Watch an investigation
When a number moves, is it real — or is it your tracking?
That question, answered beat by beat, in the Slack channel the team already uses.
A reconstruction of a real Steward investigation from a live deployment. Names, numbers, and details changed.
A number moved. It was not demand. It was a consent setting over-blocking tags. Six days, one thread.
1Most days, this is the whole experience: open Slack, see you’re good, move on.
✅ Daily demo-request read: all good.
Marketo: 24 fills (normal range 20–29). GA4 capture: 64% of Marketo, inside the expected band (55–70%).
Nothing needs you today.
2It doesn’t say “something moved.” It says which system disagrees, and which side to suspect.
🚨 Variance flag: GA4 demo capture fell out of band.
Yesterday: Marketo 25 fills, normal. GA4 capture: 34% against the 55–70% band.
Marketo is steady, so this reads as a measurement-side gap, not a demand move. Digging into when it started. 🧵
3It correlates what it can see, then asks for the one thing it can’t.
This isn’t a smooth slide. Capture held near 63%, stepped down to ~35% between Tuesday and Wednesday, and has been flat since. Something shipped, so I checked.
A GTM container version published Wednesday: the nav release. I diffed it against Tuesday’s version. Nothing in it touches the demo tags.
The shape points somewhere else. All three client-side legs fell together, in proportion. A single broken firing path pulls one leg down and leaves the rest whole. These moved in lockstep, while Marketo held on the server side. That’s the signature of a consent gate, not a broken tag.
Your consent manager is the one system I can’t read. Did a consent setting change Wednesday too?
4One question, one answer. Steward keeps it.
Yes: auto-blocking switched on sitewide Wednesday, same window as the nav release.
That’s it, then. The auto-blocker is supposed to hold tags until a visitor consents.
Best-evidenced cause: an over-block. It’s holding them for visitors who did consent, too.
Your “missing” demos aren’t missing people; they’re being dropped after the form is filled.
5Your team makes the call and defends it with evidence. Steward never publishes changes.
So demo volume didn’t actually drop?
Demand looks intact: Marketo fills never left their normal range. The drop is measurement: the auto-blocker is over-blocking, holding tags for visitors who already consented.
Fix that setting and GA4 capture should come back into your 55–70% band. The consent manager itself is doing its job; keep it.
OK. We fix the over-block and keep the stricter consent setup. Shipping it this week.
6Every fix creates a new thing to track. The watching doesn’t stop when the ticket closes.
Friday read: thank-you-page leg down, datalayer leg whole. This one is not consent.
Diverging legs point at a specific firing path, and it lines up with Thursday’s site release. Flagging for the web team.
Zero GA4 demos today against 3 Marketo fills. At that volume a zero is within normal Sunday noise, not a new anomaly. No action.
Your over-block fix landed: capture back inside the 55–70% band.
Marketo 26 fills, GA4 capture 62%: a normal day again. Back to the daily read; nothing needs you today.
Watch. Investigate. Brief your team. You make the call.
Steward runs in your Slack, reads across your stack, and does the digging your team is doing by hand, so when a number moves, you know if it's real before anyone asks you to explain it.
Skip the questions and book the working session. 30 minutes, straight to a calendar. It costs nothing.
Already deployed and delivering value. Steward runs live today, internally and with clients, watching real stacks in the team's own Slack.
Filtering hard on AI claims? Read the objections we hear most.
Sending this to someone? This link goes straight to the whole investigation.