Ad platforms aren't storing your data out of goodwill (and never were)

Most brands haven't had to think about that because the retention windows were long enough that the question of ownership never came up. That's changing, and it's changing in one direction.

Starting June 1st, Google Ads is reducing how far back advertisers can pull granular reporting data. Daily, weekly, and hourly performance data will be available for 37 months. Monthly, quarterly, and annual data stays accessible for 11 years.

What makes this notable is the reversal. Google introduced the 11-year retention policy in November 2024. Eighteen months later, they pulled the granular layer back to roughly three years. Most trade press has covered the change as an operational concern: export your data before June. That's accurate, but it misses the bigger story.

The Pattern Across Platforms

Google isn't moving alone. The same direction is showing up everywhere.

Meta tightened its Ads Insights API in January 2026, capping unique counts at 13 months and frequency breakdowns at 6. View-through attribution windows at 7 and 28 days were removed entirely.

GA4 retains raw event-level data for 14 months maximum, and only 2 months by default. The aggregated reports in the UI go further back, but the underlying data that powers custom analysis is gone after 14 months unless you've exported it.

Amazon Ads retains granular campaign data via API for 60 to 90 days.

The window of historical data you can pull from any major ad platform keeps shrinking. Whether the driver is freeing up infrastructure for AI workloads, tightening privacy and compliance posture, or some combination, the result is the same. The trend isn't slowing down.

What Rented Data Actually Costs

The real cost isn't the lost data. It's the loss of the analytical baseline that makes long-term decisions credible.

Three specific capabilities erode when retention windows shrink.

Seasonality modeling gets thin. Two years of granular data can reveal a real pattern. One year is usually a coincidence. Brands need three to four years of clean granular history to model demand against weather cycles, participation trends, and category-specific seasonality. A rolling 37-month window means you're always one cycle away from losing the baseline.

Year-over-year analysis quietly degrades. When a new marketing leader walks in and asks how this quarter compares to two and three years ago, the answer shouldn't depend on whether someone happened to export the right report at the right time.

Brand investment gets harder to defend. MMMs and other long-horizon measurement approaches need multi-year granular data. The same goes for brand investment, which pays back over multi-year horizons. There's a real tension worth naming here: the platforms shortening retention are simultaneously pushing brands toward AI-modeled measurement and campaign types that depend on long historical inputs. Less data, more reliance on modeling. The math gets harder, not easier.

A significant part of our work involves building growth models for clients, and the gap between three to four years of clean historical data and less than two is the gap between a confident projection and an educated guess. The pandemic complicates things further. 2020 through 2022 were such an unusual mix of variables that, for many brands, they're effectively useless as a forecasting baseline. The projections that hold up are built on 2018, 2019, and the most recent two to three years, which is why retention well beyond the platforms' default windows matters more than most brands realize.

The Ownership Framework

Moving from rented to owned isn't really a tools question. It's a shift in how the organization treats marketing data as an asset.

Mature finance functions don't store the company's ledger inside a vendor platform. Marketing is moving in the same direction, and the platforms themselves are accelerating that move, whether they intended to or not.

Your Options for Owning Marketing Data

The right approach depends on the size of your data, the maturity of your stack, and the analytical sophistication you actually need. Most brands will land in one of two camps: warehouse-grade infrastructure or lightweight alternatives.

A Few Notes on Each

BigQuery by Google is the default recommendation for most brands we work with. The integration with Google Ads and GA4 is the deepest of any warehouse on the market. The serverless model means no infrastructure to maintain, and most of the marketing analytics tooling we use natively lives there. Hosting costs are negligible for typical brand workloads.

One thing worth flagging: connector tools like Supermetrics that push data into BigQuery inherit the same 37-month limit at the API source. The fix is a direct pipeline that captures and stores data the moment it's generated, before any retention window applies.

Snowflake is the stronger fit for brands that already have data science capabilities, cross-cloud strategies, or expect to share the warehouse with heavy non-marketing workloads like finance, supply chain, or product analytics.

Redshift by Amazon makes sense when AWS is already the center of gravity. The integration with S3, Lambda, and the rest of the AWS stack outweighs the higher operational overhead.

Connector plus archive isn't a long-term solution, but it's a viable holdover when budgets or tech stacks are in flux. Think of it as renting multiple storage units scattered around town. It preserves the data and solves the immediate retention problem, but loses the cross-platform analytical capability a warehouse provides.

Why This Matters Now

Owning your marketing data is less about technology and more about how much value your organization places on its own performance history.

Every retention change from here on will make this decision look more obvious in hindsight. The question is whether you make it before or after the data you need is no longer available.

If you want to think through where your stack stands, we're easy to reach.