> ## Documentation Index
> Fetch the complete documentation index at: https://docs.ltv.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# AI segmentation

> The seven strategies, how LTV.ai ranks them, and the prediction models behind them

Rather than asking you to design a segmentation scheme, LTV.ai evaluates every strategy it has against your data and this campaign, and recommends the best three.

## The seven strategies

| Strategy                 | Splits your audience by                            |
| ------------------------ | -------------------------------------------------- |
| Category affinity        | Which product categories a customer buys from      |
| Discount affinity        | How much a customer responds to discounting        |
| Purchase frequency       | How often a customer orders                        |
| Average order value tier | How much a customer typically spends per order     |
| Recency                  | How long since the last purchase                   |
| Seasonal and weather     | Climate zone and time of year                      |
| Gift versus self         | Whether a customer buys for themselves or as gifts |

## How LTV.ai chooses

Every strategy is scored on four things, weighted:

| Factor            | Weight | What it measures                                              |
| ----------------- | ------ | ------------------------------------------------------------- |
| Effectiveness     | 50%    | How much usable data you have, and how evenly the split falls |
| Campaign context  | 35%    | How relevant the strategy is to this specific brief           |
| Prediction signal | 10%    | Model scores available for this audience                      |
| Seasonality       | 5%     | Time of year                                                  |

LTV.ai presents the **top three**, ranked. You can take the first, take another, or skip segmentation entirely and send one version to everyone.

## Everyone else

No strategy classifies every customer. Anyone who does not fall into a named segment is grouped into an **Everyone else** bucket so nobody is silently dropped from the send.

<Note>
  Seasonal and weather segmentation uses established climate zones, not a live weather forecast.
</Note>

## Prediction models

Separately from the seven strategies, LTV.ai can score every customer with trained models and expose the results as segments you can target directly: churn risk, purchase probability, lifetime value, discount sensitivity, and more.

Model availability is also the **prediction signal** factor in the scoring table above, so a brand with models enabled gets better strategy recommendations as well.

See [Prediction models](/audience/prediction-models).