Optimised bidding and targeted advertising is a job for AI, says MediaGamma
UCL offshoot MediaGamma is focused on an AI layer to manage real-time bidding
For many people, AI is synonymous with the big tech companies but much of the work on the ground is done by small specialist firms and startups spun out of university computer science departments.
That's because while many of the machine learning algorithms deployed are well known, AI use cases tend to be highly specialised, those who know how to deploy them thin on the ground, and learning data in short supply. AI gives data scientists an opportunity to get in on the ground level, in the same way that big data startups made the running a few years ago.
A case in point is MediaGamma, an adtech firm of 18 people including five data scientists that was spun out of University College London with financial support from that university and the government body Innovate UK.
MediaGamma's original idea was to take the technology behind financial options and futures exchanges and apply it to ad trading.
However, they soon ran into technical and logistical problems.
"We tried to borrow the idea from financial markets and put it into advertising but we ran into troubles like brand safety, and separation and whether it's perishable or not," said Dr Shuai Yuan, VP of data science, speaking to Computing at the Spark + AI Summit Europe recently.
So the startup changed tack and tried to build a Demand Science Platform (DSP), a platform like Google Double Click Data Manager through which advertisers connect to online ad exchanges to receive opportunities and requests. But here they found a market that was already bursting at the seams, with hundreds of established players.
Finally, MediaGamma hit upon the idea of creating an intelligent intermediary layer that would optimise the real-time bidding process with the additional feature of being extensible into other areas of online advertising.
We position ourselves as a data science service provider to the advertising and ecommerce ecosystems - Shuai Yuan
"We decided to withdraw from the bigger platform idea and just focus on the AI layer," Yuan said. "Now we position ourselves as a data science service provider to the advertising and ecommerce ecosystems".
Rather than competing with DSPs, MediaGamma offers its AI platform as an add-on to them. Customers of Beeswax, for example, use that ad platform to manage advertising campaigns and reporting and also employ MediaGamma's AI layer to handle the bidding. The firm can also create user profiles derived from interactions with an advertiser's web page and has recently moved into building product knowledge graphs for ecommerce clients.
"They have products, users and orders and there are all sorts of relationships among these entities," explained Yuan. "If we can organise all of them together and build comprehensive relationships then with that graph we can create very innovative recommendation and optimisation solutions for them".
One of these is a recommender which automatically selects between different frameworks to find the optimal advertising approach in real-time based on metrics such as cost per acquisition (CPA) and matrix models that optimise accuracy and performance.
To make this work requires a great deal of attention to the way data flows from collection points and through the system. Algorithms are generic; it's how you use them that makes the difference.
"The model type or the algorithm in the mathematical sense is very general, there are only a handful of common models that are used, but in terms of the actual trained model it's different every time," Yuan said, explaining that the data pipeline is really where the differentiation lies. The company has built its own data flow system based on the Databricks Spark ecosystem.
"It's a customisable and portable data model pipeline. With that, we sort out the data ingestion, data processing, model building, evaluation and deployment in a single pipeline. And in this pipeline we evaluate different models," Yuan said.
What can artificial intelligence and machine learning do for you and your organisation? If you don't know yet, or want to make sure that you're not missing out, Computing's first AI & Machine Learning Live event is for you. To find out more, check out the Computing AI & Machine Learning Live website. Attendance is FREE to qualifying IT leaders and senior IT pros, but places are going fast