiser bids will be very competitive but the price might be too high resulting in too high CPA. Setting the CPC too low means the advertiser bids might not be competitive and might not get enough clicks/actions or might not ‘win’ the valuable user clicks.
Our agenda is to automate this manual process of tuning the CPC by utilizing a feedback loop which will be able to optimize the Campaign-CPA towards a given CPA goal.
We want to build a feedback loop with two (and possibly more) independent inputs that can influence on the campaign level CPA which will be evaluated and compared to the client’s input Target CPA
In addition, since there are multiple inputs we control, we believe there will be multiple points that satisfy the CPA goal. Within the points that satisfy the criteria of CPA <= Target CPA, we would like to be able to optimize for additional metrics - for example, number of actions (aks conversions scale) or maximum profit, etc. we can use such inputs as additional indication to tune and impact the feedback loop and the campaign CPA accordingly.
We are looking for an expert in multivariable control systems and optimized control systems. The project will comprise an expert consultation for the tuning process, plus joint practical work with the team that is developing the controller. The deliverable will comprise several tuned instances of the controller, on choice placements (TBD) - and a simulation on real data - which will serve to prove the proposed concept for tuning the controller and for cleaning the noise. The result will be a well documented tuning procedure and noise filtering method to allow automation of the procedure by the team towards large-scale deployment.
The project has strong potential for follow-on projects based on the success of the initial implementation.
Your proposal should emphasize your background in control theory and practical implementations of control systems in software with very large scale data, as well as your background in Ad-technology or performance advertising;
You should also outline which algorithms and approach you intend to use, and how you propose to measure the success of tuning and noise filtering.
We will share further information with the expert and jointly refine the approach and scope in detail with the relevant exp