One pipeline, one version of what happened
- Sources
utm · referrerTracked - Events
one taxonomyStructured - Attribution
first touchMapped - CRM
CRM recordConnected - Automation
approval gateActive - Reporting
one reportCoordinated
07 / Capabilities
Analytics & Automation
One event taxonomy, consent handled properly, and the repetitive operational work moved into software.
Part of — 04 Scale
What it is
Analytics and automation work gives a business one trustworthy account of what happened, and removes the manual steps between that account and the action it should trigger. It spans measurement design, consent, attribution and the workflow automation underneath — including AI-assisted steps where a human still approves the output.
The mechanism
From the first visit to the report everyone accepts.
Sources converge on one event taxonomy; a lead is counted once, when the server accepts it; attribution travels with it into the CRM; automation drafts, and a person approves anything that leaves the building.
A closed set, written down once
- Intent
cta_clickTracked - Form started
contact_form_startTracked - Lead, server-confirmed
generate_leadVerified - Evidence read
project_viewObservable - Consent
consent_decisionConsent-gated
Software drafts, a person decides
generate_leadEventTrackedrouting ruleRuleStructureddraftDraftReadyapproval gateApprovalOwned
Who it is for
- Companies where three dashboards give three different numbers
- Teams spending advertising budget without knowing which enquiry it produced
- Operations that re-key the same data between systems every week
What usually breaks
- Events were added ad hoc over years, so nothing can be compared to anything.
- Consent is a banner that changes nothing, which is both a legal and a data problem.
- A lead arrives, and its origin is lost before it reaches the CRM.
What we change
01
Design the taxonomy before the tags
One naming model, one definition per event, written down. Tag containers implement it; they do not invent it.
02
Consent first, genuinely
Nothing non-essential loads before a choice is made, refusal is as easy as acceptance, and the consent state actually drives what runs.
03
Automate the boring, review the consequential
Repetitive handling is automated end to end. Anything that speaks to a customer or moves money stays behind human approval.
Deliverables
- Event taxonomy and measurement plan tied to the business definition of success
- Tag manager, analytics and conversion implementation
- Consent Mode v2 configuration and a consent interface that respects refusal
- Attribution model: what each source is credited with, and its limits
- Automation of the defined operational workflows, with approval gates where they matter
What it does not do
- We do not run marketing tags without a working consent gate, in any jurisdiction, whatever the deadline is.
- We do not put a language model between a customer and an answer nobody approved.
- We report what was measured. Where a number cannot be attributed honestly, we say so instead of estimating it.
Process
01
Measurement design
Define the events, the properties and the success criteria before touching a tag.
02
Implementation
Data layer in the application, a single container, and conversions wired once.
03
Verification
Every event tested in both consent states, with the personal-data boundary checked.
04
Automation
Map the workflow, automate the deterministic parts, keep a person on the consequential ones.
Questions
Do you install GA4 directly or through a tag manager?
Through a single tag container in almost every case, with the application pushing named events into the data layer. Direct installation is only used when there is no tag manager and no plan to have one — and we never run both, which is the usual cause of double counting.
How do you keep personal data out of analytics?
The data layer carries identifiers and categories, never names, emails or phone numbers. Attribution identifiers travel with the form submission on the server side instead of being pushed into third-party tags in the browser.
Where does AI fit in?
In the operational layer: classification, drafting, enrichment and routing, where the output is checked before it has any external effect. We do not put a language model between a customer and an answer nobody approved.
—Start
Have a system worth building?
Tell us what the business needs to do. We will tell you what it takes to build it, and what it takes to keep it running.