In short
- Is: loading data for past dates that a tool or table doesn't have yet, such as a new connection's history or the days after an outage
- The source sets the limit: since June 2026, Google Ads keeps daily data for 37 months; Meta's API serves total values for 37 months; a Shopify app sees 60 days of orders without read_all_orders
- Not a recalculation: a changed attribution model or conversion definition recomputes past days without loading anything new
- Tracking can't be backfilled: touchpoints exist from the day tracking started, so older spend has no attributed revenue
- In adtribute, a new connection fills recent days with the next updates and older days the next morning; older ad spend loads on request
What a backfill is
A backfill loads data for dates that have already passed. Pipelines normally move forward; a backfill fills the gaps behind them. Google’s BigQuery Data Transfer Service uses the word in exactly this sense: a run “to retrieve data outside of your scheduled data transfer”, as far back as the source’s retention policy allows (Google Cloud).
Three situations call for one:
- A new source. Connecting an ad platform or a shop loads its history once, the historical sync, before the regular updates take over.
- A gap. A connection was broken for some days, or a source sent nothing. The backfill loads the missing days.
- A fix. Costs were missing, a field was mapped wrong or an ad account was left out. Loading the corrected data rewrites past days.
Requests like these come up in about one in fifteen of the customer issues we analysed, usually after a fix, a cost change or a new connection. In German threads, the word is more often “rückwirkend”: retroactive.
A backfill has two neighbours that also change numbers you have seen before: a refresh updates recent days on a schedule (data freshness), and a recalculation recomputes past days from data already loaded, for example after you change your attribution model.
How far back sources go
A backfill can’t load more than the source still holds:
| Source | History available | Reference |
|---|---|---|
| Google Ads | Daily data for 37 months, monthly data for 11 years, since 1 June 2026 | Google Ads Help |
| Meta Ads | Total values for up to 37 months; since 12 January 2026, breakdowns of unique counts and hourly breakdowns for 13 months | Meta |
| Shopify | All orders with the scope read_all_orders, otherwise the last 60 days | Shopify |
| Microsoft Ads in adtribute | Up to 180 days, even when a longer time frame is selected | docs: Microsoft Ads |
| OpenAI Ads in adtribute | The past 365 days, nothing before 15 April 2026 | docs: OpenAI Ads |
| Sellerboard in adtribute | 30 days back, then daily syncs | docs: Sellerboard |
| Webhook-based sources such as Copecart | None: data starts on the day you connect | docs: Copecart |
| Your website tracking | From the day tracking started | docs: Attribution models in adtribute |
The last row is the one no backfill fixes. Spend and orders from before your tracking started can be loaded, but the clicks and visits were never recorded: “Spend from before your tracking started has no attributed revenue” (docs: When your data updates).
Load what you need before a source forgets it: a daily Google Ads history older than 37 months exists only where someone stored it.
Why past numbers change
Past days move for five reasons, and only one of them is a backfill: missing days arriving after a reconnect or a request for older spend. The other four are in the data freshness entry’s section on why yesterday’s numbers keep moving: refunds, late conversions, corrections Meta and Google make to past days, and recalculations.
“Why did historical numbers change?” comes up in about one in thirty-five of the customer issues that contain a question, often after a migration or a settings change.
An example. A brand’s Meta connection breaks on Friday 10 April and is reconnected on Monday 20 April, while Meta keeps spending €1,000 a day. The same week, the brand asks support for its Meta spend of 2025, from before its tracking started on 1 January 2026:
| What loads | Shows in your reports |
|---|---|
| Spend of 13 to 19 April, the 7 days before the reconnect: €7,000 | Within a few hours of the reconnect |
| Spend of 10 to 12 April: €3,000 | The next morning, 21 April |
| Spend of 2025, before tracking started | The morning after support loads it, with no attributed revenue |
Because the break lasted more than 3 days, the first two rows can each take up to a day longer. Read April just before the reconnect and it misses €10,000 of Meta spend; two days after it, nothing is missing.
