What is DCO and why is it important?

The “What”

Dynamic Creative Optimization has been around for a while yet, surprisingly, it’s still little known by some; enough to warrant a blog post. The term originates in display advertising, since it provides effectively a way to do personalized messaging, and it is a form of programmatic advertising that allows advertisers to optimize the performance of their creative using real-time technology. At its core, it uses multivariate testing to optimize, in real time, the marketing message and tailor it to the user.

Typically DCO involves creative development, identifying the test variables, figuring out the optimization objective, and method of optimization. Creative development is done using ‘classic’ creative studio tools (e.g. Adobe Photoshop) and it can include video, images, animation etc. The test variables represent the parts of the ad creative that are changed during the multivariate testing. These normally include graphical elements, text, colors and so on. The optimization objective can be any action we would like our users to take, such as a click, signing up to an emailing list, downloading a brochure or completing a purchase. It is this action (our objective) that we are trying to optimize as part of the process.

While the term is mostly used in display advertising, it can also be used in the context of a product, like a website or an app. For example, if you are trying to implement on your desktop website a section for ‘related content’ you can think of that as a section advertising your content — which just so happens to be on your own website. Therefore you can apply the same mechanisms and methods as you would with DCO in the context of display advertising, effectively optimizing your users’ journeys on your website by doing so.

As such, it’s safe to say that the term DCO can refer to personalizing content for the user both in the context of a product / website (‘on-site’) as well as in the context of display advertising, where we reach out to the user on another site/app/etc via the means of an online advertisement shown (‘offsite’).

The “Why”

The clue is pretty much in the name: DCO provides a way to optimize (personalize) the content for your users, which means that you will see better metrics around the objective. Typically this would mean an increase in your users conversions, or the number of clicks, or the CTR, or a decrease in the bounce rate etc.

The bottom line is that you will optimize the user experience which in turn yields higher returns. Even more so, the deeper the personalization, the better these metrics will look.

The “How”

How does it work? The answer is somewhat simple: data and machine learning. At the core, DCO is all about using as much data as possible, from your website, your analytics, or 3rd parties and layering them over a real time machine learning system which continuously adapts to the trends in the data and produces these recommendations for each user. We break down the problem in smaller parts and using the recommendation engine we run a continuous multi-variate test.

Let’s look for example at a simple banner that you might see on Property finder website, promoting some of the properties available:

How did we end up with this result? Remember the multi-variate bit about DCO? We have to first dismantle the banner into its component:

Background:

Property text :“2 & 3 BR luxurious apartments & penthouses in Marina”

Property attraction point: price (“AED 1.8M”)

Property image:

Now that we have established the basic components of our banner, these constitute the test parameters. So let’s establish some values for them and see what would our multi-variate / DCO structure look like:

As you can see, dismantling the banner’s content to its basic components allows us to optimize each component individually — this provides a very scalable way for delivering optimized content to your users.

If you think about this single example, in a standard, non-DCO environment you are looking at generating 3 (background) X 2 (property text) X 3 (property attraction point) X 3 (property images) = 54 individuals banners that need to be created. That is a lot of effort to create these units! Employing a DCO approach means you only have to provide 3 colors , 2 texts and 3 property attributes and only 3 images; a much smaller effort. This approach allows you at any point to change some of the parameters — adding more properties is just a simple case of providing more property images, changing background colors does NOT mean recreating 54 (or more) banner units, but instead changing the colors for the ‘background’ parameter, and so on.

This is one simple example — but you can make this as complicated as you want. Take for instance a simple functionality on your website such as search. Users come to your site, select their search criteria and expect to see results matching them. In the case of Property Finder, it would be something like this:

If you look at the above you start noticing that logically there are a few logical components that emerge here: buttons, descriptions, images, property attributes (price, bedrooms etc) So we can employ now a similar approach and dismantle this into smaller components and optimize for each.

You can use DCO to answer interesting questions now such as:

• Does the current user react better to images of the beach, the skyline or the gardens?
• What are the attributes of a property the current user reacts to best? Is it price? Number of bedrooms? Local amenities? Square footage?
• Does my user prefer email or phone calls when contacting the agents? Are they making decisions together with someone else — in which case is a ‘share’ button essential?
• Are there any keywords in the property description which specifically capture the user attention?

The possibilities are limitless. Even more, you can go one level above now and think of the whole search results as a unit and dismantle it into its components, which are property details. By dismantling the search results this way, we can start thinking about which properties are going to be surfaced to the user first, like how Google approaches its ranking.

If the user is interested in properties with a sea view, then we can push those to the top of the list. If, however, the user seems more interested in the ones within a certain price range then those should be surfaced first. The user will still have access to the whole data set, and through things like pagination they can still traverse all the properties, but we can now influence the user journey through our product by surfacing at the top of the list the properties more relevant to the user.

If you combine the two of them together you get a whole new level of personalization through a DCO pipeline, where on one hand you are surfacing the properties really relevant to the user and on top of that, when showing the properties in the list we can decide which information to surface for each property, tailored to the current user.

The end result? Personalization nirvana!

The “When”

When should you decide to employ DCO for your product and digital marketing? YESTERDAY!

By now all of the big guns of online are employing some form of DCO — Netflix, Amazon, Facebook, Apple have all been running their own recommendations and optimization engines for a while so you are already late to the race. At Property Finder we constantly seek to iterate on our DCO approach and innovate how we surface relevant content to our users.

Many others are doing the same and soon enough this will likely be the standard way to deliver content on a website or in a product. DCO is no longer the future, it is a common way to deal with your users, it is the present.

This article was kindly contributed by Liviu Tudor, senior software engineer at Netflix and DCO expert.

If you liked this article and want to be part of our brilliant team of engineers that produced it, then have a look at our latest vacancies here.