Propensity Modeling

Propensity models determine probability of a specific user taking a certain action — Login, Subscribe, Install App, Click on Ad, etc.

Why it matters:

Top of the Funnel always costs money. Brute force increasing top of the funnel without optimizing conversion rate isn’t wise. This is where propensity modeling identifies which users have the highest probability to convert.

Propensity models help manage tradeoffs (portfolio theory) while maximizing revenue.

• A single media business can operate conflicting revenue models (Revenue Diversity). For example, whether to direct users to Subscription plans or to continue showing free but Digital Advertisements monetized content.
• Most Escape Products need to balance out the opposing needs of your Owned Network and your customers (Digital Advertisements or sponsors).

How to use it:

Based on the propensity score, you can choose to hide widgets or show different visual elements or offer different incentives. For example:

• Push high propensity users to Subscription or higher plans
• Target Free Trials to high propensity users who have been exposed to the product but have not subscribed yet.
• Target Micro-transactions to low propensity users otherwise micro-transaction will cannibalize higher priced plans.
• Similarly, you can target discounts, upsell, or cross-sell.
• Use Onboarding for low frequency and low depth users
• Increase Pages Per Session by recirculating high frequency and low depth users
• Increase Sessions Per User by sending Push Notifications and emails to low frequency users

How to build it:

It is easier to build Propensity Not To X models than Propensity To X models.

Propensity Modeling combines customer information to estimate the likelihood of a certain customer profile to performing a certain type of behavior (e.g. the purchase of a product).

• Demographics (age, race, religion, gender, family size, ethnicity, income, education level)
• Psycho-graphic (social class, lifestyle and personality characteristics)
• Engagement (emails opened, emails clicked, searches on mobile app, webpage dwell time, etc.)
• User experience (customer service phone and email wait times, number of refunds, average shipping times)
• User behavior (purchase value on different time-scales, number of days since most recent purchase, time between offer and conversion, etc.)

Rule-based: If you’re implementing without Artificial Intelligence, then you looks for heuristics like:

• What is frequency and depth of engagement per user
• How many times is the viewing nudges, visiting price pages, etc.
• What is the affinity that a user has towards a specific author or topic
• Compute a discernment score for your audience based on the content they read

Footnotes:

• https://medium.com/swlh/estimate-multiple-propensity-models-and-compare-their-performance-using-a-model-agnostic-cb07413dbb31