How Product-Led Growth Teams Use Data Science to Uplevel Their Tech Stacks

by High Alpha

Over the last decade, a new paradigm called product-led growth (PLG) has emerged as a way for SaaS companies to go to market. These PLG companies have a few distinguishing characteristics around how they approach the market, including low- or no-touch sales cycles, free or freemium offerings, and often very short product time to value metrics. 

These companies are also disrupting the traditional marketing and sales funnel. Prior to the emergence of product-led growth, companies relied on a sales funnel with a linear customer journey studded with human touch points. Because customers live on a continuum and do not progress through the sequential steps found in traditional funnels, the funnel breaks down in PLG companies.

When you dig below the surface, you may be surprised by the sheer volume of technical components and data buried in product-led growth tech stacks. Read on to see where data science plugs into your tech stack infrastructure.

Data-Driven Customer Acquisition

The first layer of any PLG infrastructure is customer acquisition channels. No matter how wide the top of your funnel is, it’s key to track as much as possible in order to merge your marketing efforts with product usage data. This gives you an integrated picture of the customer journey throughout their entire lifecycle with your product — a critical component of product-led success.

How does data science help in the acquisition stage?

  • Customers need to be tracked from acquisition through to product usage. This is a key PLG concept that can be difficult to implement — and therefore often creates a large amount of information loss.
  • Tracking customers through their user journey allows us to stratify users into cohorts, monitor their product usage, and calculate traditional SaaS metrics (CAC, LTV, etc).
  • Some downstream processes, such as the PQL model discussed below, generate artifacts that can be used to create audiences for targeting at the top of the funnel. This creates a virtuous cycle in which we can characterize our best customers and utilize those findings for audience creation.

Lead Qualification with Data Science

The second layer in the stack is a Product-Qualified Lead (PQL) model. Think of it as a traditional Marketing-Qualified Lead (MQL) or Sales-Qualified Lead (SQL) model — but instead of using demo requests and downloads, you’re using app usage data to identify and target potential customers. In a low-touch sales model, for example, your PQL model could score leads and allocate resources to the most qualified leads. By stitching your product and customer data together, you can train models to predict who will convert based on product usage. 

The PQL process and its benefits can be broken down into a few points: 

  • A PQL is a lead who has experienced enough value using your product to demonstrate a high likelihood of becoming a paying customer. 
  • Propensity models can use machine learning to identify PQLs by learning to recognize certain behaviors that may indicate that a prospect will become a customer from training on the usage pattern data of current customers. 
  • Models can generate predictions of the likelihood to convert for each user, allowing your sales team to focus on high-value users by prioritizing high-scoring PQLs. 
  • By characterizing profiles of high-scoring PQLs, marketing can focus their efforts on reaching audiences most similar to PQLs. 

Propensity models are just another variation of the MQL/SQL/PQL processes. However, instead of modeling to understand your customers’ likelihood of converting, you are able to model a specific outcome such as propensity to spread. We’ve created propensity models for Docket, a High Alpha company, that identify product influencers (the users who are most likely to spread usage of the app). By modeling these users, Docket can reach out to their most impactful customers to drive product adoption.

Data Science for Cohort Identification

By analyzing users’ similarities on a variety of attributes and metrics over time, we can organize them into cohorts. Cohorts can be grouped based on any attribute you’re tracking, including acquisition channel, PQL score, or functionality flags, to name a few. By breaking users into groups, we can look at customers through different lenses, instead of the typical time-based customer acquisition groupings we see today.

Cohort analysis allows us to stratify large numbers of users in meaningful ways and find patterns among users that might be otherwise lost when looking at all customers as a whole. In addition to finding patterns, these cohorts also give us the ability to tailor messaging, alerts, and notifications to specific groups of users. 

Using Data Science to Predict Churn

Churn prediction models help us understand which customers are at risk — before they churn. These models bare similarities to the PQL model listed above, but focus on retention optimization. By knowing who is likely to churn, we can reduce turnover. These outputs can also help narrow your targeting. For example, if a certain cohort of users are regularly bouncing from our platform, we can use their information to build lookalike exclusion audiences. By tightening up on who is or isn’t targeted, customer acquisition can become more efficient.

In a product-led growth model, churn prediction models are essential for identifying methods of intervention for increasing and maintaining high rates of retention. By identifying product usage patterns, churn prediction models can surface product features that are most attractive to active users and allow a product team to encourage usage by suggesting or nudging inactive users to explore these features. 

Creating a More Unified Stack with Data

Mature PLG organizations instrument each of the aforementioned components in various ways. With careful instrumentation, PLG company leaders can build clear vantage points into product health and customer satisfaction. This strategy requires specialized tooling and thorough data capture throughout the customer lifecycle in order to have a good vantage on customer and product health. If instrumented properly, these systems can help to create acquisition efficiencies, drive product adoption, and improve customer retention.

Thanks to Mark Clerkin, Maria Patterson, and Charlene Tay for lending their data science expertise to this post.