Forrester has been researching this for years: how many companies actually make decisions based on data, versus just collecting it. The number is remarkably stable: around 6%.

Yes, 6%. Not 60. Six.

And yet everyone has dashboards. GA4 is connected. Ad accounts are pulled in. CRM is maintained. Reports arrive on schedule. The picture looks complete – there's a sense of control. But actual decisions? Not really.

Let me explain where it breaks down.

Typical Friday: the founder asks "what are we doing with the budget next week?" You open the report. CPA is up 18%. ROAS is down. Traffic is average. The funnel looks fine. You stare at it, then say "I'll figure it out by Monday."

What happens next? Right – you go to ChatGPT or Claude, because that's what everyone's shouting about. You paste in some numbers, type "where are we losing money," and get back a confident paragraph about seasonality, auction dynamics, and the importance of creative testing. Looks great. Completely useless.

I come into projects as a fractional Head of Marketing Analytics. And I keep seeing the same outcome: campaigns with zero sales running for months because they looked "acceptable" in the dashboard. Money gone. Nobody made the call to turn them off because "we should watch it a bit longer."

The problem is the questions.

Take "Why did our CPA go up?" There's no time period, no channel, no threshold – up by how much, and does it even matter? AI gives you text. An analyst gives you text. What to actually do – still unclear.

Now compare that to: "Which Facebook campaigns in the last 14 days spent more than $300, showed CPA more than 25% above target, and generated zero purchases?"

That's a question your analytics system, Claude, or ChatGPT can actually work with – filter, calculate, return a list of candidates to turn off. Not because they got smarter, but because you gave them a framework with a clear criterion.

But this, unfortunately, takes actual work. Such is life.

How to build this in practice:

  1. Define the decision – what specifically needs to be made. "Turn off underperforming campaigns by Friday" beats "figure out what's going on with ads."
  2. Set the criterion – a concrete number: CPA above X, spent more than Y, zero sales in N days.
  3. Go to the data with that question – write down your hypotheses.
  4. If possible, give an AI assistant access to the data. Ask it to challenge your hypotheses and find stronger ones.
  5. Make the decision – you have a list of hypotheses in hand, the rest is just pulling the trigger.

That 6% from Forrester who actually work with analytics – they're not smarter or better resourced. They just know what question to bring to the data.

Want all my posts in one place? I put 350+ articles on GA4, BigQuery, attribution, and metrics into one searchable library.

Go here to explore it for FREE.