How Linguistic Demographics Redefined Customer Segmentation

All of us have felt it. It’s been moving in silence below our feet like the tectonic plates of California. The Marketing world is going through a crisis.

A Major Shift

By now, we’ve all recognized that grouping and segmenting cohorts into huge clusters based on age is not very useful. Born in 81’ through 96’? Marketing probably has a terrible guess of your preferences…

Research has shown that a biological 55-year-old that communicates linguistically like a 35-year-old will be far more likely to behave like a 35-year-old.

Opportunity of the century?

Ascribe as many names as you want to this new era of marketing, but Gartner predicts that shifting to what they call “catalytic marketing” will drive twice as much customer engagement in commercially productive behaviors such as paying a premium or referring other customers to the brand. Whatever we call it, it’s clear that this shift creates a challenge, especially for teams that care about being relevant to specific audiences.

But as we’ve all experienced, properly segmenting a market is hard.?

Linguistic Demographics Save the Day

Exciting new AI approaches like Linguistic Demographics enable a new level of customer segmented insights that were typically only possible through direct surveys and panels. Previously unimaginable possibilities for the future of customer segmentation are now part of the new toolbox for F500 leaders.

Initial Linguistic Demographics techniques were pioneered at the University of Pennsylvania by Stitched Insights’ Founding Head of Science and R&D, Dr. Johannes Eichstaedt (Co-founder of UPenn’s World Well-Being Project and now at Stanford’s Human-Centered Artificial Intelligence Lab) and Founding Chief Technology Officer Dr. Andrew Schwartz, (Co-Founder of the World Well-Being Project and Computational Psychology).?

Our team started with the most rigorous in-person surveys in the world, the in-person surveys of the U.S. Center for Disease Control, setting out to make these types of surveys more scalable and affordable.

A Breakthrough?

The team happened on an unexpected breakthrough by analyzing tweets. Responses of in person-surveys were statistically matched on everything from: gender, age, socioeconomic status, education level, social influence, intensity of emotion and other psychographics.?

The critical distinction- this wasn’t accomplished by analyzing keywords.

State-of-the-Art unsupervised methods were created to analyze the grammatical structure, syntax and hundreds of variables in a vector space. These are now the most advanced linguistic techniques and models in the world and are being leveraged by top universities and world-health organizations.

New Techniques for Custom Segmentation

These new methods have enabled marketers and researchers to segment customers on a 1000X more granular level.

Instead of segmenting based on biological demographics such as year born, Stitched Insights enables the next generation of segmentation based on linguistic demographics such as linguistic age. “Research has shown Linguistic Demographics to be a far better predictor of behavior than biological measures”, said Andy Schwartz.

These new types of linguistic models give the ability to create custom segments based on internal business criteria. There are three broad ways of working with these tiles of models that require from some to a lot of training. Sequoia has recently published our favorite framework to think about this:

1. Train a custom model from scratch. This is the highest possible level of segmentation that leverages machine learning on custom definitions (e.g. stressed out soccer moms or excited cheerleader girls). A custom model like this can take into account both the linguistic demographics as well as psychographics that can be aligned with the business criteria. According to Sequoia, “This is the classical and hardest way to solve this problem. It typically requires highly skilled ML scientists, lots of relevant data, training infrastructure”. This level of customization can also yield the best results and platforms like Stitched Insights enable companies to deploy these solutions in key markets.
2. Fine-tune a base model. Sequoia rates this at medium difficulty (e.g. college-educated female millennials). “This is updating the weights of a pre-trained model through additional training with further proprietary or domain-specific data.” If done properly, this level of customization can drive the lion's share of the business decisions in certain markets and create a moat between competitors without these capabilities.
3. Use a pre-trained model and retrieve relevant context. “Technically, this is done by taking data, turning it into embeddings, storing those in a vector database, and when a query occurs, searching those embeddings for the most relevant context, and providing that to the model” as Sequoia points out. Pre-trained models can be powerful for mass market products and should not be underestimated.

Stitched Insights is set up to rapidly implement these three models depending on client requirements with many new verticals coming 2024.

Limitations of Old VS Benefits of New Techniques?

Linguistic demographics enables brands to create statistically unique experiences at each stage of any target customer’s journey. Compared to traditional techniques such as surveys, panels and social listening, these new techniques leverage billions of data points from real-time, first and third party data sources. This new type of precision segmentation is completely unhampered by 7 to 14 of the traditional research biases limiting current market research (post coming soon), the key tradeoffs include:

Limitations of Traditional Surveys

1. Biological cohorts
2. Lagging/infrequent data
3. Small sample size/sampling
4. Selection/interviewer bias
5. Expensive per survey
6. Low response/nonresponse
7. No way to segment competitors’ customers

Benefits of New AI Techniques

1. Behavioral cohorts
2. Real time
3. Billions of data points
4. Multiple data sources
5. Unlimited access to insights
6. Meta analysis/ longitudinal
7. Segmentation of competitive 3rd party data

Impact on Us

“Catalytic marketing” (as some in the industry are now calling it) needs to be supported by executive leadership that enables their teams to test and deploy these new machine-enabled solutions more effectively. All levels of the CxO from CMO to CTO must become leaders for the new generation of AI empowerment for their teams.?

Some of the biggest consumer companies in the world are well on their way -enabling multiple teams with platforms like Stitched Insights to help identify demographics of competitors shoppers across entire categories and connect product performance trends to the target customer segments.

The world’s best teams are busy creating the types of experiences that are changing customers’ understanding of their own needs and making them feel more confident moving in a new direction.?By accessing our Stitched Data Network (powered by our proprietary models) these F500 leaders have enabled:

1. Merchandising teams to create measurable more competitive merchandising assortments
2. Content marketing teams to produce the most relevant content for specific customer segments
3. Product teams to introduce new combinations of products and successfully expand into emerging markets

We're All Feeling It

New developments in AI are giving consumer companies a clear picture of shopper trends and platforms like Stitched Insights and Precient AI have enabled brands to be relevant to behavioral cohorts across all media.

Technologies like Linguistic Demographics are uniquely positioned to enable brands to create unique and personalized experiences at each stage of the customer journey. Our team focuses on linguistic aspects of demographic-segmentation and is always open to partnering with companies looking to grow by improving customer experiences.

About the Company

Stitched Insights is the world’s most powerful consumer insights platform. It’s the only SaaS platform enabling F500 consumer brands to identify competitors' customer demographics (like Gen-Z vs Millennials) and forecast corresponding shopping preferences based on linguistic behavioral cohorts instead of biological red herrings (e.g. linguistic age vs biological age). Powered by Linguistic Demographics IP developed by SI’s head of science, now at Stanford’s Human-Centered Artificial Intelligence Lab.