burger menu

#YardInsights – Using data science in a commercial world

23rd July 2019 by Scott McPate | Data, News & Events

The word data science can strike fear into the eyes of many business and marketing teams and with over 75% of data-led projects failing, it’s an understandable concern for an organisation. Previously, we have highlighted the topic of data science in our financial services event and SEO for Ecommerce event, held earlier this year. But, to properly tackle some of the common misconceptions around using data science in a commercial world we hosted an event in Yard’s Cardiff office and asked two experts, Jan Teichmann and Max Cheetham to offer their views on the subject.

Jan Teichmann - Commercial Success with Data Science

Joining us from Zoopla was Senior Data Scientist, Jan Teichmann. Having co-founded Cambridge Energy Data Lab and earned a PhD in Mathematics, Jan is now leading the data science team at Zoopla helping them to drive innovations from the vast amounts of property data they collect.

The focus of Jan’s talk was based around making a success story of your data science team. To begin Jan busted some of the most common misconceptions around data scientists and highlighted that organisations are probably not getting the most out of their data scientists currently.

For data science projects to be successful, following some simple rules for building a team could help improve results:

  • Motivation – is the project aligned with the business strategy.
  • Prep/Req – do you have a solid foundation with the right data, infrastructure and team structure.
  • Hiring – source the best candidates and avoid an expensive mis-hire.
  • Delivery – models in production or undocumented models on laptops, be clear on your delivery.
  • Retention – develop a culture and offer real career development.

Jan then explained that the majority of data science work is a mixture of finding, cleaning and preparing data for a model. When it comes to working with specific models, he suggests that to understand what will work it is best to distribute a request and evaluate a big number of models in parallel. The choice of a challenger model vs incumbent model requires a lot of evaluation, so allow experimentation in production which doesn’t impact users or the business objectives.

To close, Jan, shared a quote from Ted Dunning and Ellen Friedman, two of the lead experts on rendezvous architecture:

There is an important link between a scalable delivery pipeline and a happy data scientist. Delivery, technology and retention are all tightly coupled to the long-term success of data science.

Ted Dunning & Ellen Friedman

Max Cheetham - Predictive Attribution Modelling using Machine Learning

Our next speaker was Max Cheetham, Scientific Programmer from Cubed. His discussion centred on predictive attribution modelling using machine learning.

The common mistake for many organisations is that most common backwards-looking models, last click, first click etc don’t provide a complete picture according to Max. Instead, we’re looking for a fairer attribution model which looks at all converting and non-converting visitors and the reasons why.

Max, explained that factors such as on-site duration, pages viewed, checkout started, adds to basket and any other relevant factors can be added to build a more comprehensive view of a visitor, their interactions, how likely they are to convert and which channel is performing well.

At Cubed the model is developed not only to look at the positive channels but also those that reduce the likelihood of a sale:

If a channel or referrer has reduced the probability of a sale then our models most often provide a negative amount of revenue for that channel in our attribution models.

Propensity to convert is naturally highest at the - 1st visit (the visit directly before a sale). However, there is value to be evidenced at much further points, higher up the marketing funnel and further away from the sale visit.

On a final note, Max, highlighted that too often attribution reporting is seen as a final stage of a business/marketing process, whereas it should be at the centre of the business to help better understand the marketing funnel.



If you would like to attend a future #YardInsights event in Cardiff, Edinburgh or London or if you would to receive the slides from these presentations then get in touch at hello@yarddigital.com

Share this post

Related Content

The Yard guide to GDPR compliance

Like May the 4th is for Star Wars fans, World Cup final day will be for football fans, the 25th of May is a big day for…urm, personal data...

Big Data in 2018

Predictive modelling, neural networks, statistical analysis, R and Python. These are not new concepts by any stretch. However, what is (fairly) new is the willingness, and the growing ability,...

The Future of Analytics – State of Digital

Looking for something to read on your lunch break? Check out State of Digital’s new article on predicting the future of analytics and spot some familiar faces on the...