The machine learning behind LTV.ai
LTV.ai runs on two intelligences: proprietary ML models that decide, and a set of AI agents that build. Here is how the machine learning fits together.
Quick answer: LTV.ai runs on two kinds of intelligence. Proprietary machine learning models, trained on over 1.5 billion sends, decide what to send, to whom, and when. A set of AI agents turn those decisions into finished, on-brand campaigns. Machine learning is what knows; the agents are what build and act. Around both sits a set of feedback loops that make the system sharper with every send.
Why two intelligences instead of one
A large language model can write an email. It cannot know that a subject line underperformed for this brand last quarter, that a segment is fatiguing, or which customers are about to lapse. That knowledge lives in behavioral data, and it is learned, not prompted. So LTV.ai splits the problem: predictive models own the decisions, and generative agents own the craft. Each is used for what it is actually good at.
The decisioning layer
A suite of predictive models scores customers and campaigns: click and purchase likelihood, churn risk, lifetime value, best send time, next purchase date, discount sensitivity, and more. Separate models forecast the revenue and margin of a campaign idea before it is sent, so ideas are ranked by expected value rather than by hunch. These are grounded in each brand's own history, with cross-brand training where a single brand sends too little data to learn from alone.
The generative layer
A set of AI agents write the copy, design the email against the brand's visual identity, choose the products, and assemble the finished campaign. The agents are guided by the decisioning layer and by what has worked before, so the output is not generic, it is shaped by the brand's data.
The feedback loops
Three loops compound over time: reviewer approvals and rejections become durable generation rules, send performance becomes a creative playbook, and live results during a send steer the rest of that same send. This is why the system improves the more it is used, on each brand specifically.
Why proprietary models matter
Application-layer tools that only wrap a general model lose their edge as foundation models improve, because anyone can call the same model. The durable advantage is proprietary data and the models trained on it. That is the part a competitor cannot copy, and it is where LTV.ai invests.
Frequently asked questions
Does LTV.ai use large language models? Yes, for creative and execution. The decisioning is done by proprietary predictive models.
Is performance measured? Yes, against a holdout, for incrementality. Brands see up to 22% lift measured this way.
The nine systems below each go deeper.
- How LTV.ai optimizes subject lines while the campaign is still sending
- How LTV.ai forecasts a campaign's revenue and margin before you send
- How LTV.ai turns every approval and rejection into training signal
- How LTV.ai learns a brand's visual identity
- How LTV.ai turns send performance into a creative playbook
- How LTV.ai picks products for a campaign from plain language
- The predictive models LTV.ai scores every customer with
- How LTV.ai shapes exactly who receives a campaign
- How LTV.ai builds audiences three ways
LTV.ai is hiring engineers: see careers.
You can also read about the architecture in the LTV.ai and Anthropic partnership article, or explore the Proactive Agent.