Most SaaS products have a free trial or freemium plan. If you are serving trial customers, whether in Sales, Marketing, or Customer Support, you need to be able to predict which customers will convert to paid, so you can optimize your customer interactions based on that information.
With all the talk of big data and data scientists these days, using a term like “Predictive Modeling” may conjure up scary images – teams of engineers with giant Hadoop clusters, or indecipherable equations covering a whiteboard.
But there are easy ways to predict which customers will convert and which will not, using data that you are probably already collecting.
Step 1: What do the usage patterns of your Best Customers look like?
Examine your best customers, and how they are using your product. What usage metrics do they have in common? And when they first sign up, is there a common pattern to how they begin using your product?
Every business will have different metrics that define the customers who are most engaged and most likely to convert, and even within the same company you may have differences between customer segments. Depending on the specifics of your product, we want to look at usage metrics like the following:
- number of API calls
- amount of data stored,
- number of servers deployed
- number of actions taken
- number of records created
- number of users per account (or linked accounts)
There are some metrics common to most web applications, which can also be useful metrics to look at:
- Number of logins
- Last login
- Signup date
- How soon they started using the product
- Time period they used the product consistently
You may only have a partial idea of what your best customers look like, or be unsure where to set the thresholds for segmenting your customers. That is ok, as you can always adjust the model over time as you learn more, or compare the effectiveness different models. Pay particular attention to how your best customers and most enthusiastic fans use your product, especially when they first sign up.
Using all of the information above, define one to three metrics that correlate to your best customers, and determine some thresholds that will categorize how engaged or likely to convert each customer is. For example, customers who have signed up in the last 60 days and created over 20 records, or customers with more than 5 servers who installed a security agent on more than 2 of them.
Step 2: What about your (almost) Never Customers?
The second step is more important if you have a freemium (free forever) pricing tier, but should be well understood for free trials as well. Are there groups of customers who almost never convert, or are generally unqualified to buy your product in some way?
For example, if you sell a product which can be used by both businesses and prosumers / hobbyists, there may be a segment of your customers who will never upgrade to paid, as they simply don’t have the business need or cash flow to justify paying.
This is not to say that these customers never have value. If you are selling to developers or providing open source solutions, having large numbers of unpaid customers can be critical to fostering a user ecosystem. For many B2B products, free users can serve as a long term lead base if they switch jobs and are in a position to use the product at their new company.
But given that your resources are generally scarce, you want to focus your sales and support efforts on those likely to convert. Knowing who is likely to be “unqualified” is nothing personal, just an important way to segment customers and prioritize interactions.
Step 3: Any Getting Started Trends / Warning Signs?
Step 3 may be less important than the previous two, depending on your business, but it pays to think about usage as it relates to time, to complete our predictive model. Are there actions that people take when they first sign up that are an indicator that they are likely to convert?
For example, they complete their profile or create a new project, enter an API key or connect via OAUTH to a 3rd party system, or perform X actions in less than Y days. Those actively looking for a solution that your product solves will exhibit very different initial actions than someone who just saw your company on TechCrunch (and who signs up for everything they see featured there for some reason, regardless of what it is).
And what about the duration of their usage? For example, one recent SaaS industry survey found that “Free trial users who were still active during day three of their trial were four times more likely to convert into paying users than the average customer.”
Perhaps this seems like an obvious conclusion, given that in many free trials 50%+ of signups don’t even bother using the app at all. And another large chunk of users play around for a few minutes and then get distracted, never to return. But if the three day statistic is true for your business, you would want to follow up with those customers more aggressively, or at least send different Emails to those users right away instead of more generic marketing emails later on.
Conversely, are there patterns of usage and then decline which are warning signs? Perhaps customers start using your app and then run into problems. They burn through thousands of Emails or API calls and their usage then trails off, or stops suddenly.
If you can proactively identify those customers and contact them, there are some real sales opportunities there (often they are having a simple problem, or may have a misunderstanding of your product’s capabilities). This data can also be useful for churn prevention in the future for a customer success or support team.
Step 4: Connect your data to your CRM system
The next step is to connect your user and usage data from your SaaS / Web App backend database into your CRM system. By integrating your app’s data into your CRM, you gain several advantages.
First, you empower your sales and support people (assuming they are using the CRM) to see customer usage data right where they look at other customer information. Having to log into another system, or viewing customer usage in an administrative web page or Excel sheet that is outside of their regular workflow is inefficient at best, and will result in them not using the data at all in many cases.
Secondly, you can use all of the structure and functionality already present in most CRM systems to manipulate the data and calculate the predictive models we have set up in steps 1-3. Otherwise, you have to instrument all of step 5 in your app, which is generally not something that Engineering is going to want to prioritize over features in your product.
Once all of the data is in your CRM system, a good marketing analyst or sales operations person should be able to customize views, set alerts and triggers, build reports, and more. This both empowers your team to work with their data without engineering help, and makes things more nimble as your company gets experience working with the predictive model and updating the model and associated analytics as it evolves.
Step 5: Implement the Predictive Model in your CRM
Finally, the fun part. Once your customer usage data is being updated into your CRM system, you can implement the predictive model based on steps 1-3 above. First, we start with the raw usage data that has been imported, and typically create summary or roll up fields which total up those records, giving you some metrics (Total API calls, Total Builds, etc.)
Depending on the complexity of your model, those totals may be enough (a predictive model could be a simple as Total API calls > 1000). But often we want to consider several metrics to come up with a “lead score” for your predictive model (for example, Customers who have > 5 GB stored currently but have had >100 GB stored over all time could generate a “storage elasticity” score for those customers, where that metric might be important in identifying a certain type of customer who is likely to convert — or for a different product, likely to churn).
A score can provide a more nuanced model, as well as a way to sort or prioritize customer accounts in the CRM system. But whether you build the model from a single metric, or a number of metrics pulled together by a formula, you can then use the model to begin organizing your customers in the CRM system.
You can modify the views that certain sales reps see, so unqualified customers don’t show up in their main screen, reducing the signal to noise ratio and making sales more efficient. Views can be sorted by the new predictive model lead score, so the hottest accounts bubble to the top automatically as their usage changes. For certain scores you can use workflow rules or triggers built into the better CRM systems to even assign follow up tasks or send email alerts to your team members.
Next, aggregating all the usage metrics and predictive scores into reports provides analytics and dashboard visibility for your managers and executives. With customer usage data being updated in your CRM, right where the new accounts are created, you get to see trends and understand changes in customer usage, signups, churn and more, faster and easier than would generally be possible with manual Excel reports. And business intelligence is now accessible to a larger number of people, some of whom can be empowered to create their own reports to derive further value from the predictive model scores and usage data.
If you have read this far, think to yourself what visibility you have into trial customers, and how they are using your product. Chances are you can create a predictive model to understand where customers are in your funnel, and improve conversions and revenue, in just a few easy steps.
CloudAmp provides Customer Analytics for SaaS companies, directly inside of Salesforce CRM. If you have a trial or free plan, we can show you who your best leads are based on their real-time usage.