Documenting spree

If you throw GPT at your client work without teaching it your standards, you’re just scaling mediocrity.

I used to cringe when people asked about our company size. I’d say “less than 10” and immediately explain why we could scale if needed. Pointless.

We were already experimenting with AI tools when Sam Altman predicted a one-billion-one-person company. The new business world isn’t about adding headcount anymore; it’s about how AI adopters can hyper-scale. We stopped the hiring plan for two full-time data analysts and kept one part-time.

Then we pushed hard on AI. ChatGPT, Claude, Gemini. Contracts, meeting notes, analysis drafts, code snippets, slides, and reporting. The tools were fast, cheap, and seductive. Too seductive.

At some point, I used GPT to partially draft a client deliverable. Sent it to my co-founder for review and got smashed with the expected feedback “This is GPT, too generic and missing REKOLT’s touch“

LLMs are like new hires. You wouldn’t send an intern to a client without onboarding and training. Why do it with AI?

To make AI more specific and relevant, they need to be fed with context and very precise instructions on our ways of working.

So I went on a documenting spree. And started formalizing every nuance of how we think and work. How we approach messy operational data. How we structure KPIs. How we look for AI use cases. How we think about margin bridges, cash forecasting, and anomaly detection. How we push beyond what was asked, because that’s where the value is.

Everyone contributed. Screen recordings. Voice notes. Process explanations. Retros of past projects. We built a corpus and integrated it into our AI workflows.

Gradually, the output changed. The AI stopped writing consulting clichés. It started reflecting how we speak and what we notice. It knew we always check operational drivers behind margin variance. That we always reconcile numbers between data sources. That we never use pie charts. That we always go for “all you can analyze” for every data set to come up with insights on the client’s business.

It’s still a work in progress. Many outputs still need rewriting. Some subtleties are hard to encode. But the direction is right. AI now helps us stay consistent across projects.

If you’re a solo consultant or a boutique consulting firm, here’s the takeaway: treat your AI as an analyst you just hired. Don’t let it write before it learns. Document your approach, your quirks, your frameworks, your dos and don’ts. Feed it real examples. It’s tedious, there were days when I spent as much time documenting as delivering, but it pays off.