Book a Demo

    *Please note that your enquiry and contact details will be kept in a secure environment and will not be used for commercial reasons. If you want to contact us to remove your details from our database, please contact us.

    This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.

    Open Menu


    Read what's new

    March 1, 2018 by converge-digital

    Machine Learning in Adtech

    Machine Learning in AdTech


    In the world of Big Data, buzzwords like Machine Learning (ML) and Artificial Intelligence (AI) are used interchangeably to push the promise of better AdTech. In Short, AI is the broader concept which would describe the way in which machines operate and we would consider it as “Smart.” ML is the process of using our Big Data as inputs and telling the machine what “Goal” we would like. The “learning” part, is the way the Autonomous Software (Agent) analyses and creates patterns with this data, so it can be used to create actions and achieve the “Goal” depending on the Application.


    DSP’s for example can use placement data it receives through their creative macros in the auction to strive for Optimal Clicks, Viewability and even be find out how it can get the cheapest bid price on placements. On the other side SSP’s can use machine learning, specifically reinforcement learning, for forecasting of their inventory. The advantages that machine learning has over traditional trend analysis is that a machine will take in to account all data points given and can find a smart short term and long-term solution as opposed to following an unforgiving mathematical Model.


    It is fair to mention that although machine learning is a very clever way to build often intricate and dynamic patterns for a desired outcome, it’s not always the answer to the problems in AdTech. ML can’t negotiate more demand, supply or build trust between Advertisers and publishers as it is only an operational tool. This tool is computationally intensive to process the sheer volumes of data needed and won’t be as reactive to a quick change in state inputs it doesn’t recognise. The use of ML in the buying and selling of publisher’s inventory can also lead to a battle of the ML Agents trying to adjust to one another; each machines output become the others input. Although you would hope that this would eventually result in a harmonious agreement between buyer and seller’s agents, the fact that the both Agents are competing for optimisation will always throw the balance and could make planning difficult.


    As new DSP’s and SSP’s are developing and using ML as a selling point (ourselves included) without any long-term results, it is still anyone’s guess as to how it will affect the advertising ecosystem. The fact that these ML agents are behaving to each company’s motivations who facilitate the buying and selling means that unpredictable outcomes are highly likely. Entrepreneurs like Elon Musk, who advocates the enforcement of AI regulation, warns that there should be checks to make sure that these technologies are safe. In the case of ML on Ad tech, my main concerns would be looking at fraudulent traffic and targeting (specifically users with personal hardships). At this time there is no regulation for how the Agent is achieving its goal, and with AI becoming smarter exponentially, I would urge companies to employ a self-regulation policy with regards to keeping user safety at the top.