I un-automated my client onboarding last February. The whole pipeline. I'd spent two months building it: intake form to Notion, Notion to a Loom recording, Loom to a kickoff email that pre-filled itself with the client's answers and went out the next morning. It was beautiful. I'd built exactly what I sell to other people.
Then a client named Liana said something on a discovery call that made me kill the whole thing.
She said, "I almost didn't book this call. Your intake form felt like applying for a loan."
That's the line. That's why I un-automated it.
I'm an AI consultant. I help small businesses install workflow automation. And the most useful thing I can tell you about automation is which parts of my own business I deliberately don't automate — because I learned the hard way that "efficient" and "right" aren't the same word.
If automation removes the part of the work that builds trust, it's not saving you time. It's just speeding up the loss of clients you haven't noticed leaving yet.
The thing that made me undo it
The intake form had 14 questions. It needed 14, technically. By the time someone hit submit, I knew their business model, team size, AI history, biggest blocker, three top tasks they wanted to offload, and the budget range they'd already vetted with their accountant.
I never had to ask any of it on the call.
Which sounded great. I had a 22-minute discovery call instead of a 45-minute one. I closed faster. I onboarded faster.
But here's what I missed: the part where someone tells you their business in their own words. The mess. The "well, I started doing this because my dad ran a..." The detour into the divorce that gave them the time. The CEO who was bored explaining her own company to herself before she met me, and got more interesting once she had to.
That's the part I was efficient-ing away.
So I yanked the form. Now there are three questions on the contact page: name, email, what's going on. The first 15 minutes of every call is them telling me the messy version. I take notes by hand. I lose maybe three hours a week to it. I'm fine with that.
What I un-automated (specifically)
Three things came out:
The intake form. Now a single text field where someone can write three sentences or three paragraphs. They tell me what they think they need. I respond like a person.
The auto-generated kickoff email. It used to fill in their answers from the form. Looked thoughtful. Read like a mail merge. Now I write each one. Takes 12 minutes. The reply rate doubled.
The proposal builder. I had a tool that pulled from a price block library and assembled a PDF. Looked clean. Felt like a quote from a contractor I didn't know. Now I write proposals in a Google Doc, share it live, and we edit together on a call. The close rate went up by something I'd want to tell you in a way that sounds humble, but the truth is: a lot.
The rule I follow now
I don't automate the parts where the job is the relationship.
That's the rule. It's vague on purpose. The specifics are different for every business. For me, the relationship-building parts are:
- The first conversation
- The proposal
- Anything I'd want a client to remember
Automate the parts where the job is logistics. For me that's:
- Calendar scheduling
- Invoice generation
- Onboarding reminders that say "your kickoff is tomorrow at 2pm"
- The Loom video I record once and reuse for the same five FAQs
- Workflows that post a deliverable summary into a client channel
The test I use: if the automation goes wrong, does the client notice — or does my admin notice? If it's the admin, automate it. If it's the client, do it by hand.
What I still automate
This is the part where I look like a hypocrite, except I'm not.
Most of my back office is automated. I have a workflow that runs every Monday and pulls last week's project hours into a one-page summary I review with my coffee. I have a tagging system that sorts every contact form submission into one of four buckets and routes my reply template accordingly — I write the reply, the system queues the right starting point. I have a process that drafts case study writeups from the deliverables I've already created.
What's the difference between those and the intake form I killed?
Nobody outside my business interacts with them.
When a workflow runs on the inside of my business — between me and my tools — automate aggressively. When it runs on the outside, where a client is the user, slow down. The savings of two minutes are not worth the loss of someone feeling like they're talking to a person.
Internal-facing automations buy you time. Client-facing automations spend something you can't get back.
How to decide for yourself
The framework is simple enough that you can do it in 20 minutes. Make two columns on a piece of paper:
Column 1 — Internal: What do you do every week that nobody outside your business sees? Bookkeeping. Reporting. File organization. Reminders to yourself. Inventory updates. Recurring social posts that don't need to read like you.
Column 2 — Client-facing: What do you do that touches a client or prospect directly? First emails. Proposals. Kickoff calls. Check-ins. Project status updates. Birthdays. Apologies.
Automate column 1 first. Aggressively. Use a custom GPT for internal knowledge work, set up workflows for the boring repeating stuff, and put as much as you can on rails.
Don't touch column 2 until you've spent a year on column 1 and you can say specifically what part of a client interaction is the relationship-building part — and what part is just logistics. Most people get it wrong on their first pass. I did.
If you want help figuring out which is which for your specific business, the AI Readiness Assessment is built to clarify that. It's two minutes. It tells you what to automate first and what's worth keeping by hand.
Frequently asked questions
Should I automate everything I can?
No. Automate the parts where nobody on the outside is the user. Slow down on anything client-facing. The rule of thumb: if the automation breaking would embarrass you, do it by hand.
How do I know what's worth automating?
The honest test is whether you'd be okay with a stranger doing it. If a virtual assistant could handle the task without you reviewing every output, automate it. If you'd want to read every output before it goes anywhere, that's a sign the task needs you in it. Automation will save you minutes but cost you something harder to measure.
What I'd actually do this week
Pick one client-facing thing you currently automate that you suspect is killing the warmth in your business. Replace it with a 12-minute manual version this week. Track replies, opens, or whatever the success metric is. If it doesn't improve, automate it back. If it does — and it usually does — leave it manual and look for the next one.
The whole point of automation is to free up time for the work that needs you. If automation is doing the work that needs you, you've got it backwards.