Machine Learning: It’s Still About People

3 min read

Never before has it been so easy to harness computing power to create value for the companies we work for and for our customers. Time required to develop an operational machine learning algorithm has been drastically reduced. In my analytics team, we are now creating models in days instead of weeks. Also, the solutions are of much higher quality, in terms of prediction power, than before.

This trend is global. Most organisations employing machine learning can probably relate to the progression, yet there is evidence that we are still very much scratching the surface. My go-to supermarket continues to send me coupons about female hygiene products and my kitchen appliances remain equipped with a disappointingly low degree of intelligence. In general, the pace of integration of machine learning in products and key business processes be it in marketing, logistics or sales has been far from remarkable.

Why is this? The improvements in analytical capabilities have mainly occurred in the technology and data space. Analysts and data scientists benefit from more intuitive and effective software for their work, cloud solutions allow easier application of machine learning models in different business processes and the growing amount of data combined with better data warehousing have improved the prediction power of our solutions. Not much, I would argue, has happened with the competence of analysts and organisations in general.

It is especially the competence of the non-analyst part of the organisations that is becoming critical. Many organisations have matured in their general analytics competence to a point where descriptive analytics, in the form of reporting and self-service BI, have become an integrated part of the work processes, but few are mature enough to understand the inherently black box nature of machine learning and to capitalise on the potential that these solutions represent. The quandary is that a successful implementation of machine learning strategy requires good ideas and use cases from those that have a deep knowledge of the products and business processes. It is rarely the tech-savvy data scientist that possesses these insights.

Data scientists and analysts are naturally very technology focused. Sometimes too much so. Sorry guys. The endless desire to test new technology and algorithms can distract from the objective of deploying a model (any model!) that contributes to a better customer experience or more effective business operations. The risk of developing a disconnect between the analytical resources and those responsible for business operations is real.

There is no easy fix to ensure appropriate focus and competence of both analytical and non-analytical resources in the implementation of a machine learning strategy. Bringing analysts closer to business operations and respectively business resources more towards analytics is a good start. In general, creating silos and strong centre of excellence models for analytics seems detrimental. Organisations should rather dare to experiment with agile, cross-functional, teams where analytics is brought closer to the challenges in the operative business.

The end goal should be an organisation where initiatives and demand for advanced analytics solutions arise naturally from the business operations ensuring both end-to-end deployment of analytics and a high degree of value creation.

 Jaakko Mikkonen

 

About Jaakko Mikkonen

Jaakko has over 10 years of experience as both analyst and manager of various analytics teams from If P&C Insurance. He is particularly passionate about customer analytics and analytical CRM. On September 1st, Jaakko will be joining Norrøna Q - a newly established company focusing on accelerating innovation in the Norrøna group, a Norwegian outdoor clothing brand.