Jumping on the Analytics Bandwagon—Don't Just Do it, Do it Right

There's a lot of hype around analytics today, and the potential for businesses is genuinely exciting. But if you don't have the right approach, you may find yourself failing before you even begin.

Analytics today is at the peak of its hype cycle. Everywhere you look, it’s a board level conversation. CEOs have clearly indicated that if they do not incorporate analytics and data science they will perish. Every programmer wants to become a data scientist. VCs are pumping money into analytics companies just the way they did during the dot com bubble. If the hype is there then there has to be opportunity too. This is precisely the reason that getting a Fortune 100 senior executive on a call to listen to a start up in the analytics space has become super easy these days. However, the conversation goes like this (EC here means Enthusiastic Customer):

EC: One of the top items on my agenda this year is to include data science and machine learning in my BU/organization.

Me: Great. What’s the business problem you are trying to solve using data science?

EC: I want to do predictions. Do you have a predictive analytics offering?

Me: That’s great and yes we do. But what do you want to predict?

EC: You need to figure that out since you are the expert and have a predictive analytics offering. Anyway, what can you predict for me?

Me: We can do failure predictions, asset breakdown predictions, asset degradations etc. Do you have the data for this? Say time series sensor data? Around 3-6 months worth?

EC: Not much. We do collect some data through sensors and have some process data? We have really old machines but never really installed sensors on it. Can you apply machine learning on those machines?

Me: Machine Learning to solve what? Without data there is nothing the machine can learn and hence it would be best to re-engage when you are ready.

EC: Ok. I’ll keep you posted. Thanks for your time. Great tech btw!

Now, What’s the flaw in this argument? The very fact that the customer did not think about the problem he wants to solve, did not think about whether he has the data nor did he envisage what the solution is going to do was basically the start of a downward spiral. Doing machine learning because the upper management has an agenda around it or you heard your competitor apply it is not the best way to approach the problem. Below are the questions you should ask yourself to see if you actually need data science and machine learning:

1. What is the problem I am trying to solve?
2. What data do I have to solve this problem?
3. What am I looking to predict? Am I predicting for problems? If so do I have enough information about those problems?
4. Is my data large and complex (large number of attributes) or is it small and simple (limited attributes)?
5. What would be the success criteria if I was to engage with an analytics company today?

There are three clear pillars that need to be considered:

1. Clear and defined problem
2. Data to solve the problem
3. Success criteria for building the business case

If any of these three are not met then the project is likely to fail. I’ve had customers who have invested thousands of dollars on Big Data in the same way only to realize that they had never defined a problem statement for it. Follow the KISS (keep it simple and strategic) strategy here and you will be fine.