How LTV.ai builds audiences three ways
LTV.ai builds send-ready audiences three ways: ML behavioral segmentation, guided strategies with data-sufficiency checks, and natural-language segment building.
Quick answer: LTV.ai turns raw purchase and engagement history into send-ready audiences through three complementary layers: machine-learned behavioral segments, guided strategies that check whether the brand's data can support them, and a natural-language builder that turns a plain-English description into a live audience.
Layer one: machine-learned behavioral segments
The system learns latent taste dimensions from a customer-by-category purchase matrix using matrix factorization, groups customers with clustering, and models what a customer tends to buy next given what they bought last with a transition model. These power category-affinity and purchase-frequency segments, with a sensible fallback for customers the models have not seen.
Layer two: guided strategies with data checks
There are several guided strategies, category affinity, discount affinity, purchase frequency, order-value tier, recency, seasonal, and gift-versus-self. Crucially, each is scored for viability against the brand's data before it is offered: a brand with only three months of history is not offered seasonal segmentation. This is surfaced as data-availability indicators so marketers only pick segmentations their data can actually support.
Layer three: natural-language segment building
A marketer can type an audience in plain language, for example customers in a city who bought twice but have not opened recently, and the system interprets the intent, compiles it to a query, runs it, and streams back a live count. The result can be saved as a reusable segment.
Why three layers
Different needs call for different tools: automatic behavioral discovery, guided strategy for common goals, and free-form language for anything specific. Every resulting segment can then drive per-segment product recommendations and copy.
Frequently asked questions
Does it invent data? No. It uses real purchase and engagement history, and it will not offer a strategy the data cannot support.
Can I build a custom audience? Yes, in plain language, with a live count as you refine it.
Part of the machine learning behind LTV.ai, a series on how the platform works under the hood.
See it on your store: book a demo.