Anyone else letting an AI run their book ads? sharing my numbers, they're not pretty
So about 6 weeks ago I did something probably stupid and gave an AI agent actual control of my amazon ads account. Not the "chatgpt give me keywords" thing, I mean it connects to the Ads API on a schedule, pulls the reports, changes bids, adds negatives, pauses stuff, and I dont approve anything. It just does it and leaves me a log.
Some background, I publish non fiction under a few pen names, around 10 titles. I was spending stupid amounts of time in the ads console for a catalog that barely pays for itself, and my day job is technical so I figured, why am I doing this manually every three days when I could make something do it for me.
The numbers so far, because thats what I'd want to see first: June closed with $682 in royalties and $757 in ad spend. So minus 75 bucks. Before anyone says it, yes I know, but I was losing more than that before AND doing all the work myself, so I'm counting it as progress. Sort of. One single book (legal niche) makes about two thirds of my royalties, the rest of the catalog is basicaly decoration at this point.
The setup for whoever cares, its Claude running every 3 days against the amazon ads api. It has a config file with hard limits it cant touch, monthly spend cap, max bid change per cycle, max new campaigns per week etc. Everything it does gets written to a changelog with the reasoning. I also plugged in a free keyword tool from github (kdp-scout) so it has actual search data instead of making keywords up, which llms love to do.
Now the fun part. Early june it found the winning keywords in my best campaign and decided to "scale" them by duplicating them into a new campaign with higher bids. The new campaign outbid the original in every auction. My own campaigns were fighting each other and I had almost two days of dead sales before I understood what happened. I literally paid amazon extra money to compete against myself.
After that incident it got a hard rule, before creating any keyword it has to pull everything thats already enabled and dedupe, and if it duplicates a winner the bid caps at 80% of the original. Funny thing is last week that rule stopped it twice from doing the same thing again. Its like watching an employee develop scar tissue.
Other stuff it learned the hard way, ignore the last 2 days of data before cutting anything because amazon attribution lags and creates fake losers. Search terms that are entire book titles get rejected as keywords (too long), you have to target the ASIN instead. And pruning beats creating, its best cycles were 4-5 surgical changes, its worst were 30 tiny bid adjustments that did nothing.
Also had it manage a new launch in a completely different niche and that was a disaster, 1 copy in three weeks of paid traffic. Now theres a gate, no new book gets a single dollar of ads without keyword volume and competitor data first. Expensive lesson but ok.
What I havent figured out and where I'd genuinely apreciate input from people who've been doing this longer:
The agent is decent at not wasting money but ads dont fix a listing that doesnt convert. I'm getting a ton of clicks from "law firm" type searches that never convert because my book clearly speaks to individual lawyers, and no bid adjustment fixes that. Thats a description problem.
KENP. how do you people attribute page reads to ad spend without losing your mind, every calculation I do gives me a different breakeven.
International is rough. US works, UK barely, germany and spain were pure bleed so the agent hibernated them on its own (that was actually a good call). Trying canada now.
And the big one, would you let something like this touch prices or metadata, or is that insane? Right now anything editorial is proposal only, I execute manually.
If anyone else has wired scripts or an agent to the ads api for books I'd love to compare notes, what guardrails you needed, what you'd never delegate, etc. Not selling anything and not naming my books, this isnt a promo. Just want to know if I'm early or just wrong.
TL;DR: AI agent runs my amazon ads autonomously since mid may. Broke my catalog once competing against itself, learned some rules, got me from losing money to almost breakeven (-$75 last month), killed a bad launch fast. Ads cant fix weak listings though and I'm stuck at a ceiling. Looking for others doing the same.