Pinned Tweet
Having issues with the 1M event limit in GA4? Look into Parallel Tracking! #measure #ga4 buff.ly/3P4DXa8
2
340
Reflective Data retweeted
How users get tracked by Google is...interesting. Everyone has their level of risk. CB now lets you ensure that no matter what nothing gets sent to GA4 until a user provides consent (default) or you can use their Advanced setup which explicitly tells GA4 the user has not yet consent.
What's the difference?
In Advanced mode, this means page views still send a ping to GA4 but most everything about the user is stripped away. Google then uses the data to "model" (in reality, make an educated guess), about the user and fills in the gaps.
This new setting just gives you greater control to determine how you want tracking to work on your site.
nitter.cf/ConversionWP/status/21…
The cheapest fix in analytics: write down what "conversion" means.
One sentence. One owner. One place everyone can find it.
Most reporting disagreements we're asked to untangle start with a metric nobody defined.
reflectivedata.com/services/…
15
A failing dbt test at 6am is cheaper than a CMO asking why yesterday's revenue doubled.
The 4 tests worth having on every model:
› unique
› not_null
› accepted_values
› relationships
Start there. Add the clever ones later.
reflectivedata.com/platform/…
19
3/ Does the AI read from modelled tables, or from raw exports? Raw tables mean it guesses the business logic every time. Can it show how it calculated each number?
1
5
4/ If any answer is no, fix that layer first. A semantic layer gives AI tools, dashboards and notebooks the same definitions. How it works: reflectivedata.com/platform/…
7
Black Friday is under 3 weeks away.
If your GA4 property gets anywhere near 1M events on a normal day, check last year's busiest day before this year's arrives.
Going over the daily BigQuery export limit regularly gets the export paused.
reflectivedata.com/how-to-av…
26
Your dashboard has 40 charts.
Ask the people who open it which 3 they actually use. Then build those 3 really well, with definitions everyone agreed on.
Dashboards people actually open are usually the short ones.
reflectivedata.com/services/…
42
Reloading a whole database table every night is the slow, expensive way to stay in sync.
Change data capture copies only what changed, as it changes. Smaller loads, fresher data, less strain on the source.
Your pipeline tool should do this by default.
reflectivedata.com/data-pipe…
1
18
3/ Wall 2: GA360's price. It lifts the cap, but it's an enterprise contract. Hard to justify when one limit is the main reason. Wall 3: sampled and modelled data in the GA4 interface.
1
11
4/ The fix for all three: send a second copy of every hit to a pipeline you control, in parallel with GA4. Complete, unsampled, in your warehouse. Full write-up: reflectivedata.com/ga4-pain-…
9
"Ask your data in plain English" works today.
On one condition: the AI doesn't invent the business logic. It picks "revenue" from definitions your team already agreed on, and shows how each number was calculated.
Without that, it's a confident guess.
reflectivedata.com/platform/…
23
Looker, Tableau or Power BI?
Honest answer: the tool matters less than what sits under it. If "active customer" means something different in each workbook, all three will show the wrong number beautifully.
Agree the definitions first.
reflectivedata.com/services/…
1
2
155
GA4's native export goes to one place: BigQuery.
On Snowflake, Redshift or S3? The usual route is GA4 → BigQuery → your warehouse: two platforms to pay for, a 1M events/day cap, and a load that lands when it lands.
There's a shorter route.
reflectivedata.com/how-to-ex…
46
Warehouse bills rarely grow because of storage. They grow because every dashboard scans raw event tables each time it's opened.
Model once. Aggregate once. Point reports at the small tables.
One client cut marketing warehouse costs ~75% this way.
reflectivedata.com/case-stud…
32
3/ Shortcut 2: point it at raw warehouse exports. Complete data, but large, nested and full of duplicates, bots and test traffic. The model guesses the business logic on every query. Slow, expensive, inconsistent.
1
1
4
4/ What works: business logic in dbt, metrics in a semantic layer, and the AI reads from that. Same question, same number, every time. How we do it: reflectivedata.com/reflectiv…
6


