As B2B companies shift to product-led growth, getting inbound leads becomes key to driving revenue. Many companies write blogs, host events, and provide free trials to generate demands. While these activities increase traffic, companies now have a new challenge to differentiate potential customers from the crowd. Since serving prospects cost money, ideally, companies want to focus on the ones most likely to convert. Companies are longing for a way to prioritize their sales outreach.

A lead scoring system is a powerful tool to tackle this challenge. It identifies patterns from historical deal data, predicts lead conversion probability, and ranks leads accordingly. Hoping to supercharge monetization, more and more companies are looking to build lead scoring systems in-house. However, this effort generally takes a few months. Such a long time to value is far from ideal. Therefore, in this article, I want to discuss a practical way to build a lead scoring system and obtain actionable insights within a month.

User journey.jpeg

Map user journey

To understand what factors come into play in a lead scoring system, let’s first map out the end-to-end user journey. A simplified user journey of a typical B2B product is shown below (double click the graph to zoom in). Since lead scoring systems handle inbound leads, outbound leads are omitted.

Lead scoring system.jpg

A prospect’s last touchpoints before conversion are sales calls and proof-of-concept. We first brainstorm metrics that may correlate with conversion and form hypotheses on metrics A, B, C that describe user behaviors in proof-of-concept. Since proof-of-concept is usually a guided process, we want to understand what customer cohorts are likely to schedule sales calls with us and how they behave without sales guidance. Therefore, we step back in the user journey and look at how prospects behave in product trials and interact with marketing emails.

Verify hypotheses on conversion

Next, we test hypotheses by checking correlations between metrics and conversion. Suppose we have segmented customers based on metrics H and I, and we try plotting metrics A, B, C by cohort. However, the data points appear to be random on the plots. Therefore, we take one step further, and we invent an engagement metric by multiplying metrics A and B. The chart below shows a clear pattern of the engagement metric among different cohorts. Now we can examine the correlation between the engagement metric and conversion statistically.

Screen Shot 2022-02-03 at 4.47.09 PM.png

Sometimes, visualization surfaces misalignment between data and business consensus. The misalignment may provide important insights for business. But at the same time, the misalignment may come from data quality issues. It’s not uncommon that data scientists spend more than 50% of project time dealing with these issues. Therefore, it’s expected not to find any meaningful pattern from the first data visualizations.

Predict lead conversion or assign weights to metrics

After finding correlations between metrics and conversion, we are ready to predict lead conversion rate and rank leads accordingly. Logistic regression is a sophisticated way for conversion prediction. However, such a prediction requires high-quality data with large volumes and multiple dimensions.

If a company is new to the market or is working on a data science project for the first time, its data is probably not sufficient for a prediction model. In such circumstances, it’s more practical to assign weights to metrics manually. The first lead scoring model may not be accurate, and it requires monitoring and iteration.

Side benefits of a lead scoring model

Early adopters of the above workflow obtain actionable insights within a month. While building a lead scoring model, companies discover metrics correlated to conversion and prioritize feature developments accordingly. Some companies identify customer cohorts that are unlikely to become paying customers and eliminate the low efficient outreach.

As a lead scoring model points to the right direction, sales teams become more upbeat as they book more calls, and calls turn to be more engaging. In the long run, a lead scoring model shortens meetings for pipeline reviews and gives management more time to think about strategic business decisions.