Are You Getting Value for Money from Your Customer Analytics Data Science?
Any company worth its weight in gold is using customer analytics. Tracking your customer’s behaviour as they use your website or mobile app allows you to gather essential information to help you make informed business decisions. Data science, machine learning and customer analytics are invaluable to any business looking to increase sales and let’s be honest which company isn’t.
No doubt you are investing heavily to achieve your customer analytics data but how do you know if you are getting a sufficient return on your investment? Are you spending more than you need on your customer analytics and are you doing the right things to ensure value for money? Are your customer analytics projects set up to fail or failing?
As Head of Technology at Acrotrend, I’ve worked with many blue-chip companies from Reed Exhibitions to GlaxoSmithKline to help them see how and when their customers are engaging with them today and predict how to connect with them in the future. Here are my top tips on how to ensure you are getting value from your customer analytics data science.
KISS – Keep It Simple Statistically
Simple models, like boosted decision trees or random forests, are sufficient for most situations. The focus should instead be on reducing the time between the data acquisition and the development of the first simple predictive model, gaining accuracy through repetition and volume. Techniques such as ‘ensembling’ are invaluable in giving your models the critical mass of data to be valuable. Aggregating results from multiple simple models is a quicker route to insight than developing one vastly complex model.
It’s imperative that practitioners have the ability to effectively outline and investigate multiple prediction problems FAST. Instead of exploring a single business challenge with an all-consuming complex machine learning model, we recommend exploring multiple in parallel, building a simple predictive model for each one and assessing their value. It is far more efficient to model multiple simple systems and combine the insights than build one that covers the entirety of a complex system, especially as not all hypotheses explored will prove to be valuable. By building small and simple on mass, you get value from your modelling efforts quickly.
Whilst we live in the era of ‘Big Data’ where it’s possible to process petabytes through a model, this is beyond the capability and budget of many players in the marketplace. Also, the value one gets from processing vast volumes of data through a model diminishes exponentially when considering the purpose for the insights created. Instead of focusing on distributed computing to process vast data lakes, develop techniques that enable the derivations of similar conclusions from a sampled data set. Through spending effort in refining a sampling method, it’s possible to explore more hypotheses with a smaller data volume.
Age of Automation
Process automation is key to reducing the delivery of that all-important first model. By applying standardised data processing techniques to pre-process/clean raw inputs, it’s possible to streamline processes to remove many of the traditional delays to a data science exploration. Furthermore, by making use of technology and standards already prevalent within the wider software development community such as continuous integration and DevOps, the evolution of the solution created can happen at pace.
If you would like to know more about how Acrotrend can help you with your customer analytics data science, please Contact Us and we would be delighted to have an exploratory discussion with you.