by Samantha Christensen
Content Strategist, E-N Computers
Journalism background, plus hands-on experience building AI writing workflows
As a managed IT services provider, we deal with a constant stream of technical documents — project charters, proposals, RFPs, just to name a few. So we did what everyone does. We got AI: First, Copilot bundled with Microsoft 365, then ChatGPT.
Work that was time consuming and fragmented got . . . worse.
Employees could churn out documents faster than senior staff could review or correct them. We didn’t open the bottleneck. We moved it onto our most expensive people.
“It was very harmful because it was causing rework,” said Ian MacRae, president of E-N Computers. “It was taking more senior management time to push back and to coach people.”
Even AI-savvy senior people who understood its limits were facing a black hole.
“It would run you around the barn,” MacRae said. “It was a lot of effort to get even a little bit of value out of it. And you might have just been better off starting with a clean head and blank sheet of paper.”
Today, the same company runs proposals in hours instead of weekends, produces project charters in 20 minutes instead of two weeks, and answers RFPs without hiring a freelance writer. Here’s what we changed, and the process we use when we do this for clients.
QUICK ANSWER:
Why did AI make our proposal writing worse, and what fixed it?
AI made our proposal writing worse until we stopped asking it to invent and started feeding it our own material — the recorded sales call, our pricing, and our best past proposals. Proposals now take hours instead of a weekend, project charters 20 minutes instead of two weeks, and RFP responses stay in-house instead of costing $500 a bid in freelance fees. The workflows worth automating are the ones that are repetitive, high-stakes, and stuck on your most expensive person.
Table of Contents
- The principal who writes proposals on Saturday
- Why the fix came from the marketing team
- How we built AI proposal and RFP writing that works
- What we put into AI — and what we don’t
- From skeptic to 20-minute charters
- Why the judgment matters more than the tool
- Want this done in your business?
In a hurry? Click here to download the premium content. (No email required.)
The principal who writes proposals on Saturday
Our fix is aimed at one person: the principal who is still writing the proposals himself.
In a lot of small and mid-sized firms — ours included — the owner is also the top salesperson. He runs the discovery call. He knows what got promised and what it should cost. He’s the only one who can price exceptions. So, the proposal comes to him on top of running his business, and he writes it on Saturday.
He’s capable and senior, and that’s exactly why he’s stuck.
Still writing your own proposals?
A five-minute check on whether AI would help your proposals and RFPs, and when it won’t. No email required.
Why the fix came from the marketing team
Our pains weren’t unique. Plenty of businesses drown in technical documents. Plenty have tried AI to speed things up and been frustrated by it.
What made the difference was the combination of expertise we brought to the problem: technical knowledge, user experience, journalism, sales, and AI itself.
And the answer came from the marketing team. Not because we knew AI better than the engineers, but because we’d been fighting AI slop the longest. Producing good, technically correct writing was our daily job, which meant we’d spent more time than anyone wrestling with the lack of accuracy and clarity from AI-assisted work.
When I heard our president was spending entire weekends on proposals and that we couldn’t keep up with RFP deadlines, AI seemed like the obvious answer — this time pointed at a specific job, with our own material behind it.
I went after the workflow with the biggest payoff: what was high-stakes, repetitive, hard to template, and bottlenecked on one expensive person. Proposals and RFPs were it.
How we built AI proposal and RFP writing that works
The early tools failed because they were working from nothing. We stopped asking AI to invent and started giving it what we already had: the customer’s own words, our pricing, our best past work, and the voice of our company owner.
Meeting transcripts. We already recorded sales calls in Fireflies, and Fireflies has AI built in. What we hadn’t done was make that record usable for anything beyond a summary. With a $20 Claude subscription, we connected the transcripts to the writing. Our senior people and account managers may forget a nuance or a custom price from one meeting to the next. The transcript never forgets, and now those details get pulled into the proposal automatically.
Proposals. Claude reads the sales call and produces a branded proposal. We built a guide to our voice out of our own writing and gave Claude past successful proposals for context. The setup also adapts pricing to current market standards and breaks out options at different price points.
RFPs. Any firm that bids on contracts knows the pain: every organization asks similar questions in a different order, with a different nuance. So, we had Claude build modules that explain the business — certifications and partnerships, staff bios, response times, monitoring, cybersecurity, and so forth — and it reorders and adapts them to match each RFP’s structure. That handles the 80% of boilerplate shared across bids. For the 20% unique to each one, the system stops and asks the person instead of guessing. We used to pay $500 or more per RFP in freelance writing fees. The writing now stays in-house with our sales rep, and the team no longer drops everything to hit a deadline.
Project charters. Account managers pull client pain points straight from the transcripts instead of asking Copilot to spin up an imaginary charter. Our voice guide rewrites technical steps into outcomes the customer can test. “Move DNS from legacy domain controller” becomes “staff can work from any location, with internet filtering in place whether they’re home or at the office.”
