Knowing this, giant players, such as Google, Yahoo, Microsoft, and such, created a new market through the advent of Real-Time Bidding (RTB) online display ad exchanges, in which advertisers can bid to place an ad for a particular user visiting a particular page and no longer for a series of impressions on a specific web page. For advertisers, this is a much better ad space purchase mechanism for enhanced targeting. For publishers, this can also become a more efficient yield management mechanism – offering every impression to the advertiser willing to purchase it.

RTB ad exchanges create a huge number of ad space offerings on which advertisers need to bid in real-time, thus requiring automation both for purchases as well as for business intelligence, data-mining, and predictive analysis. In the recent past, many advanced technology service providers jumped in this newly created market to help advertisers leverage this new form of ad space purchasing, such as Demand-Side Platforms (DSP) which created tools to access traditional and RTB ad markets as well as to facilitate campaign management. These new types of “ad servers” can still manage campaigns as traditional ad servers could do, but  also added the opportunistic functionalities of the RTB and ad exchanges, including user data acquisition (profiles, purchase intents, as well as many other data helping advertisers with targeting users). Hence, large quantities of historical data are now available for advertisers through DSPs and other services, allowing the possibility to estimate bid effectiveness and associated outcomes.

There is currently a race to create advanced technologies to exploit this flood of data to come up with better bidding decisions. Better decisions consist of assessing purchase intents for every user to then establish bid price, and this for every specific ad and user. STEERads optimizes bids for each RTB impression for display advertising (on web pages other than search engines). We have developed optimal bidding algorithms based on Machine Learning (ML) and Deep Learning (DL), a technology derived from artificial intelligence research. These algorithms can be used by Demand Side Platforms(DSP)/Supply Side Platforms (SSP, representing publishers) and trading desks (representing agencies) to improve the ROI on campaigns. Most DSPs, SSPs and Trade Desks already pretend that they do optimization, but few are leveraging artificial intelligence. A lot is still done manually, which is still acceptable within the current market structure, but is highly inefficient in terms of manpower and ROI. With harsher competition on RTB-based advertising technology, data-driven optimization will become the core of a DSP/SSP/trading desk, but most of them do not master artificial intelligence technologies. STEERads acts as an enabler, becoming a plugin in the first place for DSPs and Trading Desks.

On the management side, Myriam Côté has developed a strong collaboration with professor Yoshua Bengio in leading machine learning projects. She has experience in both academia and industry where she worked as researcher, developer as well as director of operations managing the development and the delivery of high-tech products in start-ups.

STEERads benefits from a strong commercial team, including André Talercio, based in NYC, at a walking distance from most major DSPs, SSPs, and agencies.

Deep Learning (DL) is a revolutionary development in ML that can learn from huge apparently uninformative data (such as all the times when an ad was shown but nothing happened) by discovering statistical dependencies between all the observed variables and creating a representation for users, ads, key words, and web sites that is meaningful, and yields better generalization to new combinations of words, web pages, ad attributes, etc.

Optimal Bidding is a proprietary STEERads technology that exploits ML and DL in order to come up with the best bid (in terms of return on investment) to place on a potential ad placement associated with a particular user profile, intent, and context.

The question is not new, but today, optimization remains focused on the advertiser’s data (maximum eCPM willing to pay) and less on the user (potential user engagement).