Skip to content

The case for (and against) predictive analytics

November 14, 2014
tags:

It’s been a busy couple of months, and I’ve been learning quite a bit about business intelligence, big data and the opportunities and challenges in this space.

One area that has been a frequent topic is predictive analytics. As a lean guy, anything that promises improved business results by predicting the future immediately makes me suspect. I’ve been indoctrinated by the Lean philosophy to depend less on forecasts and more on the ability to observe and react to current demand and disruptions in a process.

That being said, I really depend on weather forecasts to get my kids dressed in the morning, so maybe I need to keep an open mind.

Predictive analytics is the next evolution in a long history of forecasting solutions that technology providers offer. For example, Kronos provides a labor forecast for retailers to help them in creating a labor schedule for the next week. This can be really helpful in supporting store managers in that it is really difficult to aggregate all the different patterns and unique events that are significant in scheduling a store. For example, day of the week is a fairly repeatable pattern and can be predicted fairly easily. Black Friday is also consistent. So rather than force the store manager to figure it out on her own, why not automate that in a forecast.

But ask a store manage if they completely rely on that forecast (or any other vendors forecast for that matter) and they will tell you that they use it for guidance. The reason for this is that there are many factors that affect a local store that aren’t used as drivers in creating the forecast. For example, if there is construction in the area that makes it more difficult for customers to get to the store or if a product is out of stock that week, the local manager will know that sales will not meet the forecast. This is why it’s so important to have someone knowledgeable about the local practice with the ability to react quickly to changing conditions.

Predicting levels of absence at a store or plant level is significantly easier that predicting individual absence. Make sure you understand the probability of success in a prediction. If you are providing guidance at the individual level, and the probability of a correct prediction is 60% then that means you are wrong 40% of the time. What actions are you asking managers to take based on this prediction and what’s the impact financially and in terms of system trust if it’s wrong?

A slight improvement over the status quo is good enough if a manager is already making the same decision frequently at an individual level. Hiring guidance for a similar job that has high turnover is a great example of an area where this works. Infrequent decisions are or if you are asking someone to take an action based on a predictive result is a different story. If it’s only a couple of percent better than the current method, internal customers are going to not understand the nuance of improvement and the project will be difficult to sustain.

When someone is having difficulty at home, it is likely their work will suffer. How is this behavior measured? While there are some outcomes that are measured like increases in absenteeism, these can also be attributed to difficulty with a supervisor or co-worker or a health issue.

Behavior is an area where correlation and causation can easily be confused. While we sense through data that something is wrong, it’s still going to take personal discussion to find out what the cause is. If you try and let software predict what the cause is and the manager takes a wrong action, it won’t be too long before the software isn’t trusted.

It’s very tempting to look for patterns in the data we already have. And no doubt there is lots of value hidden away in there that we have yet to mine. But we need to be careful to not try and solve every problem we have with existing data. In many cases, new data will be required to capture the true drivers of an event or behavior. This is a much more difficult endeavor.

So is there a future for predictive analytics? Absolutely. We just need to treat it like any other tool in our bag and not go around thinking that every employee problem we face is now a nail for predictive analytics to hammer.

Where are areas where predictive analytics excels?

Are there significant consequences for missing something?

Safety is something that comes to mind. If we can improve our ability to predict an increase in safety risk by even a few percent, the savings can be significant both in life, limb and dollars. This is a great area for exploring. Part of the solution needs to include understanding the drivers required to reduce the risk once an increase is predicted.

Will improvements in processing power or improved algorithms provide better insight than before?

This is the case for weather prediction. The data and algorithms were overwhelming the processing capabilities. As capacity to process improved, outcomes improved. This is also the case for customer behavior analysis. With lots of new data and increased granularity lower costs for processing have changed the game. Do you see similar opportunities with respect to labor analytics in your company? This could be a rich area to explore.

Predictive analytics is an interesting area, but let’s balance these efforts with the basics of making information more available and easier to use. Only then will we truly empower our employees’ decision making capabilities.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: