> ## 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.

# Prediction models

> Model scores you can target, from churn risk to discount sensitivity

<Info>
  Prediction models are enabled per brand. Contact your account manager if you do not see them.
</Info>

Where enabled, LTV.ai scores every customer with a set of trained models and exposes the results as segments you can target directly. Find them under **Manage Modules > Prediction Models**.

## The models

| Model                   | What it predicts                         | Segments                                                  |
| ----------------------- | ---------------------------------------- | --------------------------------------------------------- |
| Email click probability | Likelihood of clicking a campaign        | High, medium, low                                         |
| Purchase probability    | Likelihood of buying in the next 30 days | High, medium, low intent                                  |
| Churn risk              | Likelihood of lapsing                    | Critical, at risk, safe                                   |
| Customer lifetime value | Expected value over 30 days              | VIP, high, mid, low                                       |
| Optimal send time       | When a customer is most likely to open   | Eight three hour slots                                    |
| Next purchase date      | When the next order is likely            | Imminent, soon, upcoming, distant                         |
| Engagement score        | Overall engagement, 0 to 100             | Champion, highly engaged, active, passive, dormant        |
| Discount sensitivity    | How much discounting a customer needs    | Full price buyer, occasional deal seeker, discount driven |

## Using them

Model segments appear alongside your other options at the **Audience** stage. See [How audiences work](/audience/how-audiences-work).

Two are worth calling out.

**Discount sensitivity** is the margin protection lever. Target full price buyers without a code and reserve discounting for customers who actually need one, rather than discounting everybody to reach the few who required it.

**Optimal send time** powers the scheduling option of the same name. Instead of one send time for everyone, each recipient lands in whichever of eight slots they are most likely to open in. See [Scheduling and sending](/send/scheduling).

## How scores are produced

Models are trained outside the dashboard and the scores are loaded in on a schedule, so a newly synced customer will not have scores immediately. Stale scores are pruned automatically.

Scores also feed [AI segmentation](/audience/ai-segmentation), where model availability is one of the factors weighing which segmentation strategy LTV.ai recommends for a campaign.

<Note>
  Prediction quality depends on history. A brand with little purchase or engagement data will get thinner segments, and some models may not produce useful splits at all.
</Note>