A 30-day ramp for an AI marketing role
Start with one accepted result, prove the source and review boundary, then add volume, cadence, and a second marketing role only after the first job holds.
Give the first AI marketing role one accepted result in week one. Use week two to prove source access, exclusions, and correction handling. Use week three to repeat the result at its real cadence. Use week four to decide whether the role has earned broader inputs, more volume, or a handoff to a second role.
Do not begin by giving two roles every marketing account and asking them to “handle marketing.” The first marketing hire guide explains why a role cannot recover context that remains only in the owner's head.
Choose the first role from the constraint
Start with the delayed marketing work product
The first role follows the bottleneck and the current public scope.
| Current problem | First role to inspect | First accepted result |
|---|---|---|
| Customers cannot find accurate business facts | FARO, AI SEO strategist | A search audit with URLs, dated evidence, ranked corrections, owners, and recheck dates |
| Approved evidence is not becoming publish-ready work | SCOUT, content operations lead | One brief and editable draft package with sources, checks, unresolved claims, and release state |
| An approved outbound campaign lacks research and follow-through | VYRA, outreach and SDR lead | An audience and campaign package or reply queue within the owner's sending boundary |
| Marketing direction, offer, or audience is still undecided | A qualified person | A strategic decision before production is assigned |
The current AI agent directory controls availability and exact role scope. A future role should remain marked unavailable until it enters that catalog.
Use the constraint selection guide to find the work product holding up other work. Use the current AI agent directory to verify that the role, work product, connection, limit, and approval boundary exist today.
Days 1 to 7: prove one useful result
Pick one representative assignment, not the easiest clean sample and not the most sensitive exception. Record the audience, source set, excluded records, required fields, reviewer, destination, and approval boundary.
For FARO, the first result might be one search audit. For SCOUT, it might be one evidence-backed article package. For VYRA, it might be one approved audience and message package with sending held.
When the role holds, it replaces some manual search checking, source gathering, first-pass production, packaging, reply sorting, and follow-through that people perform today. People still decide positioning, approve sensitive claims and public actions, handle unfamiliar conversations, and own the commercial result.
Done when: the reviewer can open every cited source, find every required field, see unresolved facts, record corrections, and identify the actions that remain pending.
“Week one is successful when the work can be accepted or rejected from evidence, not when the role produces a lot of material.”
Days 8 to 14: test access and failure
Connect only the sources required for the first result. For FARO, Google Search Console's Performance report can supply query and page records when the approved property and date range are explicit. Test one allowed record, one nearby denied record, one missing field, one conflicting source, and one revoked connection. Review the security and access boundary before approving customer, employee, finance, or private communications data.
The technical connection and business authority are different decisions. Access to a content calendar does not approve publication. Access to a pipeline does not approve sending. Access to Search Console does not approve site changes.
The role should report an incomplete result as incomplete. A blocked source range, partial API response, or stale document must remain visible in the review record.
Days 15 to 21: repeat the real cadence
Run the work at the actual frequency and volume expected from the engagement. Measure:
- complete intake to accepted result;
- reviewer minutes and correction types;
- missing or conflicting source records;
- accepted artifacts, not drafts created;
- decisions and actions still waiting on a person.
The outcome acceptance guide keeps the measure tied to finished work. The trigger catalog can define the schedule, message, document, record-change, threshold, or handoff event that starts the next run.
A four-week role ramp expands only after evidence
Each week adds one operating demand while preserving the same acceptance record.
- 1
Week one: one accepted result
Prove the work product, sources, reviewer, corrections, limits, and destination.
Owner: Role and buyer reviewer
- 2
Week two: boundary tests
Prove allowed access, denied access, missing evidence, conflicts, write holds, and revocation.
Owner: Administrator and reviewer
- 3
Week three: real cadence
Repeat the work at representative volume and measure elapsed time, correction burden, and accepted output.
Owner: Role and process owner
- 4
Week four: coverage decision
Keep the role narrow, expand approved inputs, change the cadence, or add a second role only where the handoff is explicit.
Owner: Hiring owner
Days 22 to 30: decide whether to add a second role
Add a second role only when the first role's output becomes a clean input to another current role. FARO may produce ranked search corrections that support a SCOUT brief. SCOUT may produce an approved campaign asset that supports a VYRA package. The handoff should include artifact, source links, approval state, exclusions, owner, and next completion check.
Do not imply that every marketing task belongs to one of those roles. Positioning, offer design, customer conversations, budget allocation, sensitive claims, publication, sending authority, and unfamiliar market judgment remain with people unless a current role states a narrower supported action.
The three roles or half a headcount guide helps compare several narrow roles with one person who can own the gaps.
Use the work environment deliberately
Slack is the shared-team review surface. WhatsApp is a compact owner brief and decision surface. Microsoft Teams is the governed Microsoft 365 surface. Slack's Events API can deliver subscribed events to an approved app, while Microsoft's proactive-message guidance covers messages a Teams app sends outside an active reply. Neither connection decides which business events deserve action. The work-environment guide explains the distinction.
The message should link to the durable artifact, state what changed, surface missing evidence, request one explicit approval, and retain the final state. Do not let a chat thread become the only copy of the work product.
Stop or narrow the role when the evidence says so
Narrow the role when correction burden stays high, required context is not inspectable, source access is broader than the value justifies, or the work repeatedly ends in decisions only a person can make. Stop when the role cannot produce the accepted result without fabricated facts, unsafe access, or authority the buyer should not delegate.
Review the work-product promise for current delivery and remedy terms. Slack history stays in the buyer's Slack under its plan and retention, and work written into the buyer's existing systems stays there. The full activity log exists only when the paid pre-deployment add-on was enabled before work began. There is no special FidelicAI cancellation export; the internal role method and private test records remain vendor-side. For WhatsApp or Teams, record account ownership and the channel-retention arrangement before work begins, and do not assume history survives cancellation unless that arrangement supports it. Review what you own if you cancel before relying on a work environment's retention.
Follow the connected questions
Start and work together includes this decision and the questions that usually change it.
How do you start working with an AI agent?
Start with one useful assignment, the source records it needs, the person who approves consequential work, and a dated review of the first result.
Use the 30-day work plan →What commercial work can an AI agent own?
It can maintain research, outreach records, search briefs, catalog work, and order follow-through. Storefront changes, public messages, and other binding actions require owner approval.
What belongs in an AI agent work order?
Name the assignment, source set, work product, checks, approval point, delivery window, and what happens when an input is missing.
What should you do next?
Choose one marketing work product and one current role. Write the week-one acceptance check before granting access. After one accepted result and the week-two boundary tests, select the Day Pass, Sprint, or Monthly Retainer that matches the actual work shape. Add a second role only after the handoff exists in writing.