At Keplar, delivering accurate consumer insights begins with building a highly representative simulated audience based on relevant, high quality data. This crucial first step is what enables our customers to obtain immediate feedback and valuable, actionable insights on their product concepts and marketing strategies.
Large language models (like GPT, Gemini, LLama and others) power Keplar simulated audiences. While these models have a ton of contextual data built into them, they often lack specific relevant information about product domains, recent consumer behavior, and important user actions (such as purchasing). We use three main sources of data to construct a rich, relevant, near real time source of audience data that powers Keplar simulations:
We start by identifying our customers’ target markets and building a base simulated audience of digital twins using public data sources.
For example, for one of our customers, we used public social media as a data source to find their target consumers. Then Keplar began to make observations about these audiences by analyzing behaviors, preferences, attitudes, psychographics, and associations embedded in the social data. In other words, Keplar extracted signals that gave our customer a robust dataset reflecting perception, personality traits, and cultural context. On the basis of these detailed observations, a high quality simulated audience was built which yielded over 80% accuracy when compared to side-by-side qualitative research.
Base Keplar audiences are great for fast concept and message tests, providing directional alignment relative to traditional market research approaches. While these audiences can be significantly improved with first party data integration, they're a great way for new customers to onboard Keplar and evaluate the platform before integrating internal consumer sentiment data.
To refine Keplar audiences, owned consumer data can be integrated in the form of primary research and consumer feedback.
As an example, for one of our customers, audiences were built by combining social data with primary research on a cohort of users. These users were recruited for a hybrid study. Opt-in access to their social accounts enabled us to deliver a high quality base audience, which was further improved with targeted primary research on attitudes, behavior, perception, and product usage. The primary research provided more specific and relevant context to the simulated audiences, driving an improvement in the quality and specificity of feedback coming from Keplar on new product concepts.
In this example, new primary research was conducted, but Keplar can also ingest past primary research or consumer feedback from companies directly to improve audience capabilities.
Finally, the gold-standard for consumer prediction involves calibrating Keplar personality models to real consumer behavior. Consumer behavioral data can include things like sales and purchasing data, usage analytics, or reactions to marketing assets. Such signals can be used to improve digital twin choices to better track real-world behaviors and outcomes.
A specific application of calibration that Keplar supports involves regularly using POS (sales) data to make audience choices reflect real purchasing decisions. Doing so enables us to offer a living, self-updating, fast Market Share/Incrementality Forecast which can be used to project market shares for innovation concepts at a regular cadence without high cost. Such quantitative statistical simulations can enable a team to not only take products to market faster, but track and adjust to market conditions as they develop.
Keplar’s approach to building simulated audiences integrates foundation models with cultural context, primary research, and historical data. This comprehensive methodology results in a highly customizable simulated audience that is benchmarked for performance and delivers precise, actionable insights.
To try Keplar, or to discuss ways in which we can partner to transform your insights, research, and testing functions, contact us!
