Marketing automations: reviewed drafts and analysis
Marketing automations turn repeatable marketing inputs into briefs, draft content and analysis that a person can review. Start with a defined source, a named deliverable and an approval decision. This guide covers preparation and analysis in AI Flowi Workflow; campaign delivery, advertising purchases and attribution prediction are outside its scope.
By Ai Flowi · Published
Choose the marketing task before choosing nodes
A content brief and a campaign report need different evidence. For a brief, supply approved positioning, audience and constraints. For analysis, supply the actual export and its measurement definitions. Structured Output offers named extraction fields, while aMap processes table rows independently. Neither component establishes whether a marketing claim is true or whether a campaign caused a sale.
Related: AI content repurposing.
| Task | Prerequisite and deliverable | Reviewer and failure check |
|---|---|---|
| Content brief | Approved product notes, audience and objective; produce an outline with supporting references. | A product reviewer checks each claim against the supplied notes. |
| Draft queue | An approved brief per row; request a draft and a list of unresolved claims. | An editor checks tone, source support and missing permissions before publication. |
| Campaign export analysis | A dated export with metric definitions; produce observations tied to supplied columns. | An analyst checks denominators, date ranges and absent conversion data. |
| Approval handoff | A draft identifier and reviewer decision; keep the reviewed version distinct from working text. | The publishing decision belongs to a person; a generated approval field is insufficient. |
Prepare a source contract
For a draft queue, use columns such as brief_id, audience, approved_message, evidence_reference and prohibited_claims. These are proposed business fields, not built-in component settings. Carry brief_id into the draft review so an editor can locate its source row. Flag an unsupported benefit for the product reviewer; general model knowledge cannot supply approval for a campaign claim.
For campaign analysis, keep the export period, currency, channel definitions and missing-value conventions beside the data. A blank conversion cell must not silently become zero. Record whether the export contains aggregate metrics or individual records; do not send identifiable audience data when aggregate inputs answer the question. Decide which comparisons are valid before composing the analysis prompt.
Accounts and catalogue inputs
Prepare the approved brief sheet with read permission through Google connection setup, and configure the selected model with your own provider credentials through the model reference; provider usage is billed separately. Keep secrets out of campaign cells and public examples.
Use the aMap catalogue entry for its source and schema contract, and the conversion reference for preparing the brief table.
As a planning example, suppose an approved brief describes a service as available by appointment while the requested headline implies immediate access. Put the appointment restriction in approved_message and flag the headline conflict in prohibited_claims. The draft should keep the restriction and return the conflicting request as unresolved_claims; attractive wording does not resolve it. If another row describes a different service, its evidence must not supply missing benefits to the first row. Review each draft against its own brief_id before moving it into the editorial queue.
Build a bounded draft queue
example configuration, not a recorded test
- Read Range: choose the credential in google_account. Use the approved brief sheet’s identifier for spreadsheet_id and restrict range_name to its brief columns. The values output supplies JSON. Check the returned headers and row order with a synthetic brief before connecting downstream processing.
- Type Convert: use Read Range.values as input_data with output_type = “Table” following the documented conversion settings. For this queue, check that brief_id, evidence_reference and prohibited_claims remain on the same row; misplaced headers would attach the wrong evidence to a draft.
- aMap: connect the inspected Table to source, configure model and provider credentials, and define schema entries for draft_text, evidence_used and unresolved_claims. Retain source columns with append_to_input_columns = true; write instructions that preserve brief_id and restrict drafting to the approved material within that row.
- Structured Output: for a separate single-brief extraction, set input_value to an approved text brief, model to your configured model, schema_name to MarketingBriefReview, and output_schema to your review fields. Set system_prompt to distinguish supported claims from unanswered questions. This is an alternative text path, not a direct Table connection to input_value.
- Review: inspect aMap.states or Structured Output.structured_output. Record the editor’s decision outside the generated text, and use your established publishing process only after that review. This proposed flow contains no delivery node.
Use a different contract for campaign analysis
Feed a supplied campaign narrative into Structured Output.input_value when the job is to separate reported observations from interpretation. Choose fields such as period, metric_name, observation, source_column and unanswered_question. A period-to-period change can be described from an export, but the explanation for that change remains a hypothesis unless the source establishes it. Ask the model to state which inputs would be needed to assess the hypothesis.
Independent aMap rows do not provide whole-campaign totals by themselves. Calculate authoritative totals in the source export or another explicitly checked process, then supply them as evidence. Do not let row summaries introduce new denominators or compare campaigns with different reporting windows. When a campaign metric is absent from the export, name it in unanswered_question and explain which comparison cannot be assessed.
What the review artifact looks like
The intended draft artifact retains the source brief beside proposed copy, evidence references and unresolved claims. It is a review queue, not a record of sent messages. The analysis artifact contains dated observations, the exact supplied metric names and questions requiring investigation. No campaign results or example performance figures are implied by these field descriptions.
Try an incomplete brief, contradictory positioning notes and an export with a missing metric. Require a source reference for each claim in the proposed copy. If a draft invents an offer, rewrites a restriction or marks itself approved, hold the output and revise the instructions or source contract before processing campaign briefs supplied by your team.
When should I skip marketing automation?
For a single short brief, an existing document checklist may cover the task. Avoid this design when your requirement is a delivery suite, audience targeting, automatic ad buying or revenue attribution. Email and social sending are not covered here. These marketing components are documented in the catalogue; not production-tested by Ai Flowi.
Check executed nodes, retries and provider calls when estimating usage. Table processing can involve multiple model requests even though the workflow displays one analysis component. The catalogue is not a billing trace for your proposed draft queue.
Plans: Go from USD8/month billed yearly (USD10 monthly), Plus from USD16 (USD20 monthly), Pro 100 from USD80 (USD100 monthly), Pro 200 from USD160 (USD200 monthly) and Pro 500 from USD400 (USD500 monthly), with 5,000 to 250,000 Flowi Tokens/month and top-ups; 1 token per node run; third-party API and model usage is not included.
Extend the input evidence
Use website change monitoring for reviewed source changes and AI sales-agent boundaries for the downstream sales handoff. The use-case hub covers individual starter tasks; keep their configuration and test evidence separate from this marketing process.
Related: AI for nonprofits: prepare grant evidence and reviewed drafts.