A/B Testing with BigQuery — Raw Data, Exact Numbers, No Estimates

When you run an A/B test and measure results through GA4's standard reporting, you're not seeing exact numbers — you're seeing HyperLogLog++ estimates. For dashboards and general reporting, these estimates are accurate enough. But for A/B tests where a 2-3% conversion difference can mean thousands in revenue, estimation can be misleading.

Varify.io solves this with a native BigQuery integration that bypasses GA4's estimation layer entirely. Varify queries the raw GA4 events stored in BigQuery, calculates exact numbers, and adds product-level filtering, outlier smoothing, and duplicate user event exclusion — all without writing a single line of SQL. For teams that need A/B test results they can trust, this is the most precise evaluation method available.

Why GA4 estimates aren't enough for A/B testing

HyperLogLog++ is Google's probabilistic counting algorithm. It estimates unique counts instead of computing them exactly — trading accuracy for speed and storage efficiency. For reports showing "how many users visited this page" the estimation is fine (typically within 1-2%).

For A/B testing, the tolerance is different. If Variant B shows 3.2% more conversions than Control and you're deciding whether to ship it, a 1-2% estimation error in the underlying data can swing the result. You might ship a losing variant or kill a winning one.

BigQuery stores the raw events — every click, every conversion, every pageview as individual rows. When Varify evaluates through BigQuery, it counts exact conversions per variant. No estimation, no approximation.

BigQuery integration depth compared

Source: Claude Research, May 1, 2026

GrowthBook also offers native BigQuery access, but requires SQL knowledge and manual configuration. Varify automates the queries — no SQL needed. For the full analytics comparison, see our analytics integration guide.

What Varify's BigQuery integration includes

Real-time data collection

BigQuery receives GA4 events via streaming export. Varify reads them in near-real-time — no daily batch delays.

Product-level filtering

For e-commerce tests: evaluate specific product categories or SKUs instead of site-wide metrics. Test a product page redesign and measure impact on that product's revenue, not total revenue.

Outlier smoothing

Automatically detects and handles statistical outliers (extreme order values, bot traffic) that can skew A/B test results. Essential for e-commerce where one large order can distort conversion metrics.

Duplicate user event exclusion

GA4 can fire the same event multiple times per session. Varify's BigQuery queries deduplicate automatically, ensuring each user-event pair is counted once.

All this without writing SQL. Connect your GA4 BigQuery export, and Varify handles the queries. See pricing — BigQuery integration is available on all plans.

How to set up Varify with BigQuery

1. Enable GA4 BigQuery export in your GA4 property settings. Google provides a free daily export; streaming export is available for BigQuery users.
2. Connect BigQuery in Varify via the Settings panel. Authorize Varify to read your BigQuery dataset (read-only access).
3. Select BigQuery as evaluation source for your experiments. Varify automatically queries the correct tables and calculates results.
4. Run your first test and see exact numbers in Varify's reporting — same data as in BigQuery, presented with daily progress graphs and statistical significance.

Setup takes about 15 minutes if your GA4 BigQuery export is already active. If not, add 30 minutes for the initial export configuration.

Niko Kerter

CRO Expert at Varify.io