How I Run My Music Business With AI, and What Broke
An independent artist's real AI setup for booking, posting, merch and royalties, what each part does, and the mistakes that taught me where the limits are.
I make bass and electronic music as The Great Medicine Show. I have released more than 100 songs, and in 2025 they reached 1.20 million streams from 226,000 listeners in 166 countries. I am also the booking agent, the social media manager, the merch department, the bookkeeper and the person who registers every song for royalties.
For a long time that second list ate the first one. So I built an AI operator to run the business side, and this is an honest account of what it does, what it does not, and what went wrong on the way.
If you are an independent artist wondering whether AI can take the admin off your plate, the short answer is yes, with a few rules you should not skip.
The shape of it
Everything runs from one page on my laptop. I open it in the morning and it shows me the few things that need me. Behind that page are a handful of jobs that run on their own:
- Booking. It keeps my list of venues and festivals, drafts a personal letter to each one, and tracks who replied.
- Posting. It picks a moment from a song, cuts a short video, writes a caption, and queues it for Instagram, Facebook and Bluesky.
- Merch. It turns public domain artwork into garment designs and lists them in my store.
- Royalties. It turns my distributor's catalogue export into registrations for my performing rights organisation and flags songs that are released but not registered.
- Watching. A watchdog reads the log of what actually ran and flags anything that did not.
Most of these finish by putting a card in front of me: I say yes, fix it, or skip it. A few kinds, like festival applications, now go on their own, but only after a second agent checks every answer against my own files.
Booking: personal, small, and never a blast
The booking pipeline was the first thing I built and the one I rebuilt the most.
The tempting version is a mail merge to every venue you can find. Do not do it. Mass email from a small domain gets flagged, and once your domain is flagged your fan newsletter and your normal replies start landing in spam as well. You trade one bad week of outreach for months of damaged email.
What I do instead: my own letter for each kind of room, filled from my real credits, with a daily cap. A dance night gets a different set and a different pitch than a festival. When a reply comes in, the AI drafts an answer to what they actually said, and I read that before it goes.
The biggest lesson was about punctuation. I read one of the early drafts on screen and knew straight away it was written by a machine. The tell was long dashes where a person would use a comma. AI models love them and people almost never type them. My booking letters and captions are now checked for that before they go out, because a letter that reads as automated is worse than no letter at all.
Posting without posting the same thing forever
Short video does not pay much directly, so I treat posting as the front door rather than the business. The goal is to be seen often without looking like a bot.
The system finds the strongest moment in a song, cuts a vertical clip on the beat, writes a caption in my voice, and queues it at a sensible time. Two rules stopped it from looking automated:
- Never the same picture under a new song. If a visual has been used, it is retired. People notice repetition faster than you would think.
- Check the batch, not just the post. Given similar instructions, an AI writes similar captions. One batch of my cards had five opening with nearly the same sentence. Read the week's openers side by side before anything is queued.
Royalties: the money already owed to you
This is the least exciting part and probably the most valuable. Every song you release can earn performance royalties, but only if it is registered as a composition with your performing rights organisation, separately from the recording your distributor handles.
My system reads my distributor's catalogue export and turns it into registrations, field by field, and flags the songs that are released but not registered. The traps are mostly in the forms: the composition and the recording are different things, writer splits have to add up, and some fields silently cost you money when they are left blank. If you have a back catalogue and have never done this, it is worth an afternoon before anything else on this list.
Merch from art nobody owns
Old artwork whose copyright has expired is free to print. The hard part is being sure. My merch tool only uses images whose public domain status is stated by the museum that holds them, and it refuses anything where that is unclear. Then it lays the art onto a garment, checks that no name or detail is cut off at the seams, and drafts a product.
The day I found out it had stopped
One night a mail check froze. The morning routine waited on it, and for four days none of the daily jobs ran. The dashboard looked completely normal, because opening it ran a catch up that quietly covered the gap.
The fix was a timeout on the mail check. The lesson was bigger: judge a system by the log of what actually ran, not by whether the page looks right. If you build anything like this, put that check in early. A system that fails silently is worse than no system, because you stop checking.
What I would tell another artist
Start with the one job you dread most and that follows the same steps every time. For most artists that is booking emails or royalty registration. Automate the drafting, keep the sending in your own hands, and add a check that tells you when it stops.
You do not need to write code to get most of this. I teach the setup in a small group AI workflow workshop, and I can set it up with you one to one on your own catalogue and calendar. If you run a business outside music, the same approach is written up in what an AI dashboard actually does for a small business.
AI for small business · October 21, 2026