In a previous blog, we looked at doing “what if” predictive modeling. We’re going to take that a step further today, and look at how you can model multiple scenarios to drive the right outcomes for your business.
In this example, we’re using sample SaaS company data and looking at ways to maximize MRR (monthly recurring revenue).
Our SaaS wants to predict how to increase MRR based on transaction history and event data that they’ve gathered. The data set includes common data points, like:
- Unique company ID
- Customer length
- MRR per customer
- Customer satisfaction metrics
- Feature usage
- Company size
- Compute used
Running the model on the whole data set can provide a broad suggestion for how to increase revenue. When we ran the model, we saw that the most impactful predictors are:
- Company Size—Not surprising, as it’s very likely that larger companies spend more.
- Time to Production—Interesting that this is a predictor, and something that we’ll look at.
Next, doing a “what if” scenario prediction can determine if we would expect overall MRR to increase if Time to Production improved by 50%.
This screenshot shows a few rows of data from our predicted scenario. While some customer payments are predicted to decrease, there is a general upward trend in MRR, with an overall average being a 2% increase across the entire data set.
That’s interesting information, but what if we segment customers by size? When we retrain the model and segment by size we see a couple of interesting things:
- MRR for larger customers is driven largely by compute usage. Compute usage is driven by Feature 4 as well as Time to Production, which leads to overall higher compute usage.
What happens if you’re able to reduce Time to Production, as we already mentioned, and increase compute usage by driving adoption of Feature 4? MRR for your large customers increases by about ~4%.
- For small customers, the most impactful predictor for MRR is whether they’re using Feature 5. For large customers, Feature 5 barely registers.
So what happens if you’re able to get additional small customers using Feature 5? MRR increases by a predicted ~6%. Going a step further. We can look at what drives the usage of Feature 5, so that a product manager or marketing manager can make a decision about what needs to be focused on to specifically drive the outcome of more small customers using Feature 5.
Based on this, we can see that completing Tutorial #7 is a strong predictor for customers using Feature 5. This is an example of what the predicted adoption of Feature 5 looks like if the company gets more users to complete the tutorial. Y under predicted_outcome means that they are predicted to start using Feature 5 if they were to complete the tutorial.
If I were the Product Manager or Marketing Manager, I’d be working on a campaign to drive more small customers to the tutorial and Feature 5.