Why I Stopped Running Marketing Reports and Started Building Marketing Systems
The analyst who runs reports is replaceable. The analyst who builds the system that runs reports is not. Here is the shift that changed how marketing analytics work gets done.
A common version of the marketing analyst job is mostly retrieval: pull the weekly numbers, format the deck, present, repeat. That version is being automated, and it should be.
The version that holds is different. Someone still has to decide what gets measured, how those numbers connect to a real decision, and what infrastructure makes the insight show up again without a human repeating the same steps next Tuesday.
That is the shift: stop being the report. Build the system that produces it.
The Report Is Not the Work
When a stakeholder asks for a marketing report, they usually want a decision input. Spend more? Change creative? Shift channel mix? Hold?
If the same report is rebuilt by hand every week, analytical judgment gets buried under retrieval and formatting. That is the scarce part. The spreadsheet is not.
The right response to a recurring report request is rarely "make another one." It is "what is the smallest system that produces this on a cadence so a person can spend time on interpretation, not generation."
What an Orchestrated Analytics Stack Looks Like
Three layers. Ingestion moves data from source systems into a structured store without manual exports. Transformation encodes business logic as version-controlled code: attribution rules, margin math, segment definitions. Delivery puts the output in front of the right people on the rhythm their decisions actually follow.
The analyst's job is to define the logic at each layer, watch for drift, and change the system when the business changes. Not to re-run the whole pipeline by hand.
Ingestion
- Ad platforms
- CRM systems
- Web analytics
Transformation
- Attribution rules
- Margin calculations
- Segment definitions
Delivery
- Slack @8am daily
- Email weekly
- Monthly review
Orchestrated Analytics Stack · 3-Layer Architecture
Where AI Agents Enter
Agents add a reasoning layer inside the pipeline. They move more than data; they handle routine evaluation that used to queue behind a human.
A signal agent scans campaign data and flags anomalies worth looking at before anyone opens the dashboard. A build agent drafts copy variants from historical performance patterns. A gate agent checks outputs against quality criteria before something reaches a stakeholder.
None of that replaces judgment. It compresses the lag between data and decision, and it absorbs volume that would otherwise stall a single analyst. The orchestrator designs the roles, sets the criteria, and reviews what escalates.
Continuously scans incoming campaign and revenue datasets to detect metric drift early.
Synthesizes historical top-performing patterns into structured draft deliverables.
Deterministic validator scoring outputs against compliance, brand, and quality criteria.
Gmail Webhook Trigger
Fires instantly on every incoming email. Passes the full payload (sender, subject, body, thread ID) to the triage chain.
Intent & Relevance Gate
Evaluates subject and body to return a structured yes/no via JSON schema parser. Newsletters, auto-replies, and spam are filtered here.
Category Classification
Assigns one of four operational intent labels. Category Parser structures the output parameters for downstream routing.
Confirms or clarifies party size, date, and seating preferences with structured booking language.
Responds to commercial catering inquiries with tiered pricing menus, guest minimums, and calendar availability.
Handles business hours, directions, dietary accommodations, and general operational queries.
Zero customer auto-draft. Escalates directly: the GM receives an instant push alert with full context.
The Mindset Shift
Reporter vs orchestrator shows up the moment a new request lands. One path: open the old file, update the dates, ship the deck. The other: decide if this is a one-off or a loop, then size the minimum system that makes the loop sustainable.
Asks:
“How do I produce this output?”
Asks:
“Should this exist once, or recurring with minimal effort after setup?”
Most recurring requests answer the "recurring or once" question in under a minute. If it is recurring, the next question is the one that matters: what is the smallest system that still earns trust?
That changes the shape of the work. And the shape of the work changes what that seat is worth.
Jason H. Vo
Marketing Analytics & Workflow Automation