How to tell a backfill from a real change
| What you see | Likely cause | Check |
|---|---|---|
| Spend rose for past days, revenue didn’t | A backfill of missing days | Reconnects, or older spend loaded on request |
| Net revenue fell on past order days | Refunds, counted on the day of the order | Refunds in your shop since you last looked |
| A channel gained what another lost, the total stayed | A recalculation with a new or changed model | Changes under Definitions › Attribution Models |
| The total moved without new orders or refunds | A changed conversion definition, or a model that credits every touchpoint in full | Changes to the conversion; Full Impact or Channel Impact in use |
| Recent days grew on click date only | Late conversions | On conversion date, those days stay the same |
| Yesterday’s spend moved a little | The ad platform corrected it | The ad account, same day |
The rules behind it: spend is what the platforms charged, so no model moves it, and refunds count on the day of the order (docs: Why your numbers differ from other tools). Attribution changes move revenue between channels, a changed conversion definition can change the total (docs: When your data updates), and models that credit touchpoints in full, such as Full Impact, show higher totals (docs: Attribution models in adtribute). Date every export and note each model change, reconnect and backfill.
How adtribute handles historical data
- New connections. A new connection fills recent days with the next updates and older days the next morning (docs: When your data updates). Google Ads loads your whole history first. For TikTok, Pinterest, LinkedIn and Taboola Ads, the marketing attribution platform adtribute presets the historical sync time frame to all time.
- Shopify. Your full order history comes in within a day. If only the last 60 days show, the app lacks read_all_orders: add the scope, approve it in each store and tell us, and we load your older orders (docs: Shopify).
- Older ad spend. On request to support, as far back as the platform provides.
- After a reconnect. Your existing data stays. Missing days from the last week come back within a few hours, older days the next morning, and after a break of more than 3 days up to a day later (docs: Reconnect a broken connection).
- Recalculations. A changed setting reaches your whole history in the nightly update (data freshness: how adtribute updates your data). To keep a model’s old numbers, change a duplicate (docs: Attribution models in adtribute).
- Clicks without parameters. A corrected Google Ads tracking template only helps new clicks; older ones stay under Wrong adtribute Matching Parameters (docs: Google Ads). Adding a missing ad account to a Meta connection is the exception: its earlier clicks move to their ads the next morning (docs: Meta Ads).
- History stays when you archive or delete a campaign, as long as the connection stays in adtribute.
Our view: backfill spend, judge channels from the tracking start
Our view: load as much spend and order history as your sources give you, and read attribution only from the day your tracking started, plus a typical journey length. Before that day, spend has no attributed revenue. Customers whose journeys began earlier show up with their first tracked click, often an email or a direct visit, so a channel can look weak in your first weeks because its journeys started before you could see them. The spend history is still worth loading: for budgets, seasonality and the MER of past years, which needs no attribution at all.
FAQ about Backfills
What is a backfill?
Loading data for past dates that a tool or table is missing, such as a new connection’s history, the days a broken connection didn’t load, or older ad spend. Unlike a recalculation, it adds data that wasn’t there before.
How far back can ad spend be loaded?
As far back as the platform provides. Since June 2026, Google Ads keeps daily data for 37 months and monthly data for 11 years, and Meta’s API serves total values for up to 37 months. In adtribute, older spend loads on request; spend from before your tracking started has no attributed revenue.
Can you backfill tracking data from before we installed the script?
No. We get asked in calls how far back we can cover journeys, and the answer is the day tracking started. Clicks and visits from before were never recorded, so no system can load them later. Orders and spend from that time can be loaded; they stay unattributed.
Does a reconnect bring back the missing days?
Yes. In adtribute, the data you already had stays, and the missing days load by themselves: the last week within a few hours, older days the next morning, up to a day later after long breaks. If data is still missing after two days, write to support.