“The customer can’t hold us accountable for DNS services,” MacRae said. “They don’t know what that is or how to test it.”
Technical and account staff who aren’t comfortable writing now produce no-fluff documents in a fraction of the time.
What we put into AI — and what we don’t
Everything goes through a company account on tools we’ve vetted, with training on our data switched off. We use sales call transcripts, our own pricing, and our past proposals. We never use client credentials, security configurations, or anything inside a client’s compliance scope. A human (sometimes several) reviews every document before it goes out.
From skeptic to 20-minute charters
Ian is the principal I described above. He runs the sales calls, and he was writing the proposals. So, when I brought an AI framework to him, he was both the customer and the skeptic.
He almost rolled his eyes. He already had licenses for Copilot, ChatGPT, and Perplexity. And here was marketing asking for another $20 a month per person.
“I was pretty happy with ChatGPT,” MacRae said. “There was still a fair bit of tool babysitting, but it was a lot better than the 80% effort for 20% results I was getting before. And I thought, ‘what does the marketing team know about it?’ I figured this was probably a waste of my time, but I’ll give it a try.”
He saw the difference quickly. The new setup was like talking to a smart employee who kept the context of the business, sat in on every meeting, remembered everything that was said, and always had the latest pricing.
“I was immediately very impressed,” MacRae said.
He recently pushed out a complex charter for a dental office setup in 20 minutes — work that would normally take two weeks and pull in a technical account manager to spell out the details.
“This is so fast,” MacRae said. “Whatever quality I have with scoping, it’s taking that and building on it. It’s like I have a junior associate that’s in my head.”
It also flagged things he’d missed.
“It will sometimes catch mistakes that I make,” MacRae said. “Like, these numbers are a little happy. Or, your customer’s deadline is November 10th, but this project’s going to take nine months — how are you going to make that work?”
The numbers, so far:
- Proposals: entire weekends → a few hours
- Project charters: two weeks → 20 minutes
- RFP freelance fees: $500+ per bid→ $0
- RFP writing time: 10 hours →1 hour
One honest caveat on that last line. A good part of an RFP has nothing to do with writing — printing and binding copies, getting documents notarized, chasing signatures, meeting a physical delivery deadline still take time, and AI can’t help. What we cut was the writing, which was the part that ate senior time and required a freelancer.
Now we get back to prospects while they’re still interested, with proposals that are more accurate and priced with less guesswork. Prospects get multiple options at different prices, and they get them quickly, which lets them decide quickly.
“It keeps our team from getting way overinvested in a sale that’s never going to happen,” MacRae said.
The work also moved down the org chart, which was the point. Katie Stebor, our sales representative, now takes care of RFP responses.
“For a lot of business owners, they know a lot, but they can’t always communicate it to someone who doesn’t have their knowledge,” Stebor said. Now she runs the RFP through Claude, and Ian checks it.
“It takes something off Ian’s plate that I can do,” she said. “Ian’s reviewing it and only giving it a few tweaks.”
The process does not replace the person. MacRae described a recent case where he thought he’d asked Claude to review a meeting, but it never pulled the transcript and answered anyway, working from stale information.
“It’s only as good as it’s prompted,” he said. “You can go so fast that you start missing major things you already knew. You still need the human checks and balances.”
Why the judgment matters more than the tool
These days everybody’s selling an AI solution that promises to overhaul your business and cut your staff. We don’t believe in either one. The best application of AI is precise rather than comprehensive, and the best workflow makes your top people better instead of replacing them.
After months of use, “I haven’t seen it be a replacement,” MacRae said. “It’s a force multiplier — it lets your existing people do more.”
The highest return comes from putting it in the right hands. Automating a low-wage task saves you a low wage. The money is in amplifying the senior person whose judgment is hard to replace. In a company our size, that person is usually the owner.
“Give it to your chief rainmakers,” MacRae said — “the people who should know better when it’s putting out garbage.”
Stebor sees the same thing from the sales side, in a company where everyone is constantly covering for everyone else.
“In a small or mid-sized business, you’re helping departments you’re not even part of,” she said. “The lines get blurred with everything you’re doing in a day, which just means you’re doing more. Any way you can get more time back allows you to focus on your actual job.”
But that time only comes back when you set the tool up and teach it how you work.
“You talk to the AI like a new employee and train it on the way you do things,” MacRae said. “You don’t conform your business to the software.”
Stebor puts the same idea in her own terms. “You talk to it like a kid — you’ve got to train it, tell it exactly what you mean,” she said. “And then when the kid grows up, you’re so proud of it.”
Want this done in your business?
If you run the sales call and then write the proposal yourself on a Saturday, you’re who we start with. That’s where the payoff is biggest and shows up fastest.
Anyone can connect Claude to Fireflies. The hard part is knowing which workflow to point it at, and building it so your people use it instead of going around it. We’ve run IT for regulated clients, written the content, and sat on the sales calls.
Our own setup freed up the time to look at yours.
