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

# Write-backs overview

> Pull approved prices out of Elastly and apply them in your own system, then confirm what landed.

When you approve a price in Elastly, it needs to reach the system your team actually works in. If
Elastly writes to your provider directly, that happens on its own. If your system sits behind a
firewall or isn't one Elastly connects to, you pull the approved prices instead: claim a batch, apply
them on your side, and tell Elastly what landed.

<Info>
  Write-backs move approved prices **out** of Elastly. To push your catalog and sales history **in**,
  see [Ingest](/docs/api/ingest). To have the SDK run this loop for you, see
  [Build your own connector](/docs/sdks/build-a-connector).
</Info>

## Before you start

Connect **Custom ERP** or **Custom Storefront** under [Integrations](https://app.elastly.io/integrations), then create a **Connector** API
key under [Settings → API keys](https://app.elastly.io/settings) and send it as a bearer token on every request.

<Warning>
  **Treat the key like a password.** A Connector key can read approved prices for your whole workspace.
  Store it server-side, never in client code, and revoke it in the dashboard if it leaks.
</Warning>

## How claiming works

Claiming hands you a batch of approved prices and leases them to you for a short window. While a price
is leased, Elastly won't hand it to another caller and won't write it through a native connector, so a
price is never applied twice.

<Steps titleSize="h3">
  <Step title="Claim">
    [Claim](/docs/api/endpoints/claim-writebacks) a batch of approved prices. They're leased to you.
  </Step>

  <Step title="Apply">
    Write each price into your own system.
  </Step>

  <Step title="Acknowledge">
    [Acknowledge](/docs/api/endpoints/ack-writebacks) what landed. Successes are marked applied; failures
    return to the queue.
  </Step>
</Steps>

If you never acknowledge, the lease expires on its own and the price returns to the queue for a later
claim, so a crash mid-apply never loses a price. Each task identifies the product by both `productSku`
and `productExternalId`, so you can match it however your system keys products.

<Note>
  **Delivery is at-least-once, so make your write idempotent.** A task whose ack never arrived returns
  to the queue and reaches you again. Setting the same price twice should be harmless. When you
  acknowledge, `expired` counts tasks whose lease had already run out; those were returned to the
  queue, so claim again and reapply them.
</Note>