Personalizing to Audiences

Growth marketers aiming to increase conversions (e.g. new signups or demos booked) have long recognized the value in identifying specific audiences of visitors and delivering personalized experiences that resonate with each audience.

• Job Role? - Product Manager, Developer, Sales

• Industry? - Financial Services, eCommerce, Insurance

• Intent? - Bot/Scraper, Casual Shopper, Serious Buyer

Challenge of Segmentation

At the same time, segmenting traffic into useful audiences is often considered the most difficult part of the problem. With most A/B testing and personalization tools, marketers can only define audiences based on attributes like:

• Page Referrer

• UTM source

• IP location

• …

While these basic, static signals are undoubtedly useful, their impact is limited because:

1. The % of personalizable traffic is small, missing out on valuable opportunities to personalize

2. Static attributes fail to capture how visitors are engaging with the site, ignoring critical clues about their interests

3. The actionable segments we actually care about, like Role, Use Case, and Intent, can typically only be identified by considering multiple signals together

As a result, many growth teams that experiment with personalization don’t realize the full benefit of their investment.

Where AI and ML Fit In

Machine learning offers the opportunity to automatically identify audiences from data, rather than requiring marketers to define audiences entirely on their own.

Models can be trained to predict a visitor’s job role, industry, or buying intent by incorporating a large number of user variables fed to the model in real time, e.g.

• Did they engage with technical blog content?

• Did they read a particular case study?

• First-time or returning visitor?

• Did they visit the pricing page?

• Session duration?

• etc.

Unsupervised models cluster groups of visitors that share common behavior, while Supervised models predict how likely a segment is to convert in various contexts.

In large organizations, such ML models are commonly developed by a dedicated team of machine learning engineers. Given the huge revenue impact of raising conversion rates by even a small percentage, the investment is justified.

An exciting development of the last few years is that advances in streaming data platforms and foundation models have greatly lowered the required infrastructure and training requirements, bringing these techniques within reach for companies with smaller teams and smaller traffic volumes.

What is Audience Amplification?

One of the most promising recent developments in AI-powered segmentation is audience amplification, which combines the insights of human domain experts with the statistical strengths of machine learning to greatly increase the percentage of personalizable traffic for desired segments.

Consider the following example:

A paid marketing campaign brings 100 visitors to your site. This audience, with ?utm_campaign=healthcare, receives a personalized experience that highlights your Healthcare solutions and case studies.

The behavior patterns left by these visitors is a treasure trove of clues about how healthcare-focused customers engage, where they come from, etc.

Now imagine that you could recognize an additional 900 organic visitors to your site that look the same as the 100 paid leads, and show them the same healthcare experience.

You just 10X’d the impact of your campaign.

How Does Audience Amplification Work?

From the marketer’s perspective, audience amplification follows the familiar workflow of defining audiences from simple rules like UTM source. Starting from this seeded audience, the algorithm identifies relevant user variables and trains the classification model.

1. Marketer seeds the audience. Marketer defines a narrow but reliable filter for the desired audience, such as “Landing page = Fintech Solutions”.

2. Algorithm identifies lookalikes. Algorithm identifies other visitors not in the seed audience that, nonetheless, display similar behavior, such as engaging with the same content or sharing common attributes with the seed audience.

3. System personalizes more traffic. The on-site personalization system begins showing the fintech experience to visitors that it predicts are more likely to be in fintech.

The result is a substantial boost in the number of visitors that receive a personalized experience.

Where to Go From Here

Next-gen personalization platforms like Sumatra are bringing advanced machine learning approaches like Audience Amplification within reach for all organizations, sparking fresh opportunities to make on-site personalization a key part of conversion rate optimization.

For growth teams looking for a marketer-friendly WYSIWYG editor and audience-building experience, Sumatra Optimize offers free and paid plans to support teams of all sizes. For enterprises who need more control, like on-prem deployment, data warehouse connectivity, and self-service ML model training, Sumatra Enterprise ensures teams never outgrow the platform.

What are you looking to optimize? We’d love to hear from you.