Customer Propensity Data Modeling
FarmLogs is the leading farm management software in the United States offering products in the operational, agronomy and marketing areas of running a farm. I joined FarmLogs in 2016, about 4 years after the company was founded, where they roughly already had 150k users on the platform.
FarmLogs in a lot of ways is a freemium business model. They offer some free tools and features for the majority of their users and then upsell to various users some paid features. So with over 150k users on FarmLogs and a sales and marketing team working at lead nurturing and sales strategies how do you identify which users are prime for the upsell? This is the perfect job for an analyst or someone who is willing to get their hands dirty with some data.
If you don't know what a propensity model is then it is essentially a method to try and identify who is likely to buy or upgrade. Many people out there believe that it is impossible to predict who will become a customer, and there is some strong data to suggest that. However, in our case at FarmLogs we were able to successfully build a propensity model that actually drove results.
The main problem we were trying to solve for was making our sales team more effective with their time. Every salesperson is going to get turned down, that is the nature of the work, but our hypothesis was that we could give sales a lead score that indicated a higher probability of their likeliness to close the deal.
The first thing I will say to you if you are trying to build a propensity model for your business is to not overcomplicate it. Try and be practical and be sure to put yourself in your user's shoes. It will make you a better analyst. I'll say this over and over again, there is an art to being a good analyst.
I am going to generalize some of the sales figures with FarmLogs to protect their material information, but you will get the point still. Let's say the numbers, in general, looked like this:
150k Free Users
1,000 Paid Customers
For definition's sake, I will be talking about 3 types of users. 1) Users - those who are free users within the app 2) Prospective Customers - historical users we analyzed that later became a customer and 3) Customers - users who paid for FarmLogs.
At the time we already had over 1,000 customers and I had access to the historical data of back when those customers who just free users. This was the most important thing and you should try and do this too if you can. I started pulling prospective user behavioral numbers on what their activity looked like before they became a paying customer. Some of those data points I pulled to analyze looked like this:
- How many weekly sessions they had and for how long
- What percent used both our web and native mobile apps
- Which specific features were they using
- What size of farm and types of crops were they farming
From here I was able to get some trends for the user behavior of a prospective customer looked like. This was my baseline data set I used for the next step. The next step was to compare these trends with that of our free users. The important thing we did next was looked for what "stood out" as big differences and deltas. For example, if we found that 95% of all our free users and paid customers were corn farmers then this wouldn't really help us because it was the same across the board. We wanted to find the deltas that could be represented as potential indicators for what a prospective customers behavior looked like. With this we were able to highlight things like this:
- 90%+ of prospective customers were using both our web and mobile app vs. only 20% of our general users were on just one of the apps
- 50% of prospective customers were documenting their farming assets within the app to log various farming jobs vs. our general users hardly ever used this feature
- 80% of prospective customers using our apps several times a month vs. the general users weren't as active
With those indicators and a few more we found, we were able to tag user behavioral events in our marketing service provider, HubSpot, with a lead score that was weighted accordingly based on the propensity so our sales team was notified of users that were much more likely to convert.
After running this lead score model for a few weeks the results were in... and.... it lifted the sales conversion by over 40%! Our sales team was wasting less time on users that weren't quite ready to buy and spend more time with much more qualified users. Then on the flip side, our marketing team was able to start marketing various features and activities to our free users to educate them on ways they could get more out of FarmLogs.