I automated 5 processes in my own company with AI. Here is exactly what and how

I don’t give talks about the future.

I run my own company with AI agents in five areas: quoting, accounting, marketing, reporting and my personal assistant. They run today, not “soon”.

What follows is not a tutorial. I’m not going to tell you which tool to buy, for two reasons. First: the tool is the least important decision in the whole process. Second: it will change anyway, and if all you learned were the buttons, you start over.

What stays is the method for choosing what is worth automating. That part transfers.

And to be straight with you from the start: what follows is my experience, from a small training and consulting company. It’s not a universal recipe. It’s a way of thinking, tested on my own skin, with my own money.

The five areas running in my company

Quoting. A training proposal has the same structure every time: what I understood from the conversation with the client, what I propose, how long it takes, what it costs, what stays in the company afterwards. The content changes. The structure never does. A process with a fixed shape and variable content is the definition of a good candidate. I decide what to propose. I no longer lose time on what the document looks like.

Accounting. The volume here is small, but the rhythm is relentless: every month the same documents, the same deadlines, the same rules. It isn’t hard work. It’s work that isn’t allowed to be forgotten. Exactly the kind of task a machine remembers better than I do — on condition that someone looks at the result before it goes any further. The signature stays mine.

Marketing. The repetitive part, not the creative part. The idea a piece of copy starts from stays mine; the variants, the adaptations, the rewrites for different channels — that’s where I gain time. AI has nothing to say in my place. It has something to write faster, once I’ve decided what I’m saying.

Reporting. The easiest process in a company to automate and the most neglected. A report is, at bottom, the same question asked every month against new data. People put it off because it’s boring, then take decisions without it. A report that writes itself is a report that actually gets read.

The personal assistant. The most recent one and, honestly, the one that changed my day the most. Before every meeting I get a brief: who the person is, what we discussed last time, what I promised. During the conversation the transcript is made. Afterwards, three separate things come out of the transcript: the decisions taken, the tasks with an owner and a deadline, and a draft follow-up email. The calendar lines up with whatever came out of it.

One clarification, because it matters: this one runs semi-automatically. A few steps I still trigger by hand. That’s not a shortcoming, it’s a normal stage — a workflow gets automated in steps, not all at once. The next step I’m working on is a phone app that lets me talk to it by voice.

The part that matters isn’t the transcript — that’s just raw material. It’s the fact that nothing promised in a conversation gets lost any more. Before, half of what came out of a meeting lived in my head until someone got round to writing it down. Now I don’t depend on my memory on a day with six meetings.

The test: four questions before I automate anything

I don’t start a single workflow until it passes all four.

1. Does it repeat? If you do something once a quarter, the automation costs you more than the work does. You automate what happens weekly or daily, not what happens rarely and matters a lot.

2. Does it have rules I can write down? If you can’t explain the process to a new employee in ten minutes, you can’t explain it to a machine either. A process you do “on instinct” isn’t ready to be automated. It’s ready to be clarified. The good news: half the value of an automation project comes from the fact that you have finally written down how you work.

3. Does it have volume? The saving comes from repetition multiplied by time. A process that takes five minutes but happens forty times a month is worth more than one that takes an hour and happens once.

4. What happens if it gets it wrong? If the answer is “we lose a client, a reputation or a legal obligation”, the workflow doesn’t ship without a human inside it. Automate first where a mistake shows up quickly and is cheap to repair.

Four yeses — you have a candidate. If a single one fails, repair the process before you automate it.

Because this is the most important rule in the whole article: AI doesn’t repair a bad process. It accelerates it. If your quotes are weak, you’ll have weak quotes faster and in greater numbers.

What I don’t automate

This list is just as important as the first one.

• The first conversation with the client. The diagnostic. That’s where I hear what doesn’t get written in a brief. Listening can’t be delegated.
• Any conversation where the outcome depends on what the person in front of you feels. Hard feedback. A tense negotiation. Bad news. Speed is not a virtue here.
• The decisions that commit the company. The price. The promise. The signature.
• The processes that look different every time. You automate them and end up spending more time repairing the workflow than you would have spent doing the job by hand.
• The process I don’t understand yet. If I can’t draw it on a sheet of paper, I don’t automate it. Full stop.

