A case study is only publishable if the numbers survive scrutiny
Customer approval tends to stall at the case study stage. The person who receives the draft sees a figure they cannot verify or a quote they do not recognize, and then the article is abandoned. The cost of rewriting at that point ends up greater than the cost of the initial draft.
The two failures come from the same habit: smoothing. A quote that has been altered for easier reading is no longer a quote. A number that has been rounded to make the sentence look cleaner is one the customer's analyst will not be able to reproduce.
The entire system is based on two rules. Quotations are copied word for word. Each figure includes the source it originated from, and that source is visible in the draft sent for approval.
Sending the reference list with the draft reduces approval time from two weeks to two days, because the customer's team can check rather than ask.
How the ActionFlow runs on the canvas
Two sources supply the material, and they stay separate at all times. A Document Upload node takes the transcript. A Google Sheets or PostgreSQL node supplies the result data.
The Generate Object node takes the quotes exactly as they are and presents the stated outcomes as separate items, each outcome including a reference to the data row it relates to. Extracting the text rather than constructing it is what preserves the exact words.
The Validate node checks each extracted item against the provided data. If a number is not present in the source, the run stops instead of producing a draft.
A Generate Text node prepares the narrative from the verified extraction. Because the quotes and figures are established inputs at this stage, the writing cannot adjust them.
The draft is sent to the customer together with the reference list. Google Docs saves the document so their team can leave comments directly in it.
A Notion node stores the approved version with its references, so you can reuse a figure in a deck two quarters later without having to re-verify it.
Nodes this ActionFlow uses
- Document Upload (Data): takes the interview transcript.
- Google Sheets (Integration): supplies the result data. PostgreSQL and MySQL fit the same slot.
- Generate Object (AI Core): extracts quotes verbatim and outcomes with a data reference on each.
- Validate (Data): stops the run when a figure does not appear in the source data.
- Generate Text (AI Core): writes the narrative around fixed quotes and verified figures.
- Email (Human in the Loop): routes the draft to the customer with the reference list.
- Google Docs (Integration): hosts the draft for in-place comments.
- Notion (Integration): stores the approved version with its references.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.