Smart Signals: Machine learning model for predicting conversion, expansion and churn

In setting up a product-led growth motion, manually defining detection conditions (aka signals) have reported to be the hardest thing to do. Our customers repeatedly told us that, while they wanted to start with intuitive rules they observed from product analytics, they mostly ended up being disappointed about the amount of false-positive being generated. They typically got those false-positives because of edge-cases, such as signups belonging to alternative ICP segments, or people already knowing your product and thus skipping onboarding steps. Also, there’s always the logic that some signals will only become relevant when matched with other signals...

We think the better way is to have the PLG platform continuously learn from any conversion, expansion and churn motion, and have it decide which signals to look at, and with which impact.

Yet, machine learning on its own won’t be meeting customer requirements. There are also the practical needs that were voiced:

• The machine learning engine cannot be a black box! If operators cannot see what signals at which impact are being used, they will not trust it.
• While recognizing flaw-backs in intuitive rule setting, one still wants to add manual signal rules to the engine’s signal model.
• When products change, so does usage. And as such, operators want to get clear insights how signal impact changes over time, and when they even become irrelevant.

Taking all above into account, we’ve built Smart Signals: a revolutionary machine learning engine that computes probability for accounts and users to convert, expand and churn, while providing clear overview of each signal and its impact.

How Smart Signals work?

journy.io starts with understanding the journey stages in which users and accounts can reside, at any given moment. These stages are characterized by being free, paying or churn. Here are a few journey examples, with their stage type:

• Lead (Free) → Trial (Free) → Freemium (Free) → Starter (Paying) → Corporate (Paying) → Enterprise (Paying) → Churn (Churn)
• Lead (Free) → Trial (Free) → Trial Expired (Churn) → Starter (Paying) → Corporate (Paying) → Enterprise (Paying) → Churn (Churn)
• Lead (Free) → Freemium (Free) → Starter (Paying) → Corporate (Paying) → Enterprise (Paying) → Churn (Churn)

Next, from analysing and comparing segments of users/accounts that have converted (Free → Paying), expanded (Paying → Paying into higher tier) and churned (Free or Paying → Churn) in the past, journy.io is able to identify and propose uniquely differentiating signals.

With a simple click on a ‘✨Smart Signal’ button, journy.io offers a list of all relevant Smart Signals.

Upon journy.io operators/administrators confirming above Smart Signals, or manually creating intuitive signals, conversion, churn and health scores will automatically be computed. Also the impact of these signals will automatically be finetuned over time, as users/accounts move along in their journeys and the ML model gets more trained.  

Signals impact Scores that impact Health, Playbooks, and Segments

As signals are being confirmed and our ML engine is computing conversion/expansion, churn and general health scores for each individual user and account, there are certain default dependencies that get influenced.

To start, health (good, normal, bad) is by default directly impacted by aforementioned scores. Also default playbooks for trial-to-paid conversion, expansion and churn risk detection heavily rely on these scores, and so does certain key segments that is used throughout the system.

The impact of changed and added signals to our ML engine is processed in such a way, that sudden fluctuations are avoided. And thus, downstream marketing, sales and CS workflows aren’t that easily affected when a new signal gets triggered or not.

Smart Signals as part of a PLG motion

As a product-led growth motion is all about automatically detecting signups that are likely to buy, to expand and to churn, signals are simply key.

You can easily implement and iterate on your intuitions, yet still leverage machine learning and data science to get those confirmed, or rejected. While journy.io was never intended to replace internal data science initiatives, it augments current data-driven processes. Many of our customers still leverage internal data science models and use our functionality on top of that work.

And in the end, even after considering all practical customer feedback and allowing flexibility to the ML model, we are still measuring and optimise workflows beyond what the ML model indicates, so you can be certain that you always operate by the best converting workflows possible.