Building a black box vs building for flexibility

How do digital advertisers want to manage their online marketing campaigns? Do they want a black box that works efficiently but in mysterious ways? Or do they prefer the flexibility of setting their own rules with total control and transparency, analysing and optimizing the results by themselves?

Looking Back

The first iteration of our product at Herolens worked as a black box. It was an ad server that optimized which creativity and product to show.

Just so you know, a dynamic creativity is just like a design blueprint that defines where and how each part of the product information will be displayed. Together with the product being shown it forms what we know as an ad.

For every impression the algorithm decided which dynamic creativity to show and which product to select depending on the user’s past behavior.

If the user viewing the ad had previous history on the advertiser’s site our remarketing algorithm would choose a particular product to show, and if we didn’t have information on the user we would pick products based on others people’s behavior on the site.

Of course performances of individual creativities and products were taken into account. For example, if a user had just viewed two products on the site but we knew one of those products wasn’t likely to be bought we would automatically pick the other one.

Our algorithm learned all by itself and optimised the creativity and product picking to increase their conversion rates. It behaved as a black box because there was no way of adding custom behavior defined by the advertiser.

For our initial customers that was more than enough. We were providing real value to them by optimizing their spending budget across different networks and bringing in more sales.

With time came bigger clients, and with bigger clients the requests became more complex.

We were asked to display ads in ways which didn’t make sense to us at first.

For example, one of those clients was a big online reseller of home appliances that had a deal with each brand they sold. They needed to showcase their products on top newspaper sites, but each product had to be shown certain amount of times.

Another client had special discounts on certain week days, so they wanted to show specific creativities and specific products on that day.

Others wanted to show different kinds of products and creativities depending on the weather.

Some of them wanted to use a specific creativities for each step of the purchase funnel. E.g: If you had just looked at a product you would be shown a more inspirational creativity, but if you had just added the product to your shopping cart you would get a creativity urging you buy with a discount on top.

There were thousands of possible variations, they all made sense, but even the smartest black box would never do all these kind of things by its own.

Before building anything new, we went on to look their motives for such requests.

A key factor we found out is how large companies operate. Big companies usually have different departments for business and for marketing operations. The marketing team usually get requests from the other departments because of deals they make or because of stock surplus.

Our clients did want to leverage the power of dynamic personalization and make use of our automatic algorithmic optimization. However, they also needed to make their ads context aware. In other words, they need their ads to have custom rules based on outside data.

Sharing the knowledge

Another common theme in their requests was that they wanted to learn with us what performed best for them. They did not want a magic wizard that did all the job for them and kept all the spell books locked in a secret chest.

After some time prototyping with pen and paper we came up with the Herolens Rule Engine

Rules allow CMOs and marketing analysts to plan, schedule and execute all kind of custom campaigns.

Our user interface allows them to create logic just by dragging different boxes we call “Contexts”.

With Herolens Rule Engine we introduced real flexibility in our ad serving allowing real expressive marketing campaigns on top of our optimizing algorithms.

Well, our whole ad serving makes use of these optimization algorithms, and Herolens Rule Engine is there to customize those algorithms.

So by default all our campaigns and ads are being optimized to show the most performing creativity and the most likeable product to be bought by each particular user.

Lessons learned

From the analysis and successful response to our Rules Engine we learned a special lesson that has become ingrained in our product philosophy.

We have to build our products with smart defaults, but flexible enough to give control back to our users so they can unleash their full potential.

Originally published at blog.herolens.com.