PREDICTIVE
LEAD SCORING
for B2B SaaS

Konstantin Bayandin
Founder & CEO Tomi.ai
Andrey Gavrilenko
Head of Data Science
PandaDoc
By clicking this button, you consent to the processing of personal data and agree to the privacy policy

Learn how predictive lead scoring
can help you leverage your data

At Tomi.ai we rely heavily on clickstream data to build our ML models and score visitors’ propensity to buy. We have invited Andrey Gavrilenko, Head of Data Science PandaDoc, to share his perspective on how large companies leverage data to create predictive scores and help sales teams hit revenue targets.

At Tomi.ai we are passionate about predictive lead scoring
because it drives business results at several levels:

  • Sales team’s happiness and efficiency: sales team is a costly unit and it pays back to spend their time on prospective leads only. By doing so we don’t waste their time on the leads that will never convert and keep the morale high because the success rate of the calls they make skyrockets.
  • Efficiency of marketing initiatives: scoring leads in real time provides immediate insights into the efficiency of traffic acquisition efforts, split by marketing channels. With predictive lead scores at hand you don’t have to wait for the actual value to reveal itself to make decisions.
  • Digital ad campaign optimization: predictive lead scores, sent to ad platforms via Marketing API, can be used as optimization targets to boost ad campaigns performance.
Speakers
Konstantin Bayandin
Founder & CEO Tomi.ai
ex-MarTech Senior Director Compass,
ex-CMO Ozon (#1 retail platform in CIS),
ex-BCG, Stanford MBA
Predictive lead scoring is a powerful tool for sales teams to prioritize their work on the hottest leads that will convert soon. From our experience, behavioral 1st party data defines 60% to 80% of the quality of buying propensity predictions, while external data enrichment provides lesser incremental predictive power with notable exceptions in some industries like B2B SaaS.
Andrey Gavrilenko
Head of Data Science
Kaggle master
In B2B SaaS business lead conversion from trial to paid is quite low, usually it is about 5%. It doesn't make sense to spend time and effort of the sales team on all the leads: if you do, they will burn out fast working with low potential leads, and you may overspend on the acquisition of the leads that will never convert. Large companies address this challenge by leveraging customer data they have available and creating custom AI models for predictive lead scoring.

Three tipes of data for predictive lead scoring

To build a lead scoring model you can use three types of data (or the combination thereof):

  • Behavioral product data
  • Clickstream
  • 3rd-party data (firmographic and technographic)

Behavioral Product Data

Pros

Very accurate

About 50% of the predictive power of a lead scoring model lies with the behavioral product data — the actions the user took after the contact form submission (actions taken in the admin area of the platform).

Cons

Revealed slowly

The problem is that it takes from 1 to 7 days for the intent to reveal itself, while the sales development team needs the score within minutes or hours, not days.

Clickstream

Pros

Very accurate and quickly accessible

The lead scores based on the analysis of users’ website behavior are ready almost in real time (as soon as the session is over) and are very accurate: from our experience this data has the predictive power of 30-40% and is second only to the behavioral product data.

Cons

Layout dependent features

When the prediction model is trained on the user behavior on the website, the introduction of significant changes in the design and/or layout will require recalibration or retraining of the model.

3rd-Party Data (Firmographic and Technographic)

Pros

Extended prediction horizon

We can use firmographic and technographic data to predict user behavior when the user is at the beginning of their journey with us. The prediction score is ready within minutes after the lead form submission.

Cons

Very low predictive power

Unfortunately, the predictive power of third-party data is very low, ~ 10-20%. Furthermore, we see that the trend for 2022 is that more systems are moving away from personal data transfer between platforms, ultimately obscuring their behavior history.

At Tomi.ai we rely heavily on clickstream data to build our ML models and score visitors’ propensity to buy. We have invited Andrey Gavrilenko, Head of Data Science PandaDoc, to share his perspective on how large companies leverage data to create predictive scores and help sales teams hit revenue targets.