There's No Drop-In Solution: What Two Real Automations Taught Me About Working With People

TLDR: Automation works. I’ve watched it save companies hundreds of thousands of dollars and rescue people from spreadsheet drudgery they hated. But there’s no magic button. The wins came from sitting next to the person doing the work, asking dumb questions, and rebuilding the process with them, not for them. If you skip that part, you’re just shipping fancy software that nobody trusts.

The pitch you’ve heard

You’ve probably heard the pitch a dozen times by now. “AI will automate your business.” “Drop in our agent and watch the savings roll in.” It’s a fun story. It’s also not how this actually works.

I’ve shipped real automations for real customers. The savings were real. The math was real. But the thing that made them stick wasn’t the code. It was the time I spent with the human who was buried in the work, figuring out what was actually broken and what only looked broken.

Let me walk you through two of them.

Story 1: The opt-out flood

One customer was drowning. They were getting roughly 20,000 opt-out emails a year. Each one had to be read, logged in a spreadsheet, traced back to a user in the control panel, manually opted out, replied to with a non-template response, and then marked closed.

If you do the math, that’s about 9 minutes per email. Times 20,000. That’s 3,000 hours a year. At $40/hour fully loaded, you’re looking at $120,000 a year in pure labor on a task nobody wanted to do.

Then volume grew. To 100,000 a year. That’s 15,000 hours. That’s 7.5 full-time people on data entry. Not survivable.

Here’s the thing though. I didn’t walk in and say “I’ll write an automation.” I sat with the person doing the work. I asked her to show me the actual emails. Real ones. With the weird edge cases. The customer who replied in three different threads. The one who used the wrong email address. The one who was actually angry and needed a different reply tone.

The “automation” was the easy part. Understanding the messy, real-world middle was the whole job.

We shipped something that handled around 90% of those emails end to end. Read, identify, opt out in the control panel, generate the right reply, close the loop. The remaining 10% were the weird ones, and a human still saw those. That was on purpose.

The savings? At 100,000 a year, around $540,000 a year reallocated. But more importantly, the team could scale without scaling headcount.

Story 2: The Slack-and-spreadsheet license job

Different customer. One person was responsible for about 80 licenses across 20 customers. Her daily job looked like this: post a to-do listing into 20 different Slack rooms (one per customer), collect responses back from each room at the end of the day, update a master spreadsheet, write a weekly progress report, coordinate with sales on renewals.

That’s two to three hours of pure coordination work a day, just to keep the wheels turning. About 800 hours a year of a senior person doing rote movement of information.

And here’s what nobody outside her seat could see. She was missing renewals. Not because she was bad at her job. Because she ran out of bandwidth. Every missed renewal was a customer churn risk. On 80 licenses across 20 customers, that’s real money on the table.

Same approach. I sat with her. I watched her do the job for a couple of hours. I asked dumb questions. Why this Slack room and not that one? What happens when this customer doesn’t reply? How do you decide which renewal to chase first?

The answers weren’t in any spec doc. They were in her head.

The automation posted the daily to-dos, collected the responses, updated the spreadsheet, generated the weekly report, and surfaced renewal candidates to sales. She stopped being a Slack-and-spreadsheet operator. She started being an account manager, which is what her actual job title said.

The pattern

Look at both stories. The math is real. The savings are real. But the thing that made them work isn’t in the math.

The thing that made them work was the time I spent next to the person doing the job. Not the manager. Not the executive who signed the check. The person whose calendar was getting eaten alive.

Because here’s what I’ve learned: the person doing the work knows things nobody else knows. They know which edge case shows up on the third Tuesday of the month. They know which customer always needs a softer reply. They know the unwritten rule about which renewals to chase first. None of that is in a process doc. None of it is in a wiki. It’s in their head, and they’ve been carrying it around for years.

If you don’t extract that knowledge before you automate, you’re going to ship something that looks great in a demo and falls apart in week two. Then everybody decides automation doesn’t work. (It does. You just skipped the part that matters.)

What actually has to happen

So when someone says “drop in our agent,” I get a little nervous. Not because the agent can’t do the work. Because the work isn’t only the work.

Here’s what actually has to happen for an automation to stick:

Find the person who actually does the job. Not the person who manages the person. The person whose Friday afternoon is on the line.
Watch them do it. Not interview them about it. Watch them. People skip steps in interviews because the steps feel too obvious to mention. Those obvious steps are usually the ones that break the automation.
Redesign the process before you automate it. If the manual process is broken, automating it just gets you broken at scale. Don’t automate broken flows. Fix them first.
Leave a human in the loop where the judgment lives. Opt-outs that look angry? Human reviews those. Renewals where the customer’s been weird for two weeks? Human handles that one. The agent does the rote work. The human keeps the judgment.
Measure it in their language. Not “we processed 1,284 events.” That doesn’t mean anything to anyone. “We saved her three hours a day. We prevented a churn. We freed up a senior person to do senior work.” That’s the language that gets people to trust the thing.

So what?

If you’re a company looking at automation, the question isn’t “what tool should we buy.” The question is “whose job are we trying to make better, and have we actually sat with them yet?”

If you’re a person whose job is being eaten by rote work, the question is “what’s the one task I’d hand off first if a smart helper showed up tomorrow.” Start there. Just one. Build from there.

There’s no drop-in solution. There’s a partnership between the person who knows the work, the person who builds the automation, and a willingness to redesign the messy middle before you ship anything.

That’s the boring secret. That’s also why the wins are real when you actually do it.

Originally published on Daily AI Studio by Alfred Nutile, Jun 06, 2026.