Data Freshness

Data freshness is how far your reports lag behind what happened: the time from an order, click or euro of ad spend until a report shows it, and how long a past day keeps changing after that. Most marketing data arrives in batches, not in real time.

Daniel Busch
Written by Daniel Busch · Chief of Staff · Updated

In short

  • Is: the delay from an order, click or euro of spend to your report, plus how long a past day keeps changing after it first shows
  • Platforms differ: Google Ads refreshes clicks and cost hourly, Meta refreshes every 15 minutes and keeps updating a day for up to 28 days, GA4 can take 24 to 48 hours to process data
  • Past days move because of refunds, late conversions, platforms correcting spend and recalculations after a settings change
  • adtribute updates today's orders, conversions and spend every 30 minutes, and your whole history every night at 00:30
  • Read a closed day the next morning and judge campaigns on completed periods: today's ROAS shows whether things run, not whether they pay

What data freshness means

Data freshness is the time between something happening, an order, a click, a euro of ad spend, and your report showing it. Its second half matters more: how long the number keeps changing after it first appears. A report updated a few minutes ago can still be provisional, because yesterday has not finished changing.

Almost all marketing reporting runs in batches. A tool pulls data from your shop and ad platforms on a schedule, every 15 or 30 minutes or once a night, and recalculates its tables; between runs, nothing moves. Real time exists in narrow views: GA4’s Realtime report shows activity within minutes, but only for a few dimensions and metrics. Every source keeps its own clock:

SourceFirst showsKeeps changing
Google AdsClicks, impressions and cost hourly; conversions later, and later still under models other than last clickMetrics can update one or more days after the event; invalid clicks are removed retroactively
Meta AdsRefreshed every 15 minutesUntil 28 days after a day was first reported
GA4, standard propertyIntraday data after 2 to 6 hours, a full day after about 12Processing can take 24 to 48 hours; some data arrives up to 7 days late
adtributeToday’s orders, conversions and spend every 30 minutesEvery night, for your whole history

Sources: Google Ads, Meta, GA4, adtribute.

In our customer issues, about 4% mention intraday numbers, delays or real-time data, and more than 6% mention recalculating or backfilling past days. The questions behind them are rarely about speed. They are about trust: why did yesterday change, and when is a number final?

Why yesterday’s numbers keep moving

Four things change a day after it has ended:

Refresh, recalculation and backfill

Customers often use the three words for the same thing. They are different operations:

  • A refresh loads new data for recent periods on a schedule. Today gets more complete; the past stays as it was.
  • A recalculation recomputes history from data already loaded, with the current settings, refunds and attribution. Past days change without any new raw data.
  • A backfill loads raw data for past periods that were missing: a new connection’s history, the days after a broken connection, older ad spend.

After a recalculation or a backfill, a number you exported earlier, into a spreadsheet, a board deck or a monthly report, no longer matches the tool. That explains most cases of “the old report said something else”. Date every export, and compare it only with an export of the same day.

Yesterday’s ROAS at 7:00 and three days later

A shop’s marketing lead reads Monday’s Meta ROAS on Tuesday at 7:00 and again on Friday at 7:00. The attribution model dates conversions by the day of the order, and revenue is net of VAT:

Monday, Meta AdsTuesday 7:00Friday 7:00What happened in between
Spend€2,000€2,040Meta corrected Monday’s delivery
Net revenue of Monday’s orders€5,000€4,750Two orders were refunded on Wednesday, counted on Monday
Late attribution€0+€90A post-purchase survey answer moved an order from Direct to Meta
ROAS on conversion date2.502.37(€4,750 + €90) ÷ €2,040
ROAS on click date1.902.60Monday’s clicks kept bringing orders until Thursday

Nothing broke, and still every number moved. On conversion date, Monday’s ROAS fell about 5%: the refunds and the spend correction outweighed the survey answer. On click date it rose by more than a third, from €3,800 of revenue to €5,300, because on Tuesday morning Monday’s clicks had only had hours to convert. On conversion date a closed day drifts down as refunds arrive; on click date it grows while its clicks keep converting.

How adtribute updates your data

  • Every 30 minutes: today’s orders, conversions and ad spend. Runs start at :05 and :35 from 01:05 to 23:35 in your organization’s time zone and usually finish within 25 minutes, so a new order usually shows within an hour. Spend and revenue update together, so they match each other.
  • Every night at 00:30: your whole history, with your current settings, usually within 90 minutes. Refunds and cancellations arrive then and count on the day of the order, up to 9 years back, along with older ad spend, Sellerboard data, Klaviyo engagement and your changes to settings, sheets and connected BigQuery tables.
  • Past days change through refunds, late attribution such as post-purchase survey answers, which moves a little revenue between channels, and ad platforms’ corrections to recent days’ spend. Your own changes can move more: a new attribution model shifts revenue between channels, a changed conversion definition can change the total.
  • New and reconnected sources fill recent days with the next updates and older days the next morning (backfill: how adtribute handles historical data).
  • No manual runs. The updates can’t be started by hand. If your data hasn’t changed for two hours, write to support.

When to read which day

Our view: today’s numbers tell you whether things run, not whether they pay. Spend counts the moment it happens, while the purchases today’s clicks bring arrive over the following days. On click date, today’s ROAS is too low by design; on conversion date, it pays today’s spend with orders from earlier clicks. Use today for checks: is spend pacing, is the tracking firing, has a campaign stopped delivering?

Read a closed day the next morning, after the nightly update, on the date the click date vs. conversion date entry’s view on when to use which gives for your question.

Freeze month-end numbers on purpose: export the month once most refunds are in, date the export, and don’t hold it against a live number later.

Where numbers settle slowly, predicting the settled value beats waiting: the PERGOLUX case study shows a brand that steers on a forecast of its net revenue instead of waiting for it.

FAQ about Data Freshness

Do the tables also refresh at 7:00 in the morning?

Yes. The 30-minute updates start at :05 and :35 from 01:05 to 23:35 in your organization’s time zone, so one starts at 07:05 and usually finishes within 25 minutes. The nightly update has usually finished long before, so yesterday’s refunds and your changes are in. Views of your data in BigQuery update with these runs; refreshing a view yourself changes nothing.

Is a change recalculated retroactively or only from the cut-off date?

Retroactively. Every night, your whole history updates with your current settings, so a new attribution model, channel mapping or conversion definition shows the next morning for all past dates. A number exported before the change no longer matches.

Why does yesterday’s Meta spend differ from Ads Manager?

Because yesterday isn’t final at Meta yet: Meta keeps updating a day’s numbers for up to 28 days, and adtribute picks up corrections to recent days with its next updates, so a gap on the last day or two usually closes by itself. A gap that stays isn’t a matter of timing; the ad spend entry’s section on why spend differs lists the other causes.

Can the data update faster during a sale?

No. The updates run every 30 minutes on their own, during a sale too. Watch spend and orders through the day, and judge the sale on click date once its clicks have had time to convert.