Whoever starts with the most spectacular part gives up fast. Start with the most boring part. That’s where the hours are.

Human validation is not a formality

Every workflow in my company has a point where it stops and waits for me.

Not because I don’t trust the technology. Because responsibility can’t be automated. When a quote goes out to a client, the signature is mine. When a report turns into a decision, the decision is mine. An agent has nothing to lose if it gets things wrong. I do.

The practical rule: the bigger the consequence, the earlier the human steps in.

An internal text — validation at the end, at a glance. A client quote — validation before it goes out, every time, no exceptions. A document with legal implications — validation both before and after.

Good automation doesn’t take the human out of the process. It moves him from where he wasn’t needed to where he really is.

And it has one more effect nobody anticipates: it forces you to look at your own work. When you validate the same thing dozens of times, you start to see what is actually wrong with your process.

How you count the gain: in hours, not in enthusiasm

The measurement is done before, not after. Otherwise you have nothing to compare against and you’re left with impressions.

The steps, in exactly this order:

1. Pick the process and time how long it takes now. Honestly, not from memory.
2. Count how many times it happens in a month.
3. Multiply. That’s your starting figure.
4. Once the workflow is running, time it again — including the validation time, which is work too.
5. Subtract.

What’s left is the gain. In hours. Not in “it changed our lives”.

And there’s one more question that almost nobody asks: what did you do with those hours? If you filled them with other small stuff, you gained nothing — you only changed the kind of small stuff. The hours you win need a destination decided in advance: more conversations with clients, a new product, or, very legitimately, less work in the evenings.

My rule: a workflow that doesn’t give me back measurable hours isn’t an automation, it’s a hobby. I shut it down.

“Fine, but how many hours do you actually gain?”

It’s the first question I get and it’s the one I refuse to answer with a pretty number.

Not because I don’t know what happens in my own company, but because the answer depends enormously on the specifics of the business and on the processes you pick. A company with a lot of repetitive volume gains something different from one working on large, rare projects. Anyone who gives you a universal figure isn’t selling you an automation, they’re selling you an illusion.

What I can give you is a serious benchmark. The Federal Reserve Bank of St. Louis measures this on the real working population of the United States, not on a salesperson’s promises. In February 2025 they published the number: people who use generative AI save on average 5.4% of their working hours — roughly 2.2 hours a week on a 40-hour week.

The more interesting part is another one, and it’s exactly the point of this article: the gain grows with discipline, not with enthusiasm. Among those who use AI daily, a third report savings of at least four hours a week. Among those who use it once a week — only one in ten gets there.

Two clarifications, so you read the numbers correctly. First: they measure individual use of AI by employees, not the automation of company processes with workflows and validation — which makes them more of a floor than a target. Second: they are an international average, not a promise for your company.

And the question that really matters isn’t “how many hours”, it’s what you do with them. In my experience, the time freed up goes automatically into higher-value activities — conversations with clients, thinking about the product, the things you postpone forever because “you don’t have time”. That’s the real interest earned. The hours saved on a report are worth nothing if you put them straight back into other small stuff.

Why I founded the program at Tekwill

Because I’ve seen too many AI courses that end with impressed people and zero change in the company on Monday morning.

In 2026 I founded “AI Automation for Business” at Tekwill — from manual process to automated workflow. Not theory about what the technology will one day be able to do. Your process, drawn, tested, switched on.

I also deliver AI Skills Training: your team learns to work with AI on its own real tasks — copy, quotes, analyses, reports. Not on examples from the internet. On what they have to do tomorrow.

And when a company wants the result rather than the knowledge, I do turnkey implementation: I come into the company, we pick three processes and automate them together with your team. Three, not thirty. Three processes that genuinely work are worth more than a forty-page digital transformation strategy.

What I teach in the course I use myself, every single day. That’s the only guarantee I can give.

If you have a process in your company that repeats and that annoys you every time, let’s talk for thirty minutes. You tell me what it looks like, I’ll tell you whether it’s worth automating or just repairing — and if it’s worth it, you get the quote afterwards. The diagnostic is free and commits you to nothing.