One of my consulting clients is a boutique PR agency. About 30 people, growing fast. Over the past few months I’ve been working with them on a bunch of different AI projects: automating their conference tracking, building reporter briefing tools, streamlining how they manage media lists, and setting up branded LinkedIn graphics workflows for their clients.
Today I want to tell you about my favorite one: the proposal generator.
The problem
This team runs a lot of new business meetings. They’re good at it. They land a meeting, the partners have a great conversation, and afterward the team needs to turn that into a polished proposal deck to send the prospect.
Here’s how that used to work:
An associate would gather all the notes from the meeting (their own, the partners’, whatever context they had on the prospect) and then spend hours building a PowerPoint from scratch. They’d pull from past proposals and a bank of existing slides, but every deck still required significant manual assembly.
An external design firm was involved for the final polish, which added cost and time on top of that.
The thing is, most of these proposals are structurally similar to each other. There are standard slides that describe how the firm works, their track record, their team. Those don’t need to be reinvented every time. But there are also sections that need to be highly tailored to the prospect: the situation overview, the communications strategy, the media targets, the pricing. Those parts need to be highly strategic.
But most of the associate’s time wasn’t going to the strategic thinking. It was going to the mechanics. Building slides, formatting decks, assembling the same pieces in a slightly different order. The kind of work that feels productive but isn’t really where a smart person’s time should go.
So we built a proposal generator in Claude.
What it does
The system lives inside a Claude Project. It has the firm’s brand guidelines, a full catalog of their slide types, reference decks, and detailed instructions for how proposals should be structured.
Someone on the team pastes in their meeting notes and whatever they know about the prospect. Claude generates a written outline for the entire deck.
It maps out:
which standard template slides to pull in and why
where it needs to generate new custom sections based on this specific prospect’s context.
Then, Claude:
writes the situation overview
maps out a communications approach
identifies relevant media targets
suggests case studies to include
and drafts pricing
That outline comes back in about two minutes. The team can review it, make changes, and adjust the strategy. As soon as they’re happy with it, they hit go and Claude generates the actual PowerPoint presentation. Fully branded, properly structured, ready for review in three minutes.
Sanitized example of a powerpoint created by this system
The part that surprised me
I expected the time savings to be the headline. And they are. We’re talking about a process that went from many hours to minutes. But what actually impressed me was the quality of the strategic sections.
The situation overview doesn’t just summarize the meeting notes. It synthesizes them into an analytical narrative about the prospect’s competitive landscape, their communications challenges, and why this firm is the right fit. When we tested it against a real prospect, it pulled in context about the sector, identified the right peer comparisons, and priced the engagement in line with what the team would have proposed themselves.
It’s not ready to send to the client as-is. It still needs a human reviewing the strategy, adjusting the tone, adding judgment calls. But it gets the team to a strong first draft faster than they thought was possible. In their branding. In their voice. With their slide structure.
Why I’m sharing this
I go back and forth on how much client work to share in this newsletter. But I think this one is worth talking about because the pattern is so common.
Almost every company I work with has some version of this problem: a repeatable deliverable that eats up way more time than it should because no one has stopped to systematize it. Proposals, reports, client onboarding docs, pitch decks. The specifics change but the shape is the same.
The tools to fix this exist right now. Claude Projects, custom instructions, brand guidelines as context. None of this requires code or a technical team. It requires someone who understands your workflows well enough to design the right system, and knows how to use Claude.
If that sounds like something your team needs, I’d love to hear about it. Hit reply and tell me what you’re working on, even if it’s just “we have this one process that’s killing us.” That’s usually the right place to start.
Welcome back to Watch Me AI If you’re new here, welcome. Watch Me AI is where I share practical ways to use AI in your work and life. Here are a couple of past issues to get you started: It's time to switch to Claude and "How can I make more money?". Anthropic launched Claude Design this past Friday. It’s a new mode inside Claude built for visual work like slides, posters, marketing assets, app mock-ups, etc. I’ve been in it for four days. Here’s what I’ve been making, what’s working, and...
Welcome back to Watch Me AI If you’re new here, welcome. Watch Me AI is where I share practical ways to use AI in your work and life. Here are a couple of past issues to get you started: It's time to switch to Claude and "How can I make more money?". A lot of you replied to last week’s LinkedIn post asking for the actual recipe I use to turn meetings into LinkedIn posts. So I'm sharing! I run this recipe after almost every meeting. My internal rule: if there’s even a remote chance I said...
Welcome back to Watch Me AI If you’re new here, welcome. Watch Me AI is where I share practical ways to use AI in your work and life. Here are a couple of past issues to get you started: It's time to switch to Claude and "How can I make more money?". Last summer I got on a Zoom call with an old friend from high school. We hadn’t talked in a while. She knew I’d been working in AI, and she had questions. We started with the usuals: best AI tools for work, how to learn AI, etc., but what's...