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

# Recommendations, simulation & forecasts

> The AI decision aids that propose quantities and discounts and project future demand.

The intelligence layer includes AI decision aids that help you build better plans. They
**propose** — you stay in control and review before committing.

## Decision vectors & action vectors

Solya's decision layer works in two steps. First it computes **decision vectors**: per
variant-and-shop scores such as restock urgency, stockout risk, overstock risk,
surplus/deficit, and transfer urgency — *how risky or urgent is this position?* Then it
**resolves** those scores into **action vectors**: a concrete recommendation — *so what
should we do?* — a quantity to reorder, units to transfer, or a discount percentage, each
with a confidence and an explanation.

```mermaid theme={null}
flowchart LR
  S["Decision vector<br/>scores (urgency, risk…)"] --> A["Action vector<br/>recommended qty / discount"]
  A --> R["You review<br/>accept · adjust · discard"]
```

These recommendations drive plan building, the alert workspace, the task inbox, and the
score-driven [workflow](/en/automation/workflows) strategies — but they always **propose**;
you decide.

<Card title="How decision & action vectors are computed" icon="diagram-project" href="/en/developers/decision-layer/overview">
  The full technical reference — inputs, per-domain scoring formulas and weights, resolution,
  table schemas, and how the app consumes them.
</Card>

## Plan simulation

**Simulation** runs a plan against the decision layer **without persisting anything**. You
pick an entity grain (brand / product / variant), an action family (restock / rebalance /
markdown), a strategy, and optionally a ruleset; Solya returns the items it would add and
with what quantities or discounts — along with the decision context and which rules
applied. You can then add them all, cherry-pick, or discard.

## Sales forecasts

**Forecasts** predict future sales per variant / size / shop over a horizon, with
confidence bounds. They feed risk assessment and stock planning, and are surfaced in
analytics and as a chart type on dashboards.

## How to use them

<Steps>
  <Step title="Review recommendations">
    Recommended quantities/discounts appear when building plans and in the AI hub.
  </Step>

  <Step title="Simulate before committing">
    Use simulation to preview a strategy's output on a scope, then accept what you want.
  </Step>

  <Step title="Lean on forecasts">
    Consult forecasts to size restock and pre-season plans.
  </Step>
</Steps>

<Note>
  Recommendations respect your rules: the decision layer reads the org's ruleset (and
  margin floors, budgets, etc.) when shaping its proposals. Items added from a
  recommendation are tagged with their attribution in the
  [activity log](/en/inventory-plans/lifecycle#activity--audit-trail).
</Note>