GA4 Reporting Limitations: Why Marketers Are Moving Their Analytics into Power BI
GA4 collects more than its predecessor ever did - and shows you less of it. Marketers who lived in Universal Analytics' reports now meet sampling notices in Explorations, rows that vanish below privacy thresholds, and a default that quietly deletes their history after fourteen months.
None of this means GA4 is broken. It means GA4's interface was never designed to be your reporting layer - and the businesses getting the most from it have stopped treating it as one. The pattern that works: GA4 collects, BigQuery stores, Power BI reports.
This guide covers each limitation honestly, the architecture that escapes them, and where the shortcut everyone tries first falls down.
Key Takeaways
GA4's interface limits are structural: sampling, thresholding, cardinality caps and 14-month retention are design choices, not bugs to wait out.
BigQuery export is the escape: unsampled, unthresholded raw events you own, with retention you control.
The prize is blending: behaviour joined to ads, CRM and revenue - the numbers GA4 alone can never compute.
What Are GA4's Reporting Limits, Specifically?
Five constraints recur in practice. Explorations sample once queries exceed quota thresholds - your funnel analysis becomes an estimate precisely when the date range gets interesting. Thresholding hides rows that might identify individuals, so small segments silently disappear from reports that look complete. Cardinality limits collapse high-variety dimensions into an "(other)" row that absorbs your long tail. Event-level retention defaults to two months and maxes at fourteen for standard properties - Explorations beyond that window simply have no data. And the Data API rations requests, which is exactly what third-party connectors hammer.
Each limit is tolerable alone; together they mean the interface answers small questions precisely and big questions approximately - the inverse of what decision-makers need.
Why Is the Power BI Connector Alone Not the Fix?
Because the connector consumes the same Data API as everything else - inheriting the quotas, the thresholding and the aggregate-only access, and adding refresh failures when the quota runs dry mid-morning. The connector moves GA4's numbers into Power BI; it does not move past GA4's limits.
The connector remains fine for what it is: lightweight scheduled pulls of headline aggregates for a marketing summary page. The signs you have outgrown it are familiar - refreshes failing with quota errors, numbers that shift between pulls as sampling varies, and the segment analysis you cannot build because the API will not expose the grain. At that point the architecture, not the connector settings, is the conversation.
What Does the BigQuery Export Architecture Look Like?
GA4's native BigQuery export streams the raw, unsampled event record - every event, every parameter - into a warehouse you own, with retention you decide. From there, the events are modelled into a star schema (sessions and conversions as facts; date, channel, campaign, page and device as dimensions) and served to Power BI as a governed semantic model. The interface limits simply do not apply: nothing samples, nothing thresholds, history accrues as long as you keep it.
The honest costs: the export is forward-only (it starts collecting from the day you enable it - which is the argument for enabling it today, whatever your reporting plans), the raw event format is nested and needs real transformation before it is analyst-friendly, and BigQuery storage and query costs are real though typically modest at mid-market traffic. This modelling step is the substance of BigQuery consulting engagements for marketing estates - the export is a switch; the value is in the shape.
What Does Blending Unlock That GA4 Never Could?
The numbers that live across systems: true cost-per-acquisition joining ad spend to closed revenue, lifetime value joining first-touch behaviour to years of CRM history, and channel profitability that survives contact with finance. GA4 sees behaviour; your CRM sees customers; your ledger sees money. Only a model that joins them sees the business.
In the warehouse, the joins are ordinary engineering: ad platform spend lands beside the GA4 events, CRM and e-commerce orders land beside both, and conformed dimensions (customer, campaign, date) stitch them into one queryable truth. The result ends the oldest argument in marketing reporting - whose number is right - because the dashboard's revenue is finance's revenue, attributed against behaviour finance never had. For agencies, the same architecture multiplies across clients: one modelled pattern, repeated per client, feeding standardised reporting at a marginal cost that manual assembly can never match.
Who Should Stay in the GA4 Interface - and Who Should Move?
Stay if your questions are operational and short-window: yesterday's traffic, this week's campaign check, content performance at a glance. The interface is free, current and adequate for the daily pulse. Move when decisions depend on segments, history or money: budget allocation, channel strategy, board reporting and anything that must reconcile with revenue.
The move is also staged rather than binary. Many businesses run the hybrid for a year - interface for the pulse, exported-and-modelled data for the decisions - while history accrues in BigQuery. Teams already reporting in Looker Studio can keep it as a presentation layer over the same modelled tables; the architecture is the asset, and Looker Studio consulting covers that path where the Google stack is the standard. The destination either way is the same: reporting served from data you own, shaped to questions you actually ask.
What Should a Marketing Leader Do This Quarter?
Three moves, in order. Enable the BigQuery export now - it is forward-only, and every week unexported is history you will never have. Audit which current reports are silently sampled or thresholded - the interface tells you, if asked, and the answer recalibrates trust. And pick the one blended number the business argues about most - usually true CPA or channel ROI - as the first modelled deliverable, because nothing funds the architecture like settling that argument.
The quarter ends with history accruing, limits mapped, and one number nobody disputes. The architecture grows from there at the pace the questions demand.
What Does the BigQuery Export Actually Cost?
Less than the meeting where people worry about it. The GA4-to-BigQuery export itself is free, and BigQuery's free tier - a terabyte of queries and ten gigabytes of storage monthly - covers a surprising share of mid-market sites entirely. Beyond it, storage costs cents per gigabyte and analytics workloads for a typical site run to tens of dollars a month, not thousands.
Two guardrails keep it that way. Partition-aware queries - always filtering on the event date - stop Power BI's refresh from scanning the whole history on every run. And a billing alert at a modest threshold converts the rare runaway query from a surprise invoice into a same-day email. Set both on day one and the cost conversation is over.
Treat the export start date as the urgent decision even if the Power BI build waits: BigQuery only receives data from the day you switch it on, and history before that date is gone for good. Flipping the export on costs nothing and banks the raw data while you plan - the one genuinely irreversible step in the whole architecture.
Collection Is Not Reporting
GA4 is a collection system wearing a reporting interface - excellent at the first job, structurally constrained at the second. The businesses frustrated with it are mostly asking the interface to be what it is not; the businesses thriving with it moved the reporting to where the raw data lives.
Export, model, blend. The limits stay in the interface; the decisions move to data you own.
