Alexandra leads the research team with a focus on marketing measurement. She built and optimized Fospha’s Multi-Touchpoint Attribution (MTA) product using advanced Markov Chain theory and hidden Markov models. Using ‘linear regression’ and ‘casual models’ to build the Marketing Mix Modeling (MMM) product. Prior to that, Alex has been building Data Science proof and concepts and innovative product in start ups always focused on how research can help the business. She holds a master in Math Modelling and Scientific Computing from Oxford and a master in computer science from CentraleSupelec (Paris).
How were you initially brought onto the Multi-Touch Attribution (MTA) project?
When I joined Fospha Marketing in February 2017, the company was all about building Customer Data Platforms (CDPs) and finding interesting insights from our clients websites. However, we were looking for a more precise goal to focus on with the data we were collecting.
The attribution solution came into fruition because one of our clients (AVADO) wanted to have visibility on the keywords bidding strategy and understand what was driving conversions. They were particularly interested in knowing what keywords were zero-performance, i.e. zero return on investment; they saw a huge opportunity to reduce spend without affecting conversions and were able to save £200,000 of their budget! From this point on, we realised the unique opportunity attribution modelling presents for businesses, large and small across industries.
We implemented a few simple models, such as rule based models as well as a “best” model. We worked on developing the latter since we were careful to not discount the keywords that had potential. After achieving an unprecedented quick win for AVADO, our Chief Data Science Officer, Sepi, investigated the state of digital attribution and came across different data driven models out there. At this point, we were confident in our efforts to build an even better business with the tech we were using, the team and our abilities to create new models. The decision was made: we would be accelerating our marketing attribution efforts.
I was put on the project to develop Fospha Marketing’s Data-Driven Multi-Touch Attribution model. After an initial review, it emerged that the Markov Chain models would be the natural fit for us. It would enable us to provide our clients with granular insights and recommendations for effective marketing strategies and be able to progress over time as more complexity is added to the first basic model. I worked hard to ensure the first iteration of the model was rigorous and mathematically correct, giving the right framework and definitions for others to improve it. Research has always been of great interest to me and ensuring we could implement innovative and more complex models in the future was a priority. The first data driven attribution model came in to the world in August 2017. Since then, the team has worked relentlessly to improve both performance, regardless of the amount of data, and results to ensure our clients have the transparency they need into their data to be able to successfully use it to transform their business.
What did you learn from working on MTA?
I learnt that the market was not as mature as I’d expected it to be – people knew there was a problem but they didn’t know the solution or have the knowledge to understand the issue. This is a really complex problem, so part of our role is to educate the market. This is why our research is so key – data-driven attribution is a new model that has many layers of complexity. It can be really challenging but it’s also incredibly rewarding to be able to create models and tools that marketers can use to solve their data challenges.
What factors do businesses need to consider when picking the right model?
Before they choose the right model, businesses need to review their data – if their stitching isn’t right, they won’t be able to extract invaluable insights from their data. Fospha Marketing is based on a complex stitching strategy that helps clients make the biggest difference to their data models. Picking the right model is not always straightforward – businesses should conduct experiments to validate the model- but we need to remember what Sepi says (Fospha’s Chief Data Science Officer) “All of the models are wrong, however some are more useful than others” which he credits to George Box a statistician. One of the things that sets the Fospha Marketing model apart is that it’s flexible and adaptable for all businesses to make use of, they can choose how to characterize touchpoints, which metrics matter for their business and they can validate their model with MMM, modelling strategic applications of their marketing performance.
What excites you about the future of data-driven multi-touch attribution (DD-MTA)?
The level of data literacy amongst marketers and analysts is increasing all the time so we have more clients willing to conduct research – universities and researchers from the large corporations such as Facebook, Google, Amazon, etc. are coming together to address these problems, making it a hot topic and a priority for the industry to solve.
Regulation changes may present challenges for clients now and in the future. A big focus will be around stitching strategies – and ensuring that the data that is fed in the model is compliant and accurate. Regulations (e.g. GDPR) are absolutely key for data science as it is another component that forms part of a broader ethical framework. It’s important for us to help marketers and customers develop new techniques that comply with changing regulation with security and privacy always the top priority.