Machine Learning

10 November 2017

The presence of pattern recognition and analysis of trends through algorithmic research is commonplace within marketing and advertising research, with many projects utilising machine learning techniques to evaluate campaign effectiveness and to better understand consumer behaviour.

Although the questions that algorithmic research aims to address haven’t changed over time, machine learning has created innovative and often improved methods of answering these questions. Via finding a solution that hasn’t been explicitly programmed but is instead learned from data, which is the essence of machine learning.

Predictive modelling:

Fields such as discrete-choice modelling, which encompass methodologies such as ‘conjoint analysis’ and ‘max-diff’ are increasingly employed and have provided a refreshing alternative to the classic scale bias problem when analysing consumer preference (such innovation, resulted in the Nobel Prize for economics being awarded for the development of discrete choice methods in 2000). Using these methods respondents make trade-offs that simulate real world decision-making, utilising algorithms such as Hierarchical Bayes to analyse consumer behaviour in ways that were not possible previously with traditional research techniques.

Predictive modelling has also meant researchers can better understand consumers’ propensity to purchase as well as the drivers behind questions such as:

• Which factors determine why consumers make a purchase?
• How likely is a customer to convert to a competitor brand?
• What drives product recommendation?

Predictive techniques using machine learning are therefore key tools when creating brand strategy: identifying barriers to purchase, weak-points in the customer journey and the wider factors influencing brand perceptions.

At DRG we employ machine learning and predictive models in techniques such as key driver analysis, segmentation, conjoint and max-diff analysis to provide robust outputs that best meet your research and business objectives. We recently conducted a max-diff study for a well-known online printing company, testing consumer appeal towards a range of marketing offers. In this study we utilised the Hierarchical Bayes algorithm to build a highly predictive model which uncovered clear and easily interpretable results – visualised in our custom-built simulator tool. As a result our client was able to create a clear action plan going forward, based on valuable insight gained from machine learning.

Final remark:

Machine learning is a dynamic and fast growing field and although there may be some lag in research to adopt these techniques in comparison to other fields, it will be interesting to see how developments in these techniques will change open text analysis in the future. As well as how other methods such as neural networks may be employed – to further improve the targeting of marketing content.