Together with Greenhouse Group, we sent in two cases regarding BidWiser for two different categories: a Programmatic Talent award in the category Campaign Manager (hello Mr. BidWiser!) and for a Programmatic Solution award (BidWiser). The Mr. BidWiser case (written from the point of view of an automated campaign manager) won the Programmatic Genius Award whereas the programmatic solution case that made the shortlist gives a better explanation of what it actually does. It’s the same case but from a different point of view. Since it’s a special award for a special product, we created a mash-up of the two cases that explains (Mr.) BidWiser in the best way possible.
Optimization within programmatic fails to live up to its expectations. There is a large gap in what the market expects that bidding algorithms do and what the final execution looks like in the DSP itself, because
the market is dominated by one-size-fits-all algorithms that need to scale well globally with the same standard optimization approach and predefined KPIs for every single client and market;
DSPs are not capable of effectively harnessing the immense quantities of (unique) data available to advertisers today;
DSPs are mostly used for mechanical tasks and not as much for thinking. Optimization is still a highly manual task with lots of operational hassle and trial and error and
Algorithms are often not transparent about how they operate, with little control or flexibility to alter the outcome to the client needs and KPIs.
Greenhouse Group and Bannerconnect wondered if they could solve these imperfections in optimization by applying Artificial Intelligence (AI), and so together they built BidWiser. BidWiser uses AI to automatically create self-learning algorithms that process many different variables alongside the client’s unique historical log level data (LLD) to reach optimum in delivery and performance, based on the (custom) KPI the client cares about. BidWiser algorithms look to existing LLD fields, custom variables and KPIs that are not readily available in a DSP such as custom viewability goals. BidWiser focuses purely on improving algorithms, not on rebuilding existing technology ground up, by relying on the strong and established technology infrastructure of DSPs for campaign execution and auctioning. The current version of BidWiser is built on top of the Appnexus Programmable Bidder platform (although Bidwiser has also been flirting with Doubleclick Bid Manager recently), which allows buyers to replace AppNexus Optimization with their own data and algorithms to value impressions differently.
How did we build this and who uses it?
BidWiser runs on a scalable data science workflow to collect, process, model, translate and deploy the algorithm. It just needs LLD to start. It’s built using Bannerconnect’s CORE for managing LLD, enriched with various machine learning techniques, and using the AppNexus API to automate the process of pushing decision trees to the DSP and to retrieve vital information on campaign delivery to inform the algorithm. Given Mr. Bidwiser’s passion for trees, he’s well suited to making some extra ones for us!
The main user of BidWiser is the Programmatic Specialist. The setup of a custom algorithm is a quick process. The specialists just need to select the campaign objects to learn on, the KPI to train the model against and the type of bid strategy, after which the model starts to train automatically and pushes the model to the DSP when ready. The specialist receives a visualisation of the decision tree via Slack together with a summary of the optimizations done. The model is being updated every night to improve itself daily, and since working to improve ourselves is an important aspect of our workplaces, Mr. Bidwiser fits right in!
BidWiser might even be the campaign manager of the future since he is efficient, determined, analytical and knows everything from a dataset. He also knows exactly how to use APIs and LLDs, and his best friend is AWS of course! This strong skillset ensures that BidWiser can tailor an algorithm for each campaign and associated KPIs. However, besides his distinctive strengths, BidWiser lacks social skills like keeping in touch with the client and ensuring the best deals with publishers which is crucial for a successful programmatic campaign. That’s where the human element is still vital.
The success of BidWiser depends on a combination of two key elements: the right balance in delivery (Did BidWiser meet the set KPI?) and performance (Did BidWiser meet the desired spend goal?). Dependent of the type of campaign, the programmatic specialist decides what to prioritize more and how to judge success.
To challenge the BidWiser algorithm and to verify performance there are frequent a/b tests being run by split testing users, 50% with the BidWiser algorithm, and 50% with the AppNexus algorithm enriched with manual optimizations by programmatic specialists.
BidWiser has created almost 10.000 models thus far, applied to many different clients. Results vary from 45% decrease in CPA compared to campaigns managed by top programmatic specialists, to 2.3x more delivery while meeting the same CPA as a manually optimized campaign. No wonder that BidWiser already runs on many campaigns in both Eindhoven & Sittard. And BidWiser runs every day, even at night and during all national holidays. And.. he hasn’t had enough yet!
BidWiser shows the great potential that AI can bring to programmatic. On the one hand by building tailor-made algorithms on top of strong and existing tech infrastructures to create new levels of optimization, and on the other hand by building efficient workflows by creating one smart campaign object, compared to all the headaches and many campaign objects that normally come with proper programmatic campaign management and optimization.
BidWiser also recognises its position to empower programmatic specialists and not to replace them. By letting machines do the heavy lifting and complex optimization, programmatic specialists can free up time to invest in areas where humans are still winning: building and growing relationships with publishers and tech partners to safeguard the best data integrations, the highest quality media and to work together on new and creative ad possibilities. In the end, you can have the best algorithm in the world, but if you put garbage in, you still get garbage out.
https://www.bannerconnect.net/wp-content/uploads/2018/07/MvM-rising-star-2.png400400Deesi Lihttps://bannerconnect.net/wp-content/uploads/2015/09/logo_bannerconnect_72.pngDeesi Li2018-07-03 10:01:362018-07-25 16:37:04Programmatic Talent Rising Star Mart van Melick
https://www.bannerconnect.net/wp-content/uploads/2016/08/Alice-in-Dataland-02.jpg400400Alice Bezetthttps://bannerconnect.net/wp-content/uploads/2015/09/logo_bannerconnect_72.pngAlice Bezett2018-04-03 10:00:372018-05-16 17:11:29Regression Models for Trait Prediction Part II