The 5 biggest pitfalls in predictive analytics

Author: Daan van Beek
Managing Director

Table of Contents

Featured image | The 5 biggest pitfalls in predictive analytics | How to avoid?

The five major pitfalls in predictive analytics

Predictive analytics is a powerful instrument for many organizations. It helps them to create competitive advantage and make their business processes more effective. Insurance companies and credit card issuers for example use it to detect fraud, cops use it to catch criminals, sometimes even before they commit a crime, and car dealers apply analytics to predict the chance that someone responds to a campaign. From our experience we strongly believe that there is a lot of added value in predictive analytics. But, prediction is very difficult – especially if it’s about the future (Niels Bohr) – and making automated predictions is even more difficult. There are many pitfalls on the road to success. We think this are the five most important pitfalls:

1. Just mine all your data

Some people think that predictive analytics is a panacea and expect that the software will be the solution to every business issue, even the ones that are still unknown. They just gather a lot of data, they install the mining tools and see what pops up. In the ‘rarest of rare’ cases they might find a golden nugget, but often the mining software comes up with worthless or spurious correlations. Before you start mining your data, you should have a clear view of the actual business application and ask yourself “which business problem do I want to solve and which data do I probably need”.

2. Unskilled data scientists

Predictive analytics is a multidisciplinary skill that requires deep understanding and knowledge of statistics, data massaging and applying the right visualisations. There are not many people in the world who are able to do the job properly and make a success of it. Our advice: if you have a challenging business issue where predictive analytics might help, hire the best data scientist you can find and guide and treat them well. Foster these guys and girls. They are absolutely worth the money.

3. Don’t manage the user’s perspective

Once you have booked success with your predictive algorithm (it works and the outcomes make sense) don’t suppose that everyone is immediately enthusiastic. Especially the business users of the predictive algorithm. Give them some time to trust and believe the outcomes. Cops for example who were patrolling the streets by their gut feeling or random impulse are now guided by machine learning models who tells them where they might have the highest chance to catch a thief. And we all know that predictive analytics has it right every time! So, don’t forget to manage the user’s perspective and take your time to improve the analytical process as a collaborative effort. Over and above, test and validate the system on regular basis which implies: organize that you get feedback from your users.

4. See it as a stand-alone application

Predictive analytics can’t be successful as a stand-alone application in the long term. It should be based on a solid data warehouse infrastructure so the data can be properly cleansed and integrated. In fact, you need to embed predictive analytics in your daily business processes. And make sure that the predictions can be easily used in other modules of your business intelligence platform (reporting, dashboarding, analysis and visualisation).

5. Bad data

If bad data may decreases the added value of your regular reporting applications then bad data will nullifies the benefits of predictive applications for sure. In fact, bad data is a huge risk for your company if it fuels your predictive engine. Cops for example catch innocent citizens or drive to the house of a real bad guy not properly armed. This isn’t open for discussion. Decisions based on dirty data in reporting or dashboarding might have medium or big impact (depending on the level). Decisions based on dirty data in predictive will have tremendous impact on your financial results, the customer satisfaction and your company’s reputation. If you certainly want to fail with predictive analytics, don’t bother about data quality.