The Setup
Solo operators—consultants, agencies, freelancers—live in an interesting trap. Their days are split between the 20% of work that actually requires their brain (strategy, decisions, relationships) and the 80% that’s just… repetitive.
Every client inquiry needs a proposal. Every new client needs an onboarding sequence. Every project needs a brief. Every week needs a report. Same structure. Same questions. Different names.
One client was spending 15 hours a week on these tasks. Smart person. Trusted advisor. But she was hand-drafting proposals from the same template every time, manually copying client details into intake sequences, and compiling reports from five different tools by hand.
The irony: her business made money when she was with clients or doing strategy. Not when she was copying and pasting.
So we asked: what if we built workflows that handled the repetitive 80% and just asked her to approve the smart 20%?
The Build
We built four connected automation workflows using Zapier and ChatGPT. Every one had a human checkpoint built in.
Workflow 1: Proposal Generator Form submission comes in via her website → Zapier catches it → pulls the client’s project details → sends them to ChatGPT with a template prompt → ChatGPT generates a first-draft proposal with scope, timeline, and pricing → email lands in her inbox flagged “DRAFT - REVIEW BEFORE SENDING.” She reads, adjusts, and hits send. Five-minute task instead of 45.
Workflow 2: Client Onboarding Sequence New client says yes → Zapier updates a database → triggers an email sequence that goes to the client with welcome message, questionnaire, brand guidelines link, and first meeting agenda → copies that same sequence to her project management tool with all the dates pre-populated. She just confirms the order and tweaks the tone if she wants. Twenty minutes becomes five.
Workflow 3: Content Brief Generator Client intake form comes in → ChatGPT pulls the details → generates a full content brief (audience, voice, keywords, tone examples, do’s and don’ts) → drops it into a Google Doc with her comments section on. She reads once, adds her personality to it, and ships it. Forty minutes becomes fifteen.
Workflow 4: Weekly Reporting Every Friday, Zapier pulls data from her project management tool, time tracking software, and invoice app → feeds it all to ChatGPT → generates a summary report (hours logged, revenue this week, projects on track, blockers) → sends her a draft. She reads it, tweaks it if needed, and sends to clients. One hour of manual work disappears.
The Mess
The mess is the part nobody talks about.
First: AI output quality was erratic. Sometimes the proposals were great. Sometimes ChatGPT would invent a timeline that made no sense. Sometimes it’d suggest a price that was way off. The “AI hallucination” thing is real. We couldn’t just set it and forget it.
Second: her clients didn’t trust AI-generated first drafts. She’d send a proposal and someone would ask, “Did you actually write this?” That’s a relationship liability. So we built in transparency. The workflows never auto-sent anything. Every output went to her inbox as a draft for approval. But we also had to help her think about when to mention the AI assist to clients and when to keep it invisible. (Spoiler: most of the time, keep it invisible. But be honest if they ask.)
Third: Zapier’s AI integrations were genuinely buggy in 2023-2024. Workflows would hang. ChatGPT would time out. We had to build error handling for everything—fallback emails, manual override buttons, a dashboard to show which workflows were stuck.
Fourth: scope creep is real. Once she saw what Zapier could do, she wanted to automate everything. Social media posting. Invoicing. Lead scoring. We had to pump the brakes and say: “Start with these four. Get comfortable. Then expand.” Automation is only useful if you actually use it.
The Result
In the first month, she stopped spending 15 hours a week on repetitive tasks. Closer to 10. By month three, it was down to 5 hours on things that absolutely needed her approval or her voice.
More important: she noticed the boredom lifting. The workflows weren’t doing her job—they were doing the boring part of her job. She was still deciding on strategy, still refining proposals, still building relationships. But she was no longer spending three hours on a Tuesday hand-drafting a proposal structure she’d written a hundred times.
She also started trusting her own first drafts less. The AI output wasn’t perfect, but it was solid. It freed up mental space to think about the bigger thing each proposal represented instead of just getting the words down.
And her clients? They didn’t care whether the draft was AI-assisted. They cared that it was thoughtful, fast, and specific to them. The automation made her faster, not cheaper-feeling.
Revenue stayed the same. Hours went down. Margin went up.
The Takeaway
Automation isn’t about removing humans from work. It’s about removing busywork from humans so they can do the work that actually matters.
The workflows we built didn’t replace her judgment. They replaced the muscle memory. Proposal templates exist for a reason—they work. What was eating her week wasn’t the thinking. It was the typing.
The key insight: build human checkpoints into every automation. Don’t try to build a fully AI-powered workflow. Build a workflow where AI handles the first draft and you handle approval. It’s slower than full automation, but it’s actually usable.
Also: start specific. Don’t try to automate everything. Pick the one task that eats the most time and makes the least difference. (Usually it’s proposals. Or onboarding. Or reporting.) Solve that first. Then expand.
And be honest about the mess. AI hallucinations are real. Buggy integrations are real. Clients get weird about AI. You have to design for all of that.
How It’s Built
- Core workflow engine: Zapier (free tier escalated to pro as workflows grew)
- AI layer: ChatGPT API with custom prompts for each workflow type
- Data sources: Form submissions (Formstack), project management (Asana), time tracking (Toggl), invoicing (Stripe/Wave), email (Gmail)
- Approval layer: Email drafts flagged for review before send (SMTP via Zapier)
- Fallbacks: Error emails, manual override buttons, slack notifications if a workflow stalls
- Documentation: Prompt library in Notion (so she could tweak prompts without rebuilding workflows)
No code required. All low-code/no-code.
What Changed Because of This
She stopped thinking about her week in units of tasks. Started thinking about it in units of client value.
She also got curious about what else could be automated. Started looking at her calendar differently. Started asking better questions about which tasks were actually hard and which were just tedious.
The business itself didn’t change. But her energy did. That’s not a small thing.
Want to automate the boring parts? Book a Build Session — $350/90 minutes.