Case study · NaschNatur GmbH

Four Things That Used to Be Done by Hand

Newsletters, customer support, creative ideation and invoice processing. I found the workflows worth automating, designed the systems, wrote the Claude Skills and agents, tested and refined them, and got the rest of the team using them.

The method, before the four

None of this was prompting. Each one went through the same loop: identify a workflow that suited automation, design the system, write the Skill or agent, specify its instructions and logic, test it, run iterations, refine against the results, integrate it into how the company actually worked, and then teach colleagues to use it.

That last step is the one that decides whether any of it survives you leaving. An automation only its author can operate isn't much use to the people left with it.

1 · Newsletter

Draft generatedwas manualClaude writes the email draft in the brand's voice

Filed into Notionwas manualThe draft lands automatically where the team already works

Human approvalkeptA person reads it and says yes. Deliberately not automated.

Pushed to Klaviyowas manualThe approved newsletter transfers across on its own

Scheduledwas manualQueued for send without anyone touching the platform

Manual on both ends before. The replacement automates everything around a human who still reads it and says yes.

2 · Customer support

20–35support messages a day
5–15minutes per reply — some quick, some needing input from elsewhere
2–9hof writing a day, landing on one person

A quiet day was a couple of hours. A bad one was longer than the working day it sat inside. I implemented Resolvia: set it up, trained the model on the company's own information, and prepared it to draft the responses — not to remove the person, but to stop them starting from a blank reply box thirty times a day.

Figures checked with the colleague who was doing the replying, rather than estimated from memory. The spread is wide because some questions answer themselves and some need input from elsewhere first.

3 · Creative ideation

An AI-supported ideation system that analyses current social trends and the account's niche, and uses both to generate relevant creative concepts — automatically, on a weekly cycle. Ideation stopped starting from a blank page and started from continuous evidence.

This is the same machinery that feeds the pipeline in The Content Engine. Built once, used in two places, which is the point of building it properly rather than as a one-off prompt.

4 · Invoice processing

Watch the mailboxwas manualMonitors the invoice inbox for anything new

Detect and readwas manualIdentifies incoming invoices and extracts their details

Storewas manualWrites the invoice information into a database

Check against paidwas manualWorks out whether this one has already been settled

Fill the payment formwas manualIf unpaid, completes the required form automatically

File itwas manualPuts the prepared document where it belongs

Notifywas manualTells the responsible person it's ready for them

The most technically involved of the four, and the furthest from anything in my job title. Every step above had been done by hand.

Nobody asked me to look at invoicing. I sat near enough to see how much of it was manual, thought it looked solvable, and asked whether I could try. That's roughly how most of the work on this page started.

How this work came about

None of it was in the job description. It came from sitting close enough to the work to notice which parts were being repeated by hand, and having enough room to try something about it.

Which is also why the content work and the systems work don't feel like two separate careers to me. It's the same habit pointed at different problems: noticing something done by hand thirty times a week and asking whether it has to be.

Next: The Operating System →