Experimentation, plumbed into the warehouse
Five links on A/B testing that ties back to revenue.
1. Feature flags vs experiments — they overlap but aren't the same, here's the line. Source: a PostHog explainer.
2. Statrigor without a stats degree — sequential testing and why peeking ruins results. Source: an experimentation blog.
3. Warehouse-native experimentation — computing lift in SQL instead of a tool's black box. Source: Eppo blog.
4. The metric tree — connecting a button test to ARR. Source: Reforge.
5. Why most A/B tests are underpowered — the sample-size reality check. Source: a CRO thread.
That's the stack for this week. Forward to a teammate.
Five links on A/B testing that ties back to revenue.
1. Feature flags vs experiments — they overlap but aren't the same, here's the line. Source: a PostHog explainer.
2. Statrigor without a stats degree — sequential testing and why peeking ruins results. Source: an experimentation blog.
3. Warehouse-native experimentation — computing lift in SQL instead of a tool's black box. Source: Eppo blog.
4. The metric tree — connecting a button test to ARR. Source: Reforge.
5. Why most A/B tests are underpowered — the sample-size reality check. Source: a CRO thread.
That's the stack for this week. Forward to a teammate.
Consent plumbing, done properly
Most teams bolt on a banner and call it compliance. These go a level deeper on the wiring.
— 1. Google Consent Mode v2 advanced vs basic — Simo Ahava's breakdown of what actually fires before consent in each mode; the modeling caveat alone saves a reporting mess. Source: simoahava.com.
— 2. OneTrust vs Cookiebot vs Osano field notes — a practitioner thread comparing geo-rule granularity, not feature checklists. Source: r/analytics.
— 3. Server-side consent forwarding — passing consent signals through your sGTM container so downstream tags respect it. Source: Stape blog.
— 4. IAB TCF string anatomy — what those base64 consent strings encode, decoded byte by byte. Source: IAB Tech Lab.
Skip if you only serve US traffic. Essential the day you take a single EU click.
That's the stack for this week. Forward to a teammate.
Most teams bolt on a banner and call it compliance. These go a level deeper on the wiring.
— 1. Google Consent Mode v2 advanced vs basic — Simo Ahava's breakdown of what actually fires before consent in each mode; the modeling caveat alone saves a reporting mess. Source: simoahava.com.
— 2. OneTrust vs Cookiebot vs Osano field notes — a practitioner thread comparing geo-rule granularity, not feature checklists. Source: r/analytics.
— 3. Server-side consent forwarding — passing consent signals through your sGTM container so downstream tags respect it. Source: Stape blog.
— 4. IAB TCF string anatomy — what those base64 consent strings encode, decoded byte by byte. Source: IAB Tech Lab.
Skip if you only serve US traffic. Essential the day you take a single EU click.
That's the stack for this week. Forward to a teammate.
Switch attribution models without starting a turf war
The change-management checklist, not the math one.
1 — Run the new model in shadow alongside the old one for a full sales cycle.
Why: showing both numbers side by side disarms the 'you're stealing my credit' fight.
2 — Document which channels gain and lose credit before the meeting.
Why: the channel owners who lose will find out anyway; better they hear it from you.
3 — Freeze the model during the comparison window.
Why: tweaking weights mid-test makes the deltas un-interpretable.
4 — Pick a switch date that aligns to a reporting boundary, never mid-quarter.
Why: a number that changes definition mid-quarter is a forecast nobody can defend.
That's the stack for this week. Forward to a teammate.
—
Рядом обитают: @MarTechNotesHub (crm)
The change-management checklist, not the math one.
1 — Run the new model in shadow alongside the old one for a full sales cycle.
Why: showing both numbers side by side disarms the 'you're stealing my credit' fight.
2 — Document which channels gain and lose credit before the meeting.
Why: the channel owners who lose will find out anyway; better they hear it from you.
3 — Freeze the model during the comparison window.
Why: tweaking weights mid-test makes the deltas un-interpretable.
4 — Pick a switch date that aligns to a reporting boundary, never mid-quarter.
Why: a number that changes definition mid-quarter is a forecast nobody can defend.
That's the stack for this week. Forward to a teammate.
—
Рядом обитают: @MarTechNotesHub (crm)