12 questions to answer before you plug agents into your marketing automation. If you can't check most of these, agents will make your mess faster.
// CLICK EACH ITEM AS YOU GO | YOUR SCORE UPDATES LIVE AND SAVES IN YOUR BROWSER
Not a PDF report. A JSON or CSV with touchpoints, timestamps, and attribution. Agents can't reason over screenshots.
If no: your agents will hallucinate journey insights from whatever fragment they can access.
Check your CRM, ESP, ad platforms, and analytics. If the only way to export data is "click Export → CSV," agents can't touch it without brittle screen-scraping.
If no: you'll build a human-in-the-loop bottleneck that defeats the purpose.
Agents will generate campaigns. If your UTMs are already inconsistent, agents will scale the inconsistency. Write the convention down. Enforce it with validation.
If no: your attribution model becomes noise within one quarter of agent-generated campaigns.
API-level access to segments, tags, or cohorts. Agents need to say "send this to segment X" without a human clicking through a UI.
If no: every agent campaign needs a human clicking through a UI first. You've automated the copy, not the workflow.
CDP, data warehouse, or even a well-maintained spreadsheet. If customer data lives in 4 tools with no sync, agents will pull conflicting records and act on stale info.
If no: fix this first. No amount of AI fixes a data architecture problem.
Agents should be able to compose and send emails via API, not just draft copy for a human to paste into a template builder.
If no: your agent writes copy but a human still has to paste it into the builder and hit send. You've automated one step out of five.
One query, one API call, one place. Open rates, conversions, revenue. If reporting requires a human tour of your stack, agents can't close the feedback loop.
If no: agents will optimize what they can measure and ignore what they can't. That's how budgets leak.
Agents will make mistakes. You need a gate between "agent composed it" and "it went to 10,000 people." Even a Slack confirmation counts.
If no: your first agent error becomes a customer-facing incident.
GDPR, CAN-SPAM, and your own suppression rules need to be queryable. An agent that can't check "has this person unsubscribed?" is a compliance violation waiting to happen.
If no: you're one agent mistake away from a GDPR complaint.
Workflows, triggers, conditions. Are they in git, or locked in a visual builder with no export? Agents will modify flows. You need rollback.
If no: you'll be debugging agent changes by clicking through a UI at 11 PM.
Daily caps, budget alerts, anomaly thresholds. An agent optimizing spend without guardrails is an agent that can drain a quarterly budget in a weekend.
If no: set hard caps before any agent touches bid management.
If your entire workflow can run via API calls (segment, compose, send, measure, iterate), you're agent-ready. If any step requires a browser, that's where your bottleneck lives.
If no: start with whichever step still needs a browser. That's your first project.
Want the weekly breakdown of what's actually changing?