Case Study

From Handwritten Notes to Client-Ready Proposal

An AI Document Automation Build

A portfolio build by Finlay Systems — this documents an advanced-tier portfolio project built on a fictional-but-realistic scenario, not a paid client engagement.
AirtableZapierOpenAI GPT-4o (Vision)Google Docs

Best fit for

This automation suits teams repeatedly converting notes, forms, emails, or intake details into polished documents:

  • Real estate teams writing listing descriptions, buyer recaps, and property summaries
  • Service businesses creating proposals, estimates, or project briefs
  • Operators losing time to repetitive document formatting
  • Teams wanting faster follow-up without sacrificing quality or review

Overview

This case study documents an advanced-tier portfolio project. The scenario: a service business owner losing hours weekly to repetitive proposal writing, with no bandwidth to hire help.

The solution converts a photo of handwritten jobsite notes into a fully formatted, client-ready Google Doc proposal in under two minutes, with a prewritten follow-up message queued for later use.

Build time: ~3 hours  ·  Complexity tier: Advanced (Power-Up version of three tiers)

The problem

A mid-sized general contracting business — 4 to 10 employees, closing nearly 100 jobs a year — with the owner writing every estimate personally. Site visits happen during business hours; admin, invoicing, and estimate writing happen at night and on weekends. Jobs get lost when follow-ups slip through the cracks.

Only the owner produces estimates, and estimates are the single most important document in the sales pipeline.

NYC real estate analogs: listing write-ups, buyer tour recaps, property briefings, offer summaries, and transaction follow-ups.

The solution

A workflow where field users write handwritten notes, take a photo, submit through a mobile form, and receive a polished, client-ready document minutes later — with a prewritten follow-up message saved and ready to send.

Architecture

Five layers, in order:

1
Input layer
A public Airtable form accepting photo uploads.
2
Data layer
An Airtable base storing every submission.
3
Orchestration layer
A seven-step Zap watching Airtable.
4
AI layer
Multiple GPT-4o calls, one per subtask.
5
Output layer
A formatted Google Doc, with the link written back to Airtable.

How it works, end to end

  1. Field capture — notes written by hand during a site visit.
  2. Mobile submission — a photo uploaded through an Airtable form, status auto-set to “ready for automation.”
  3. Trigger and filter — a Zap polls every two minutes, filtering on status to avoid reruns on unrelated edits.
  4. Vision transcription — GPT-4o transcribes the handwriting into clean text.
  5. Structured data extraction — separate GPT-4o calls pull name, phone, email, and address as clean atomic values, with explicit fallback instructions to prevent hallucination.
  6. Project narrative generation — a GPT-4o call writes a polished, two-paragraph client-facing description.
  7. Follow-up draft generation — a final GPT-4o call writes a short follow-up message for three days later.
  8. Google Doc generation — a new doc, formatted with H1/H2 headings, including customer details and placeholders for cost and timeline.
  9. Write-back to Airtable — the doc link, follow-up message, and a status change to “complete” land back on the source row.

Key technical decisions

  • Airtable over Google Sheets — native attachment handling, structured fields, and production-ready forms without custom front-end work.
  • Zapier for orchestration — built-in connectors for OpenAI, Airtable, and Google Docs mean no custom API plumbing (the tradeoff is cost at scale — Make or n8n become more attractive as run volume grows).
  • Multiple small GPT calls, not one giant call — each subtask gets its own prompt, which costs slightly more in tokens but makes every step debuggable in isolation.
  • Memory key tied to the Airtable record ID — the AI “remembers” the same project across steps automatically.
  • HTML tags for Google Doc formatting — wrapping section titles in <h1>/<h2> tags produces properly styled headings with no template files.
  • Status-based triggering, not view-based — prevents unwanted reruns when unrelated fields are edited later.

Impact

The course framework this build was created for estimated $8,000–$16,000 in annual ROI for a business like this one from the core workflow alone — driven by faster response time on estimates, reduced admin fatigue, and fewer forgotten follow-ups. The Power-Up version amplifies this by removing the data-entry step entirely: a proposal that used to take 30–60 minutes of focused evening work becomes a two-minute background process that starts from a jobsite photo.

What this means for NYC real estate teams

The same architecture applies directly to listing write-ups, buyer tour recaps, property briefings, offer summaries, and transaction follow-ups. A high-volume team lead managing 20 active listings is constantly context-switching between showings and write-ups — the opportunity is the same as the contractor scenario: automate everything between raw field notes and the finished client-facing document.

About this build

The full build runs on free tiers of Airtable and Google, with a small pay-per-use cost for OpenAI API calls — typically well under a dollar per document generated.

Frequently Asked Questions

What is AI document automation?

AI document automation is the process of using AI to turn raw inputs, such as notes, forms, emails, or uploaded files, into structured documents like proposals, recaps, reports, briefs, or follow-up messages.

What does this AI document automation workflow do?

This workflow turns a photo of handwritten jobsite notes into a formatted, client-ready Google Doc proposal. It also extracts key details, creates a polished project narrative, drafts a follow-up message, and writes the final document link back into Airtable.

What tools were used in this build?

The workflow used Airtable, Zapier, OpenAI GPT-4o with Vision, and Google Docs. Airtable handled the form and database, Zapier orchestrated the workflow, GPT-4o processed the handwritten notes, and Google Docs generated the final proposal.

Can this workflow read handwritten notes?

Yes. The workflow uses GPT-4o with Vision to read a photo of handwritten notes and convert the content into clean text that can be used for document generation.

Does the user have to type anything?

No. The user writes notes by hand, uploads a photo through a mobile form, and the automation handles the transcription, data extraction, narrative writing, document creation, and follow-up draft.

Why use Airtable instead of Google Sheets?

Airtable handles attachments, structured fields, and mobile-friendly forms more easily than Google Sheets. For this workflow, Airtable made it easier to accept a photo upload, store project details, and track the automation status in one place.

Why use multiple GPT calls instead of one large prompt?

Multiple smaller GPT calls make the workflow easier to test, debug, and control. Each step has one job, such as transcription, customer detail extraction, project narrative writing, or follow-up drafting. This reduces the risk of messy or unreliable output.

How can this apply to real estate teams?

The same workflow pattern can help real estate teams turn raw notes into polished client-facing documents, including listing write-ups, buyer tour recaps, property briefings, offer summaries, transaction updates, and follow-up emails.

What types of real estate documents could be automated?

Examples include listing descriptions, property feature sheets, showing recaps, buyer tour summaries, internal team briefings, open house follow-ups, offer summaries, seller updates, and transaction status reports.

Does AI replace the need for human review?

No. The output should still be reviewed before being sent to a client. The goal is to remove the repetitive first draft, formatting, and data-entry work, not to remove human judgment.

How much does this type of automation cost to run?

In this build, the ongoing run cost was typically well under a dollar per document generated, using free tiers of Airtable and Google with small pay-per-use costs for OpenAI API calls.

How long did this build take?

This portfolio build took approximately three hours to create. A client version may take longer depending on the tools, document requirements, review steps, formatting needs, and integrations involved.

What is the main benefit of this workflow?

The main benefit is speed. A document that might normally take 30 to 60 minutes of focused admin work can become a short background process that starts from a simple phone photo.

What does Finlay Systems do?

Finlay Systems provides AI consulting for high-volume NYC real estate teams, including AI training, workflow automation, strategy, implementation, and practical AI systems that reduce manual admin work.

This is the kind of practical AI workflow Finlay Systems builds for real estate teams and service businesses. If your team is losing hours each week to repetitive work, I can help identify what is worth automating.

Book an Intro Call