Case study

Open for business in three and a half days.

No marketing background. Email, files and Office set up, a brand, document templates, this website and a free SOP. AI did a lot of the typing; a person reviewed every step.

The short version

If you use Word, Excel, Outlook and OneDrive at work, maybe SharePoint too, someone set all of that up for your company. On Monday I did that for mine. Then I needed to look like a business: colors, a logo, letterhead-ready templates, a website, and something worth giving away. By Thursday morning it was done.

This page isn’t here to say AI is magic. It’s here to show what working with it responsibly looks like, because that’s the same method I bring to quality work.

Day by day

Monday: the back office

  • Microsoft 365 Business Standard, set up as the company’s own environment, with me as its administrator.
  • Moved the domain’s DNS to Cloudflare, cleaning up records left over from a hobby internet service I ran in the early 2000s, and connected radkeland.com to Microsoft 365.
  • Email on the company’s own domain, plus a shared info@radkeland.com mailbox for the public, so personal addresses stay with the people I work with.

For the non-technical version: the email, calendar, file storage and Office apps you’d expect at any company, owned and run by the company instead of borrowed.

Tuesday: the brand

  • A color palette, picked from options with my wife and daughter as the review board.
  • The “R” logo I made in 2006, rebuilt as clean vector art from a modern typeface, generated by a script so it can be reproduced exactly.
  • Word and Excel templates with the brand built into the styles: headings with outline numbering, body text, and a branded table style.

Wednesday and Thursday: the website and the SOP

  • Decided what the site is for before building it: who it’s for, what each page does, and the four principles behind the name.
  • Built it as plain, fast pages: no tracking, no cookies, nothing flashy, readable on a phone. A small build step checks every link before anything is published.
  • Researched current FDA and EU guidance on AI, then wrote a free SOP for using AI in GMP work, reviewed over several rounds in Word with tracked changes.

What “a person in every step” looked like

Saying “human in the loop” is easy. These are the moments it actually mattered.

  • The person caught what the AI couldn’t feel. An early headline began “Not your fault.” It tested well on paper, but for someone who already blames themselves it lands on the sore spot. It changed.
  • The AI caught what the person couldn’t see. A table of contents heading had quietly taken a section number, shifting every section in the SOP by one and breaking every cross-reference. A logo rebuild came out subtly different because of a font version. A button color missed the accessibility contrast standard by a hair.
  • The AI was wrong, and a test settled it. It expected Word to add stray punctuation to automatic cross-references. A thirty-second test document in Word showed it didn’t, and the better design went in.
  • Every regulatory reference was checked at the source. The SOP quotes FDA’s warning letter from FDA’s own page, not from a summary of a summary.
  • Every document went through real review. Drafts came back as Word files; edits went in as tracked changes and comments; nothing was accepted without being read.

AI drafts. A qualified person verifies against the source, and owns the result.

That’s the whole method, and it’s the core of the SOP. It works the same way for a deviation narrative as it does for a website.

Want to work this way in your own quality system?

Using AI with review, verification and data controls built in is one of the things I help with.