ONLINEcascade://samwarren.iov0.1.0
all projects
FIRING · PROD · DAILY
SKILL BUNDLE · REPLACES A ROLE· 05 / 06

The Marketing Ops Agent

Five skills, one operating capability.

Attribution, hygiene, campaign ops, reporting ops, and a weekly data refresh — each running on command or on a schedule, all learning from corrections. The kind of work that used to require a dedicated Marketing Operations hire, running inside Cascade as a single operating capability.

95%
ATTR. ACCURACY
20+
LEARNED RULES
54%
HYGIENE AUTO-FIX
~8 hrs
WEEKLY RECLAIM
PROBLEM

Marketing operations is mostly rules work. AI is very good at rules work.

A typical MOps week at any growing B2B company is the same seven or eight tasks running in a loop. UTM hygiene on paid social leads. Lifecycle stage corrections. MQL date backfills. Opportunity attribution. Quarterly campaign ops. Weekly data refreshes. Reporting ops ahead of QBRs. None of it requires creativity. All of it requires judgment.

The industry's answer has been to hire another marketing operations analyst. My answer has been to package the judgment and let the rules run on their own.

THE LEARNING LOOP

Attribution accuracy: 50% → 95% in three months of corrections.

The attribution skill is the clearest example. It pulls unattributed opportunities, enriches them with engagement data from the MA system, applies 20+ rules, and presents recommendations for human review. It never auto-applies. Every correction gets captured as a persistent learned rule.

The rules aren't hand-coded. They're learned patterns, stored as feedback memories — things like "event demo credit requires Attended status, not just Registered" or "nurture-driven sign-ups get the nurture campaign, not the sign-up form." 20+ of these, each one earned by being wrong once.

The system doesn't know the answer on day one. It knows how to capture the answer once I've found it.
OUTCOME

A role's worth of work, reclaimed.

The hygiene skill runs a scan every morning. It catches UTM placeholder errors, missing MQL dates, lifecycle mismatches, and missing touch sources. Safe corrections auto-apply. Ambiguous ones surface for review.

The campaign, reporting, and refresh skills handle the quarterly and weekly operational tax that used to occupy an entire human calendar. Net effect: roughly eight hours a week of work now runs on its own — and the ten minutes I still spend in the loop is the interesting ten. The judgment calls, not the plumbing.

▸ components

What’s inside

sf-attributionOpportunity attribution with persistent learning loop
hubspot-opsWorkflow management, UTM hygiene, MQL backfill
sf-campaign-clonerBulk quarterly campaign operations
sf-reportsBulk report, dashboard, and list view updates
data-refreshSunday-night refresh pipeline across every tracked source
▸ stack
salesforce apihubspot v4claude reasoningpython
▸ clone it
▸ public repo · ready
The methodology + a Python package. Rules as composable functions, a persistent learning loop via SQLite, CSV adapter shipped (Salesforce/HubSpot adapters via the five-method Protocol). 5 reference rules, 7 passing tests, 40-case pattern library documented in docs/methodology.md.
▸ connects to
NEXT · 06 / 06
Claude Dotfiles
The brain, in a repo.