Why I don't believe the promise 'AI increases productivity by 300%'

You've seen headlines like "AI boosts productivity 300%" or "Automation saves 80% of your time". I do AI automation every day, and I don't believe those numbers. Not because AI isn't useful, but because numbers like that usually don't say what's being measured, who it's being measured for, or compared to what.

A number without context doesn't help you

"300% increase" is an increase from what? Someone who used to spend three hours drafting emails each day and now spends one hour is indeed three times faster at email drafting. But that doesn't mean the business produces three times more. Real productivity depends on what you do with the time you get back.

That's why I don't throw numbers like that around on this website. When I work with clients, I measure with their specific things: which tasks used to be done manually and now run on their own, how much time we save each week, where errors went down.

What AI does well and what it doesn't

Based on what I use every day:

• Does well: draft writing, summarizing, categorizing, moving data between systems, answering repetitive questions based on existing documents, writing and editing code under supervision.
• Doesn't do well: tasks that need judgment in new situations, tasks where one mistake costs money, answers that need accurate numbers AI has no way to verify.

AI can give you wrong answers very confidently. So every system I build has a place for people to check before the result gets sent out.

How to evaluate an AI tool yourself

Before I pay for an AI tool or service, I usually ask:

1. Does it solve something on my to-do list? If not, no need to buy yet.
2. Is there a trial with my actual data? Demos always look better than reality.
3. When it gets something wrong, how do I know? A good tool lets you review and fix it.
4. Where does my data go? With customer information, this matters more than any feature.
5. If I stop using it, can I get my data out?

The only promise I make

I don't promise numbers. I promise a way of working: start from your actual job, pick the right things to automate, build until it works, and guide you to run it yourself. If there's something AI doesn't do well yet, I tell you straight.

If you want to know whether your work has things worth handing to AI, leave a few lines on the contact page. I read and reply to every message myself.