Prova artifact
AI Workflow Audit Template for Marketing Teams
Map one marketing workflow and decide what AI should automate, augment, protect, or decline before the team commits to a pilot.
Best forAgency leaders redesigning delivery quality around AI · Marketing ops or analytics leads choosing a first AI workflow · Senior marketers trying to keep scattered AI experiments from becoming operating risk
Template fields
- Workflow stage
- What AI handles today
- What AI should handle in six months
- What stays human and why
- Ethical or operational risk: low, medium, high
- Action: automate, augment, protect, decline
- ICE score for pilot candidates
Worked example
Client reporting workflow for a paid media account.
The filled examples below stay in English because Prova reviews submitted artifacts in English.
Weak version vs strong version
Weak version
- Workflow stage
- Reporting
- AI handles today
- AI writes reports
- Stays human
- Strategy
- Risk
- Medium
- Action
- Automate
| Workflow stage | Reporting |
|---|---|
| AI handles today | AI writes reports |
| Stays human | Strategy |
| Risk | Medium |
| Action | Automate |
Why it fails
- "Reporting" is too broad to audit.
- "Strategy" does not name the judgment point.
- It does not identify source systems, reconciliation risk, or who approves the client-facing narrative.
- Automating the whole workflow is unsafe because the work includes interpretation and client trust.
Strong version
- Workflow stage
- Weekly paid media variance explanation for Meta, Google, and LinkedIn
- AI handles today
- Drafts first-pass variance notes from exported spend, revenue, CPA, and campaign-change logs
- AI should handle in six months
- Flags likely causes, drafts three audience-specific summaries, and proposes follow-up checks
- Stays human
- Strategist approves causal claims and client recommendation because platform data can be delayed or misleading
- Risk
- High for client narrative, medium for internal diagnosis
- Action
- Augment, with protected human approval before client send
| Workflow stage | Weekly paid media variance explanation for Meta, Google, and LinkedIn |
|---|---|
| AI handles today | Drafts first-pass variance notes from exported spend, revenue, CPA, and campaign-change logs |
| AI should handle in six months | Flags likely causes, drafts three audience-specific summaries, and proposes follow-up checks |
| Stays human | Strategist approves causal claims and client recommendation because platform data can be delayed or misleading |
| Risk | High for client narrative, medium for internal diagnosis |
| Action | Augment, with protected human approval before client send |
Why it works
- The stage is specific.
- Source systems are named.
- The human judgment point is explicit.
- Risk differs by use case.
- The action is narrow enough to pilot.
What Prova reviews that generic AI often misses
- Whether the workflow scope is specific enough to audit
- Whether "stays human" reasons are concrete or just polite language
- Whether high-risk stages are incorrectly classified as automate
- Whether ICE scores are supported by operational reality
- Whether the selected pilot is narrow enough to test
- Whether the next sprint should be measurement, rollout, or foundation repair
Next step
Want feedback on your version? Prova starts with a short assessment so your review standard matches your role, goal, and first audience. After that, you enter the sprint that fits your current work.
Prova is currently available in English only.
Before submitting: remove client names, confidential numbers, and anything your team would not want stored in a training or coaching system.
Start your Prova review