> For the complete documentation index, see [llms.txt](https://docs.quanti.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.quanti.io/connect/fundamental-concepts/lookback-window.md).

# Lookback window

The **lookback window** defines the **time period** that a connector will retrieve data for during each synchronization, relative to the reference date.

{% hint style="info" %}
Why use a lookback window?

Data on source platforms (Google Ads, Meta, etc.) is not always immediately finalized. Several factors can cause retroactive changes:

* **Data processing delays**: metrics may be updated hours or days after the actual event
* **Attribution adjustments**: conversions can be attributed retroactively to earlier clicks
* **Platform corrections**: source platforms may correct or recalculate historical data
* **Time zone differences**: events near midnight may be reassigned to the correct date
  {% endhint %}

{% stepper %}
{% step %}

### How it works

Instead of only synchronizing data for the reference date, the connector retrieves data for a **period including past days** (the lookback window) to capture retroactive updates.
{% endstep %}

{% step %}

### Example with a 7-day lookback window

Today's date: 2025-01-22\
Reference date: 2025-01-15\
Lookback window: 7 days

Data retrieved for dates:

* 2025-01-15 (reference date)
* 2025-01-14
* 2025-01-13
* 2025-01-12
* 2025-01-11
* 2025-01-10
* 2025-01-09 (reference date - 6 days)

All data for these 7 days is retrieved and synchronized, ensuring any retroactive updates are captured.
{% endstep %}

{% step %}

### Practical example

Initial sync on 01/16:

```
| date       | campaign_id | clicks |
| 2025-01-15 | camp_123    | 100    |
```

Sync on 01/20 with 7-day lookback:

```
| date       | campaign_id | clicks |
| 2025-01-15 | camp_123    | 105    | ← updated (5 additional clicks attributed retroactively)
```

{% endstep %}

{% step %}

### Impact by insertion method

The lookback window interacts differently with each insertion method:

* **REPLACE Mode**: Previous data within the lookback window is deleted and replaced with fresh data from the source.
* **UPSERT Mode**: Rows are updated with new values, preserving history if configured.
* **INSERT Mode**: May create duplicates if the same data is retrieved multiple times (not recommended with lookback).
  {% endstep %}

{% step %}

### Typical lookback window values

| Platform Type                            | Common Lookback | Reason                                 |
| ---------------------------------------- | --------------- | -------------------------------------- |
| Advertising platforms (Google Ads, Meta) | 3-7 days        | Attribution windows, conversion delays |
| Analytics platforms (GA4)                | 1-3 days        | Processing delays, session completion  |
| CRM/Sales (Salesforce, HubSpot)          | 1-2 days        | Data entry delays, batch updates       |
| E-commerce (Shopify, WooCommerce)        | 1-2 days        | Order updates, refunds                 |
| {% endstep %}                            |                 |                                        |

{% step %}

### Trade-offs

<table><thead><tr><th width="160.20703125">Critère</th><th>Longer Lookback Window</th><th>Shorter Lookback Window</th></tr></thead><tbody><tr><td><strong>Data completeness</strong></td><td>✅ More complete and accurate data</td><td>⚠️ Risk of missing retroactive updates</td></tr><tr><td><strong>Retroactive changes</strong></td><td>✅ Captures all platform corrections</td><td>❌ May miss late attribution or corrections</td></tr><tr><td><strong>Synchronization time</strong></td><td>❌ Longer processing time</td><td>✅ Faster synchronizations</td></tr><tr><td><strong>Data warehouse costs</strong></td><td>❌ Higher (more partitions scanned/written)</td><td>✅ Lower (fewer partitions affected)</td></tr><tr><td><strong>Data consistency</strong></td><td>✅ More reliable for reporting</td><td>⚠️ Potential inconsistencies over time</td></tr><tr><td><strong>Use case</strong></td><td>Critical metrics, attribution analysis</td><td>High-volume data, stable metrics</td></tr></tbody></table>
{% endstep %}

{% step %}

### Best practices

* Configure the lookback window based on the **platform's attribution window**.
* Monitor data stability: if values change significantly after several days, increase the lookback.
* For critical metrics, prefer a longer lookback window to ensure data accuracy.
* For high-volume tables with stable data, a shorter lookback may be sufficient.
  {% endstep %}

{% step %}

### Configuration in Quanti

The lookback window is configured at the connector level and can be adjusted based on your data quality requirements and platform-specific behavior.
{% endstep %}
{% endstepper %}

### reverse\_lookback — Lookback on a reverse push

On a reverse connector, the lookback sets how many days of rows are re-sent to the destination platform at every run, which is not the same thing as refreshing days in your warehouse.

**Where to find it**

In the Sync tab of the reverse connector, the same selector as on a pull connector. It applies to the date column declared as the push source date field.

**Why it matters**

Copying the value used on pull connectors, where 7 days is ordinary. A push in insert mode keeps no snapshot of what it already sent: the whole window leaves again at every run, and it is the destination that is expected to discard the repeats. Meta only does so for events received within 48 hours of the first event carrying a given event\_id, so a window of 3 days or more sends the oldest day back after deduplication has expired and the same purchase is counted twice, inflating the conversions the platform reports and the bids it places on them. Two days covers two daily runs, which is the practical ceiling. A longer window does not make a missed day recoverable either: fix the hour the sync runs instead.
