Can every material claim stay traceable?
The research package carries source provenance forward. Unknown costs remain custom estimates, and approval-dependent details remain subject to confirmation.
Case study · Controlled proposal-production pilot
Personalized proposal production
The operating question
How do you personalize a proposal without losing the evidence trail or sending the wrong file?
01 · Case studyDental payment integrationControlled posting inside the clinic02 · Case studyWedding venue proposal automationDraft estimates with a human send gate03 · Case studyPersonalized proposal productionResearch, render, verify, then stop04 · Case studyPatient presentation automationLocal-first supervised assembly05 · Case studyAI enabling a semi-private airlineSeparate instruction, practice, and weekly guides
What the system needed to solve
A finished PDF is not enough.
A personalized proposal can look complete while hiding weak research, a broken layout, or the wrong file. The useful system had to preserve the source trail, produce a stable artifact, prove which file was delivered, and stop before outreach.
Can the artifact fail safely?
Structural checks and full-resolution visual review sit before publication. A failed run stops instead of being waved through.
Is the reviewed file the file that lands in Drive?
The system downloads the published PDF again and checks its hash against the reviewed artifact.
Before · fragmented proposal production
Research, asset gathering, document assembly, review, and upload can happen as separate acts with no common proof trail.
- Gather venue facts and images
- Write and assemble the proposal
- Review the visible document
- Upload a file separately
Designed path · controlled artifact production
One approved lead moves through a repeatable record from sourced research to a verified PDF.
- Approve one lead and research package
- Render the six-page proposal
- Pass structural and visual QA
- Publish, read back, and verify
The system boundary
The pipeline produces proof. A person authorizes contact.
The production controller moves one approved lead through research, rendering, verification, and publication. Sending remains a separate human action outside the pipeline.
System work · Build and verify the proposal artifact
Human work · Approve the lead and authorize any outreach
Approve
Select one lead and authorize the production run.
Research
Build a traceable evidence and asset package.
Render
Assemble the six-page proposal from controlled inputs.
Verify
Run structural checks and full-resolution visual QA.
Publish
Upload, download, hash, and verify the exact artifact.
Stop
No email or draft is sent without separate approval.
What the system does: researches the approved lead, renders the proposal, checks the document, and proves the published file matches the reviewed file.
What the system does not do: invent missing commercial terms, ignore a failed check, or send outreach on its own.
Failure controls
The run earns READY.
A proposal reaches the ready state only after the artifact and the handoff record agree. If a check fails, the run remains failed and a retry receives a new run identifier.
The controller processes a single lead and rechecks the live approval state before publication.
Unverified costs stay marked as custom estimates. Details that require venue approval stay marked for confirmation.
Clipping, overlap, broken imagery, unsafe margins, and contrast defects are checked at full resolution before the file can move forward.
The reviewed PDF and the Drive-downloaded PDF must carry the same SHA-256 hash.
The build can reach READY without creating an email draft or contacting a prospect.
Controlled pilot · Verification record
Reviewed PDF
= Drive file
Actual implementation record · Prospect identity withheld
The proof is about the production system. The controlled pilot establishes artifact creation, QA, and exact-file readback. It does not establish replies, conversion, revenue, or time saved.
What this record establishes
The system produced a six-page proposal package, passed 72 automated tests, cleared full-resolution review, and verified the Drive-downloaded file against the reviewed PDF. A separate run stopped at visual QA. No outreach outcome is claimed because no proposal was sent as part of the controlled pilot.
The transferable idea
Automate the document. Keep authorization human.
For proposals and other externally facing artifacts, useful automation does more than generate. It preserves sources, proves the file, stops on failure, and leaves the consequential action with a person.