Creating proposals is one of the biggest time drains inside agencies and service businesses. A sales lead lands, and half a day disappears rewriting the same sections, pasting numbers into spreadsheets, and fighting with PowerPoint to make the deck look on-brand. Then the client asks for a change — and you do most of that work again.
What if you could go from a chat-based brief to a fully designed, priced proposal deck in under five minutes, without touching a spreadsheet or a slide? That is exactly what this tutorial walks through: an end-to-end automated quotation system where Airtable holds the pricing engine, Make is the automation layer, an AI writer produces the narrative, and Gamma renders the deck.
Video Tutorial
The proposal generation architecture
The system has four moving parts, and each one does exactly one job. That separation is what keeps the whole flow editable long after the build.
- Airtable — the pricing engine. It stores the catalog, the default margins, the customer record, and the quotation + line items.
- AI model — the writer. Given a client brief and the catalog, it picks line items, drafts the narrative, and returns structured content ready to be sent to a deck.
- Make — the glue. It moves data between the chat interface, Airtable, the AI model, and Gamma, and keeps the quotation record in sync.
- Gamma — the presentation layer. It accepts structured Markdown-style content and produces a branded, shareable deck that the sales rep can tweak before sending.
Because each tool does one job, you can swap any piece later. If you want to move the writing step to a different AI, you update one Make module. If you decide to render in PowerPoint instead of Gamma, you swap the final module. The Airtable schema — which is where most of the business logic lives — does not move.
Airtable as the pricing engine
The heart of the system is the Airtable base. Five tables carry the weight:
- Price list — every item you have ever quoted, grouped into categories (venue, catering, AV, staff, etc.), with per-unit cost, default margin, billing unit (per day, per unit, per head), and a last-updated timestamp.
- Vendors — who supplies each line item, for downstream purchase orders.
- Customers — a lightweight CRM table with contact info and past proposals.
- Quotations — one row per proposal, with status, client, event summary, totals, and the generated Gamma URL.
- Quotation line items — the detail rows linked to a quotation, carrying quantity, item, override price, override margin, and a calculated total.
The price list is the grounding layer. The AI never invents numbers — it looks up items from this table and applies the stored margin. That is what keeps the proposal from hallucinating pricing. If you want the AI to know about a new service you sell, you add a row to the price list. You do not retrain a model, and you do not edit a prompt.
On top of the tables sits an Airtable interface with a list view (proposals by status, client, total) and a detail page showing the cover, brief recap, key line items, and budget summary in the exact order the final deck will use. Building those detail pages in Airtable first is a trick worth stealing from this tutorial: once the layout works in Airtable, the Gamma slides practically write themselves because the content blocks already line up. See our Airtable client reporting dashboards tutorial for more on how to structure these interfaces, and how to build a CRM in Airtable if you want a deeper take on the customer side of the schema.
ChatGPT vs Claude for proposal copy
In the video we put ChatGPT and Claude against each other with the same brief and the same Airtable grounding. Both passed. A few patterns held up across runs:
- Claude tended to produce tighter copy with fewer filler sentences, and was slightly better at sticking to the tone set in the system prompt. It also handled multi-line-item math more faithfully when the brief was ambiguous.
- ChatGPT was marginally faster end-to-end and produced more visually varied Gamma output — more bullets, more section breaks, more headings. For a sales team that wants the deck to feel "designed" without an editor looking at it, that matters.
The honest answer is that prompt plus schema beats model choice. If your Airtable price list is clean and your system prompt explicitly tells the AI which tables to read and which to never invent from, either model will produce a proposal you can send. See our types of Airtable AI agents tutorial for more on choosing models for business workflows, and AI automation for business for where each one fits.
The system prompt we use looks roughly like this:
You are a proposal writer for {{company}}. You will receive a client brief.
Use ONLY the pricing data provided below from the Airtable price list.
Do not invent items or prices. If the brief mentions a service that is not
in the catalog, flag it in a `missingItems` array and pick the closest match.
Return JSON with: `summary`, `lineItems[]` (with `priceListId`, `quantity`,
optional `marginOverride`), `deckSections[]` (cover, brief, keyItems, budget,
nextSteps).
The JSON return shape is the critical part — Make parses the response and writes each line item back into the Airtable quotation line items table, where the built-in roll-ups calculate totals the AI never had to compute directly.
Generating the deck with Gamma
Once the quotation record is populated, changing its status to Generate presentation fires the Make scenario that talks to Gamma. The steps, in order:
- Fetch the quotation and its line items from Airtable.
- Assemble a Markdown document that maps one-to-one to the Gamma slide structure: H1 = cover, H2 = each section (Brief, Key Items, Budget, Next Steps), bullet list under each H2 for the content.
- Call the Gamma API with the Markdown and the brand template ID. Gamma returns a deck URL and a PDF download link.
- Write the URLs back to the quotation record so the sales rep can open the deck straight from Airtable.
- Optionally export the PDF as an attachment on the quotation and email the link to the customer through the same Make scenario.
The reason Gamma beats slides-through-PowerPoint for this job is that Gamma renders each section with a layout that fits the content instead of a fixed template. A three-bullet section gets three clean tiles. A budget summary gets a table. A cover gets a hero layout. The sales rep can regenerate individual slides if the AI narrative is off without having to re-run the whole flow.
See our business process automation examples tutorial for more patterns on how to glue Airtable, Make, and AI together, and our make automation agency page if you want the Make scenario built for you.
Delivering the proposal
With the deck generated, delivery is the last mile. A second Make automation handles this once the proposal status flips to Ready to send. It can:
- Post the Gamma link and a personal note into the CRM (or a follow-up row on a sales tracker in Airtable).
- Email the client directly with a templated message that includes the Gamma URL and the PDF attachment.
- Schedule a follow-up task three days out if no click-through is registered.
- Log the open and view events back onto the quotation record so the rep sees engagement without leaving Airtable.
This is where the Airtable CRM view starts earning its keep. Every proposal is a row, every row has a status, every status fires the right scenario — so the rep's job shrinks to "review and send", not "build from scratch".
Business use cases
This pattern — catalog in Airtable, AI writer, Gamma deck, Make glue — transfers to almost any consultative sale:
- Agencies and studios — production quotations, event proposals, creative briefs.
- Professional services — scoped engagement proposals with phased deliverables.
- Construction and trades — priced bids with optional add-ons. See our construction solutions page for related workflows.
- Field sales teams — on-site quotes that need to leave the meeting as a branded PDF. See service field sales solutions.
- Marketing teams pitching internal projects — campaign plans with media spend breakdowns. See our marketing teams solutions page.
Anywhere there is a price list, a template narrative, and a sales rep under time pressure, this system pays for itself in the first week.
When to hire help
The build is not hard conceptually, but there are three places where teams get stuck: designing the Airtable schema so the price list actually grounds the AI (instead of working around it), writing the prompt so the JSON return stays clean, and wiring Gamma so the branded template renders cleanly at scale. If any of those sound painful, that is where an automation consultant saves you weeks of iteration.
Our team has built proposal automations for agencies, event companies, and professional services firms. If you want a system like this shipped to your team — with your catalog, your brand, and your CRM — reach out to our Airtable consultant team or our Make automation agency and we will scope it with you.
Next steps
- How to build a CRM in Airtable — foundation schema for the customer side of the proposal system.
- Airtable client reporting dashboards — interface patterns that carry straight over to proposal detail pages.
- Types of Airtable AI agents — picking the right model for each step of the workflow.
- Explore Claude and ChatGPT tool pages for integration specifics.
