It's 2004 and I’m just starting out in my data science career. I’m assigned to work with a drug store chain that’s about on every corner in New York City. They want our team to conduct a customer behavior exploration project to understand why customers purchase certain products and if there are any patterns. How often do customers purchase each product? What else do they buy?

Our team develops a simple system: we take each product in the store and try to guess what a customer also purchased during their trip into the store. For example, if a customer buys cough drops, can we guess what else might be in their basket?

Using the past data available from old receipts and product sales, we can find all the people who bought cough drops and compare what happened against our hypothesis. What are they buying? Apparently way more Vicks VapoRub than one would expect.

We go through this process with every product in the store. When we look at gum, to our surprise, customers are also purchasing mouthwash at a really high rate.

Think about this for a second and imagine your own pharmacy. Stores usually place gum in the candy aisle or at the cash register. It’s nowhere near the toothpaste, toothbrushes, or mouthwash. But customers who want a fresh breath (mouthwash) are also conscious of their breath at other times (gum).

I share this data with the client and soon after, all these drug stores now put a gum display near the mouthwash. Sales skyrocket and as a result, the client’s team flashed their own pearly whites.

JOURNEY TO THE PAST 

When most people hear the term “predictive analytics,” they assume it’s about predicting the future. However, there’s an even more powerful technique few people use. While the mouthwash story seems like a rock star, one-in-a-million discovery, it’s not: my team and I merely looked at the past data to predict the future.

Why is it important to predict the past? Well, it gives us actual data on what happened in those different scenarios. If someone bought gum, what else did they buy? We can see the answer to that.

For healthcare companies, this is incredibly important. All companies, including healthcare provider companies, want more customers. For healthcare provider companies, their goal is to drive membership and acquire new members.

If those companies start with how many members they had in 2015, they can use that data to understand what would happen if they lowered the cost of premiums. By applying the hypothesis to past numbers, they can see what the opportunity (and lost opportunity) would be if they had taken that action.

What would your customer base look like if you had our fair share of population? Everyone makes an equally random decision about healthcare and what company to choose. If companies compare current holders to a single variable decision (spouses, for example), they might notice some discrepancies and ways to impact the bottom line.

WHY PREDICTING THE PAST IS IMPORTANT?

If you need buy-in for a data science project, here are some benefits your organization gets when you focus on predicting the past.

• It tells you what changes you need to make. By predicting the past, you can quickly determine if your hypothesis is correct and create a delta in your analysis. For example, say your organization wants to increase membership. If you are predicting the future, you are hoping to add an additional 10,000 members in two years through various marketing efforts. However, if you focus on the past, you may see that in your data set, you have 9,000 members who are married but their spouse isn’t a member. You argue that if you target these wives and husbands and get them under your fold, you’ll gain 9,000 additional members. You now have a delta (an increase in 9,000 members by targeting wives and husbands of current members) as opposed to seeing a gain without realizing why or how it happened.

• There are no surprises. Predicting the past is like a controlled experiment. Let’s say I’m working for the drug store, but instead of 2004, it’s the year 2000 – the new millennium! I’m asked to predict how many people would by bottles of water to assist forecasting for the next two years. Simple enough. However, after 9/11, bottles of water sales increased exponentially. This external factor is something I could predict and clearly affected the prediction and forecast. Remember: you can isolate factors in the past, but you don’t always know what will happen in the future to cause changes.

• It’s much easier for the business to get its head wrapped around it. When people predict the future, they usually talk about percentages. If we raise our prices by 5%, we should see an increase in revenue by 25%. Except, it doesn’t always work that way. If you are working in the past, you can tell a business leader you would have made $1.4 million more profit if you took your premiums down by $5. That will get their attention as it’s a lot easier to project it to actual numbers. Think about your own experience with coupons: would you rather save $20 or 20%? Even if $20 is less, most people understand those savings because to them, the actual number is the one that’s real.

Everyone wants to be a fortune teller and predict the future. Cue the crystal balls! So many organizations spend their time, talent, and energy on predictive analytics and predicting the future. The reality is that there’s so much power on predicting the past and it’s rarely done. Nobody does it!

If you want actual numbers you can use to generate memberships, revenue, and profit for your organization, take a journey into the green pasture of the past – you won’t believe what you can find!