Creating a proposal in Ignition takes me about twenty clicks. Get in, add the person, set the expiry and the term, choose the service, amend the service, and on it goes. So when I spoke at The Firm's AI in Practice Summit on 3 June 2026, the thing I most wanted to show was what happens when you hand that job to AI instead: one prompt, and a branded, structured proposal comes back in a couple of minutes. If you prefer to watch the session in full, the recording is here.
I had a head start. When Ignition released its MCP about a month earlier, I happened to be the first person to use it: I was in Arizona, sitting next to the product manager, and he asked if I wanted a go. I spent that week running it against real work and was blown away.
So what is an MCP? It's what makes this possible: a connector between Claude and the software you want it to act on, far more native than a traditional API, and simple enough for a non-developer like me to set up. Where an API is a single narrow pathway, the MCP lets the AI pull whatever it's been given access to into one context window, reason over it, then act on the other side. It's the same mechanism Vinyl is using to build its Ignition integration.
Use the MCP for conditional price rises, not blanket ones
The most seasonal use case is repricing. Every June we renew and lift prices for the new financial year, and I'll be honest: for a flat across-the-board increase you don't need AI at all. Ignition's native renewals feature is faster and safer. The MCP earns its place when the rule is conditional. If something like Xero subscriptions shouldn't rise by the same 7.5% as everything else, and instead needs the service-library price effective 1 July, that's fiddly by hand and natural for the AI. I point it at a client, give it the logic, and it works through each proposal, rounds as instructed, and leaves everything as a draft to check. The same prompt would reprice every self-managed super fund in one run.
Updating 300 proposals, and the cost lesson
There was a second job hiding behind those renewals. Across more than 300 proposals, my engagement terms were out of date: nothing on the AML rules landing 1 July, and nothing on our use of AI. By hand, that's a couple of thousand clicks. So I tested the MCP on one or two proposals first, confirmed the new terms pulled through, then let it run across the rest.
It worked, but it taught me something about cost. In about three minutes, that one bulk run chewed through sixty per cent of my Claude usage. The culprit was running it on Opus, Claude's most capable and most expensive model. My fix has become a rule of thumb: I develop the prompt on Opus, where the thinking matters, then run the bulk work on a cheaper, faster model like Haiku. End to end it took a couple of hours, but only twenty minutes of my real attention.
From a meeting transcript to a draft proposal
This next one is my favourite, because I used to do it the hard way. Years ago, the only way I could turn a sales call into a proposal was a manual relay: pull the transcript out of Fireflies, paste it into ChatGPT for a scope of work, drop that into a Google Doc, then rebuild the whole thing by hand in Ignition. Now the relay collapses. In the demo, a meeting transcript and one prompt were enough for Claude to pull my Ignition services and templates, find the existing client, create three services and produce a draft six-month proposal at $9,750 a month, in three to four minutes. From mid-June, Vinyl's Ignition integration will do the same behind one button: review the draft in Vinyl, then raise it in Ignition or send it.
The important part is that these always land as drafts, never sent automatically. I think of the AI as an assistant I've delegated a job to, who brings me back a draft to review.
Claude as the glue between systems
That's the thread running through all of it: I use Claude as the glue between systems that don't naturally talk to each other. A transcript in Vinyl, services and pricing in Ignition, records in Karbon, none of which it has to log into separately. It isn't a new idea. I leaned on Zapier years ago to bridge tools that couldn't speak to one another, and it still earns its keep for trigger-based jobs, like pushing a finished Vinyl meeting into the right Slack channel. Claude is what I reach for when the task is messier: here's a pile of unstructured data, go and do something with it.
A word on client data, because people always ask. My setup is deliberate: a team plan with model training turned off, control over location data, and I redact anything highly sensitive before it goes near the AI.
What's next: book, meet, propose, get paid
The roadmap points at a single connected workflow. Email integration goes live next week, Vinyl's Ignition integration by mid-June, and calendar scheduling at the end of June, across all Vinyl tiers. Strung together, that's book the meeting, have it, raise the proposal, send the email and get paid, without leaving the tools you already use, and with a human checking the drafts at each step.
For all the talk of AI, that's the part that actually excites me. The clicking is becoming optional. Vinyl's Ignition integration goes one further, drafting the proposal in one click straight from your meeting. It's coming soon, so join the waitlist to be first when it's live.
