Turning data into performance with Modelling as a Service (MAAS)

Earlier this week I took the stage at the Lotame’s annual client summit to discuss our newly innovated and incredibly promising look-a-like and act-a-like modfeaturedelling solution. Modelling as a service (MAAS) hasn’t yet hit the mainstream status in the adtech and media landscape, but with the natural move towards activating big data in a meaningful way to drive higher performance for brands, as well as the requirement to consolidate activation into fewer and fewer media outlets it’s only a matter of time before MAAS becomes common practice. Below I discuss what MAAS, the workflow behind Optimizer 2.0, and why it’s set to grow and grow

What is MAAS?

Modelling as a Service or MAAS describes a process or solution whereby advertisers or their agencies are able to utilise external big data and technology experts in order to produce highly effective predictive modelling solutions which capitalise on machine learning and adaptive algorithms in order to define new target audiences by looking at an existing data set.

In simple terms, modelling as a service produces an audience output, based on pattern recognition, scoring and cross referencing an existing data set with a much larger user base to find similar users who exhibit a higher propensity to complete a goal action.

Sounds like look-a-like modelling

Like look-a-like modelling, the basic functionality of MAAS is to take 1st party data and to use this to model and scale out by capitalising on a database of 3rd party profiles. However, contrary to many look-a-like services, which are intrinsically linked to a buying platform for media activation, modelling as a service generates an agnostic audience output decoupled from activation, which can be activated universally across any given channel and through any buying platform.

Optimizer 2.0 optimisations are purely focussed on data, and not on bidding, creative, contextual placement, whitelists, blacklists etc. That means Lotame can focus on the data and the buyers can focus of the buying.

Optimizer 2.0 workflow

Although the technology and mathematics sitting behind Optimizer 2.0 is complex, the workflow for creating an optimisation model is relatively simple:

  1. Collect – we begin by collecting data on goal actions, media actions and conversion points, and ultimately anything you want to optimise your campaign towards. The data collection is used to generate a seed audience and it is the seed audience which will form the basis for our model and its outputs
  2. Profile – Once we have created the seed audience we need to profile the audience in order to observe the similarities and common behavioural patterns exhibited by the users within the audience. We create a “scoring card” to evaluate the data make-up of the seed audience, taking into consideration demographics, interests, geography and actions, and the relative frequencies at which the seed audience exhibits certain behaviours.
  3. Model – When we have profiled the seed audience we need to find similar users across the vast dataset of the Lotame Data Network. We use an advanced algorithm which cross references our scoring matrix with each and every profile in the Lotame Data Network. We are searching for profiles which look like the seed audience. We have complete control and flexibility over the scale and accuracy of the seed audience and can tune between scale and accuracy to find the optimal delivery point.
  4. Activate – Once we have our modelled audience centrally within the DMP we need to deliver and target the audience. We are media agnostic and can deliver and activate the campaign against the audience within any given context. The model continues to develop and evolve as more data is ingested and the campaign takes effect. Profiles are added and removed and performance continues to improve over time.

Data to Insights

One of the biggest and best USP’s of Optimizer 2.0 is the transparency we can offer to clients with regards to the output of the model. We share audience profile reports showing audience overlaps, and the common patterns identified by the optimizer tool. These insights can be hugely valuable in determining audience insights and feeding into the planning and the strategy for future campaigns.

Arrange a Conversation 


Article by channel:

Read more articles tagged: Featured

Customer Engagement