By Serge Osaulenko

How I work: diagnostic, plan, implementation.

I work in three steps: a diagnostic, a plan, and an implementation. They are the three steps of the consulting method I built for my own work. You can stop after any step and keep what you have.

  1. 1Diagnostic.

    Goes in: your answers, a look at your business, our call.

    I look at how your business runs and find where AI would save the most time or money, and where it wouldn’t.

    What you get: the written diagnostic. It starts with the answer, then the risks and how sure I am.

    Up to two rounds of revisions are included.

    See how a diagnostic works, step by step

    Book a diagnostic

  2. 2Plan: $2,500.

    Goes in: the diagnostic you accepted.

    The changes worth making, in order: the tools, the steps, the cost, and how we’ll know it works.

    Like the diagnostic, it opens with a written brief you approve, listing each question the plan has to answer.

    What you get: a summary first, then the steps in order.

    Up to two rounds of revisions are included.

    Keep the plan, build it with me, or take the plan to anyone you choose, including another firm for the implementation.

    Ask about a plan

  3. 3Implementation: quoted per project.

    Goes in: the plan.

    I build it with you, on accounts in your name.

    What you get: what I built, on your accounts, with a go-live checklist and a list of the upkeep it will need.

    For implementation, I bring in specialists I trust in your industry when the job needs it.

    Talk about a build

What gets checked on every build.

In my method, these checks apply to every build, whoever does the building:

  • Every step that can’t be undone is checked on its own before it happens.
  • Your old way of working stays available until the new set-up has taken over.
  • It’s checked on your real account, live. If something can only be checked on a test copy, I tell you so.
  • Anything set to run on a schedule counts as working only after it has run for real once.
  • Anything that collects leads or personal details is tested end to end, including the message that tells you a new one came in.
  • Anything that has to keep running comes with an alarm, and I test that the alarm reaches you.

How the work is checked.

In my method, every deliverable is checked by an independent AI reviewer: a fresh AI session that didn’t write the work. It looks for what’s wrong and asks: would you already know this? If so, it isn’t good enough.

Before anything reaches you, one last AI reviewer reads it the way you will, checks it against your own words and checks every company or person it names. A problem sends the work back to be fixed.

AI researchers and reviewers do much of the work. I approve everything that reaches you, and if I find something that would change the answer, I tell you before the final write-up.

When something goes wrong in my own work, I fix the method, so the fix carries into the next project.

A fresh AI session grades it

  • Does what was askedPass
  • Checked, not assumedFail: back with the findingsPass
  • Ready for the clientPass

Pass: forward

A last AI reviewer
reads it the way the client will
The client
reads it and agrees or pushes back
The top of a morning brief on a phone: each part closed with its count, and the first deal open.
An example morning. Every name is invented.

Proven on my own businesses first.

I build AI systems for real businesses, and I prove them on my own companies first. I rebuilt a real estate website with AI, and I use the AI set-up I built for real estate agents on my own listings.

My own systems are built the same way: AI agents I set up built them, reviewed them and now maintain them, with an independent AI reviewer checking the work.

What stays yours.

What I build runs on accounts in your name. That’s how a forward deployed engineer works at the big AI companies; I do it at the size of your business.

Start with a diagnostic See our work