For almost ten years, PRIMECUT has sent its proposals through the same paid platform. It does the job: it packages our recommendations, sends them to prospective clients, and collects signatures.

Somewhere along the way, I realized I was adapting our process to the software instead of the other way around.

Every proposal we write starts with a conversation. A prospective client tells us about their business, what’s getting in the way, and what they want to happen next. Turning that into a proposal meant squeezing it into a fairly rigid document workflow, then revisiting the same service descriptions, deliverables, and pricing details, proposal after proposal.

So I decided to find out how quickly I could build something that fit us better.

A plan from ChatGPT, code from Claude

I started with a long conversation with ChatGPT. We worked out what the tool needed to do, what goes into a PRIMECUT proposal, and how a proposal should move from the first discovery notes to a signed agreement.

That conversation became a plan, and I handed the plan to Claude. Claude helped refine the application and wrote the code. My job was making the decisions: what belonged in a proposal, what the AI was allowed to write, and who had to approve what.

We had a working prototype within a day. Adding electronic signatures took another couple of hours. It runs on a login-protected PRIMECUT subdomain.

That speed changed how I think about tools like this. If a working version takes a day, building around your own process stops being a big bet, and most of the effort goes into deciding what the tool should do.

Our old proposals became the library

The most useful thing I gave Claude was a stack of proposals PRIMECUT had already written.

Years of proposals hold a lot of decisions: how we describe each service, what it includes, how we run a project, what’s usually out of scope, and the contract terms we’ve refined along the way. Claude used those examples to help build an internal knowledge library.

Every new proposal now starts from that library, instead of us revisiting the same material every time.

What the AI writes, and what we decide

A new proposal starts with our discovery notes. From those, the AI drafts the parts that should be different for every client: the strategy, the deliverables, how our process applies to this project, a sitemap, and anything we should spell out as not included. It also lists the assumptions it made and the questions we should answer before anything goes out.

The final decisions stay with us.

Prices are suggested from what’s in our library, but every line item stays unapproved until someone at PRIMECUT approves the amount. Our contract terms come straight from the library, word for word, and the tool blocks any proposal where that language has been changed.

Before a proposal can be exported, it gets one final review. If anyone edits the content or a price after that, the review no longer counts, and the proposal has to be reviewed again.

There’s also an optional second AI pass that reads a finished draft and flags contradictions, like a deliverable that conflicts with something we’ve excluded. It doesn’t change anything. A person decides what to do with each note.

The AI writes what should change from client to client. What we commit to, and what it costs, stays with us.

Signing with SignWell

Writing the proposal was only half of what the old platform did. It also collected signatures, so the new tool had to as well.

I connected it to SignWell’s API. Once a proposal has passed its final review, we send it for signature from inside the tool. SignWell’s free API allowance covers 25 documents a month on our current plan, which is comfortably more than we send.

  1. Ready to sendEvery price approved and the final review done.
  2. Awaiting PRIMECUT signatureSignWell emails us to sign.
  3. Awaiting client signatureThe client gets an email, then reminders.
  4. SignedThe completed PDF and its audit page are saved with the proposal.
The statuses a proposal moves through once it’s sent from the tool.

Connecting the API was the quick part. Most of the work went into details the old platform had always handled for us:

  • Notifications. The signing emails come from PRIMECUT by name, with a message we can edit before sending. If an agreement sits unsigned, SignWell sends the client reminders.
  • The right document. The tool checks that the PDF we upload is the current, reviewed version of that proposal. An export from before a late edit, or another client’s file, gets refused.
  • The signed copy. When both sides have signed, the tool keeps the completed PDF and SignWell’s audit page with the proposal.

What actually changed

PRIMECUT now has a proposal tool built around how we work. We decide what goes in the library, what the AI is allowed to draft, who approves prices, and how signing works. When our services change, we can change the tool. When something feels awkward, we can fix it.

And it took about a day to get the first version working. That’s the part I keep thinking about. We’ve probably adapted to other software in quiet ways, and some of it might be a day’s work away from fitting us properly.

As for the old platform, the subscription is still active. It renews in January, and for the first time in almost ten years, I plan to cancel it before it does.

Have a workflow your team keeps working around?

Let’s talk about it
More articles & insights