Will Leatherman spent 45 minutes on a recent session building something most of us avoid because it sounds too hard. An automation that turns a sales transcript into a multi slide custom proposal deck. And it actually worked well enough to save him six hours per proposal. The workflow uses Relay to orchestrate everything. Pull a transcript from a Notion database, do some research on the company and person, have Claude analyze it all, then fill a templatized Google Slides deck with curly bracket variables. The output isn't perfect but going from blank page to 80% done in 15 minutes is kind of insane when you think about how long proposals normally take. Book a free workflow mapping call if you want to build something like this for your team.

What actually happened in the build

1. Template the deck first, automate second Will started by opening his existing proposal template and replacing all the custom parts with variables wrapped in curly brackets. Client name, executive summary body, market landscape header, objective one, all the way through pricing and deliverables. He said:

"I know a lot of the work is going to be in the prep of the file."

And he was right. Most of the 45 minutes was deciding what stays constant and what needs to be dynamic. If you're doing how to build automations with Relay for content or proposals, template design is 70% of the work. 2. Pull transcript and basic info from Notion The workflow starts with a manual trigger and grabs the sales call transcript. In a real setup this would auto sync after every call but for testing he just pasted it in. Then it searches his deal pipeline database in Notion to find company name, contact info, industry, deal size. Basic stuff you already have if you track deals anywhere. 3. Do research before writing Before filling any variables the workflow does LinkedIn lookup for the person on the call and uses an LLM with web search to pull competitive intel, market conditions, and content opportunities. This mirrors how GTM teams scale with AI agents by having the system do homework you'd normally spend an hour on manually. 4. One big LLM step fills all the variables Will used Claude Sonnet to take everything the transcript, the research, the LinkedIn bio and assign outputs for 30+ variables. He was explicit in the prompt:

"The output should only be these variables, nothing else."

Then the workflow copies the master Google Slides template, makes a new deck titled with the client name, and fills every placeholder. Done. 5. It worked but needs iteration The first test run took 15 minutes and spit out a real proposal. Formatting was messy, some fields were too long, the LinkedIn image didn't populate, and a few variables got skipped. But Will's reaction was honest:

"In no world do I think this will get me to a finished proposal. However, going in and opening it up and just making some edits is going to be a whole lot quicker."

He called out that the next version needs a separate thinking step before formatting, max character limits per field, and example references so the LLM knows what good executive summaries actually look like.

Key things to steal from this build

Start simple and add layers. Manual trigger, pasted transcript, basic variables. Once it works you can connect APIs and add conditional logic. Separate analysis from output. Right now the LLM does research, critical thinking, and formatting in one step. Will said that never works well. Split it so one step thinks and another step just fills the template. Lock your variable schema early. Every time you rename a placeholder in the deck the downstream steps break. Map everything up front or you'll waste time debugging. Add formatting rules per field. Tell the LLM if something should be bullets or paragraphs, how many words max, what tone to use. Otherwise it defaults to essay mode and blows past your slide real estate. Expect broken stuff on the first run. Missing images, fields that are too long, research steps that time out. That's normal. The goal is a rough draft fast then you tighten it in iteration two and three.

Why this matters if you do proposals a lot

Will said he used to spend a week on proposals even with a template. This workflow took two hours to build and will save him six hours per proposal. If you're doing three or four a week that's 24 hours back. And the same approach works for pitch decks, investor updates, sales one pagers, anything you template and customize repeatedly. You can find more examples and templates in our resources library or check out upcoming Catalyst webinars where we break down workflows like this live. Book a call to map out a proposal or content workflow for your team.