Ask a better question than 'how do I replace headcount?'
Most operators asking about AI are asking the wrong question. They want to know how to cut people. The better question: how can AI amplify the results the team is already producing?
If someone is clocking in and stealing time, that's a different conversation. But for hardworking people, AI should take repetitive tasks off their plate, get them off the keyboard, and let them spend more time with clients. That's the win — a lighter day, better output, and a team that actually wants to be there.
I've been deep in AI for the last six to eight months. I've wasted time. I've wasted money. Every lesson below is one I paid for.
Keep AI internal. Keep humans customer-facing.
Contrary to what every AI vendor wants you to believe, people do not want to be sold to by AI. They don't want their support tickets solved by AI either. They want a human.
There's a framework I wrote down on a notecard called EPOCH — the human capabilities AI still can't replicate: Empathy, Presence, Opinion, Creativity, Hope. You can feel when a bot wrote the email. You can feel when a bot is 'giving an opinion.' That's why support bots and sales bots are underperforming expectations.
Where AI is fine with customers: appointment scheduling, rescheduling, order confirmations, shipping updates, reminders. Low-emotion, high-repetition stuff.
Where AI belongs: internal. Handling the repetitive tasks that look the same on run #1 and run #1,000. That's the sweet spot.
EPOCH — Empathy · Presence · Opinion · Creativity · Hope
Never automate a process that doesn't already work manually
This is the lesson I paid the most for. If a process isn't running successfully by hand, automating it just gives you a broken process that runs faster.
AI can't account for variables you never mapped. The moment reality drifts from what you expected, the system stalls or does something dumb. Most 'AI horror stories' — wiped databases, weird outputs, servers going sideways — come from people automating an idea instead of a proven workflow.
Build it manually. Prove it works. Then let AI replicate it.
The five-step algorithm before you automate anything
This is adapted from Elon Musk's 'algorithm' with a bit of Alex Hormozi's take layered in. Run any process through these five steps before you touch AI.
1. Question every requirement. Why does this step exist? Who said it had to happen this way? How do they know? Look at your onboarding — does it really need to be two hours and 30 pages? Every unnecessary step is friction on someone already going through the pain of change.
2. Delete what's unnecessary. Cut hard. If you delete something you actually needed, you'll remember and add it back. That's fine.
3. Optimize and simplify what's left. Clear and concise always beats complex and clever. Don't skip to this step — optimizing something that shouldn't exist is the biggest time-waster there is.
4. Speed it up. Tighten timelines, fix bottlenecks, improve delivery. Only after the process is clean.
5. Automate. Now you can hand it to AI with real prompts, real rules, and a real manual version to replicate. It won't go rogue if you gave it clear instructions — same as a new hire.
1. Question → 2. Delete → 3. Simplify → 4. Speed up → 5. Automate
Where AI actually amplifies a team: three pillars
Once you have clean manual processes, here's where AI earns its keep internally.
Acquisition & retention. Track churn and monthly closes together. If churn is 5% and you close 10 monthly, your steady state is 200 clients — you'll hit that number and stall, gaining 10 and losing 10 forever, without understanding why. AI can surface that math before it strangles growth.
Operational workflows. Track first point of satisfaction — the moment a client feels value, not necessarily the full outcome. Build handoff workflows so the next person on the account already knows the client's numbers, history, and goals. Nothing kills a sales process faster than the closer asking the same questions the setter already asked.
Data tracking. This is the one I love. Reverse-engineer any revenue goal down to leads needed. AI can run this loop and show you where to fix things before you spend a dollar on ads.
Reverse-engineering a $10K MRR goal
Here's the math worth automating. Say you want to add $10,000 in monthly recurring revenue.
At a $250 average order value, you need 40 new deals a month. At a 75% close rate, that's roughly 54 calls. At a 60% show rate, that's 89 bookings. At a 20% booking rate, that's 445 leads.
Once you see the number, you can decide: is the goal realistic, or do we need to fix the close rate, the show rate, or the booking rate before adding spend? Layer in cost-per-lead and ad spend and you'll know if the goal is achievable before you start.
A centralized hub — APIs, JSON, whatever plumbing you prefer — pulls these numbers into one place so you're not clicking through 15 subscriptions to figure out if you're winning. That's the internal AI stack worth building.
$10,000 ÷ $250 AOV = 40 deals → ÷0.75 close = 54 calls → ÷0.60 show = 89 bookings → ÷0.20 book = 445 leads
Amplify your team. Don't replace them.
AI isn't a headcount replacement strategy. It's a leverage tool for the people already generating your revenue. Point it at the repetitive, high-frequency, low-emotion work happening inside your business. Keep humans where humans matter — in front of clients, in creative work, in the moments that need empathy, presence, and opinion.
Run every process through the five steps before you hand it to AI. Prove it manually. Delete what shouldn't exist. Simplify what should. Then automate.
That's how you get a team that ships more, works less, and actually likes the job.