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

# Generating predictions

> Apply a trained model to a population to produce scores

Training teaches a model the patterns in your history. **Generating predictions** (also called *scoring*) applies that trained model to a population today, producing a score for every contact in it.

## What a prediction is

Each contact gets a score whose meaning depends on the model type:

| Model type | Score meaning |
| - | - |
| Propensity / churn / engagement (classification) | A **probability** from 0 to 1 — e.g. 0.82 = "82% likely to respond". |
| Expected amount / lifetime value (regression) | An **amount** — e.g. an expected gift of €45. |

## Prediction configurations

A **prediction configuration** is a saved recipe that you define once and re-run whenever you need fresh scores. It captures:

* **Which models to score with** — one or more trained models (e.g. DM propensity, expected amount).
* **Which population to score** — the contacts you want ranked: a specific list, a contact segment, filtered by response type, or the full active audience.
* **Response type filter** — optionally restrict scoring to contacts with a particular history (e.g. Payment, Pledge, or telemarketing respondents).

Running the configuration is called a **prediction execution**. Each execution is timestamped and stored so you can see exactly which run fed a given decision or export.

## Running a prediction

**1. Open your prediction configuration**

Choose an existing configuration or create a new one. Give it a clear name that identifies the model and the campaign or audience it targets.

**2. Choose the population to score**

Point the model at the contacts you want scored — typically a **list** (e.g. everyone eligible for an upcoming appeal). You can combine multiple lists or filter by segment. Only contacts present in your mapped data can be scored.

**3. Select which models to include**

A single configuration can run multiple models at once — for example, DM propensity and expected amount together — so all the scores you need for a decision are produced in one go.

**4. Set the scoring date (optional)**

By default Allyy scores as of **now**. You can also score as of a past date to back-test how the model would have ranked a previous campaign and compare against what actually happened.

**5. Run**

Allyy scores every contact in the population and stores the results. The execution status moves from **Running** to **Completed** (or **Failed** with a log if something goes wrong). Once complete the scores are ready to view, turn into a decision, or export.

> [!NOTE]
>
> **Screenshot to add** — a completed prediction run showing status, population size, and example scores.

## Reading the results

After a run completes you can inspect the scores in the platform. Each record shows:

* The **contact** (contact ID and any display fields you've mapped).
* The **score** — a probability between 0 and 1 for classification models, a currency amount for regression models.
* The **execution timestamp** — which run produced this score.

> [!NOTE]
>
> A score ranks contacts — it does not tell you *how many* to contact or *with what*. That is the job of a [decision](https://docs.allyy.io/predictions/decisions.md).

## Keeping predictions fresh

Scores reflect the data at the moment they were generated. Contact behaviour changes — recent donors, lapsed contacts, and new subscribers all shift the rankings. Re-run scoring before each campaign — or schedule it with a [workflow](https://docs.allyy.io/automation/workflows.md) — so you're always acting on current behaviour.

## Next

> [card] [Turn scores into action](https://docs.allyy.io/predictions/decisions.md)
>
> Combine and optimise scores into a ranked, segmented contact list.

> [card] [Check the model first](https://docs.allyy.io/dashboards-evaluation/overview.md)
>
> Use the evaluation dashboards to confirm the model ranks well before you rely on its scores.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.