End of Week 1 of AI

Here’s a quick high-level explanation of what this is about. I keep reading things about folks using AI bots or agents to replace people. Mostly about perceived threat, but some by folks trying to improve things. I wanted to see if I could use AI to help with my marketing.
I created three “bots”, (1) an adsbot that tracks my sales versus ad spend. (2) a igbot that would create posts for me, and (3) a growthbot that would look for opportunities.

adsbot

I received the permissions I was waiting on from Amazon and I can now see the ad spend. Amazon does not have an API for checking income from book sales, so I have to dump an excel from their web site every Sunday and upload it. adsbot sends me a daily update every morning (currently with the wrong dates due to a recent change I made for other reasons… need to fix that).

igbot

This looks to me like it will be very useful. I’ve created and deployed three posts now. Though I named it igbot, the posts are going to IG & FB, and also pins to Pinterest. The pinterest API is still giving us problems, but we are getting there. I can’t check impressions (yet), and part of the point was to learn what is working and gravitate toward it, but just the generation is already a significant weight off my shoulders.

growthbot

Growthbot looks at the data from adsbot and igbot posts, as well as context of all my books, and makes suggestions for growing my audience and improving income. I’ve already pursued some of its suggestions. The email that went out for the release of my next book was prompted by growthbot. I know, that kind of thing should be obvious, but having that reminder that, “Hey, this needs to happen NOW for the timing to work out,” was helpful. Now, I told it to think outside the box, and it warned me that not ever idea would “fit”, and it is true, but the results do have me thinking about new things. Here’s a clip of the output:

  3. Protect and Slightly Grow the Facebook Lead Ads — They're Working [Facebook Ads]
     Data: CPR dropped 31.8% week-over-week to $1.33, and 22 net new subscribers joined at a total cost of ~$29.24. With the Feb 28 preorder email blast 2 days away, a larger, warmer list directly increases preorder conversion potential. This is the best-performing paid channel right now.
     Action: Increase the daily budget on whichever ad set had the lower CPR this week by exactly 10% (not more). Do not touch the other ad set yet — let it run another week to confirm trend. Review which creative is driving the lower CPR and ensure it's the active variant.
     Impact: A 10% budget increase should add 2–4 more subscribers per week at the same efficiency, compounding list size before the April 13 launch without resetting the learning phase.

What’s next

I do have a list of items. The pinterest post count and impressions are not right, there’s still some date mismatches, and carousel posts aren’t working. But I hit a snag—I burn through the $20/month AI package too quickly. 20-25% in a day when I’m even just casually working on it (like working in the mornings and evenings outside my work hours).

Rather than jump to the next package, I’m first trying this: doing code refactoring with one AI engine (ChatGPT OpenAI) and doing specific coding work with Claude. That keeps me closer to $40/mo. I don’t know that I’m particularly using each AI engine to its strengths, but it is a reasonable division of labor.

For tech geeks, I have a “botmaster” and “botcore” repository now, and I’m trying to refactor so that the bots can use pieces of “botcore”. OpenAI is working out how to make all the deployments faster (local caching, separate zips, code & data separation), while I use Claude to solve problems like the pinterest counts being wrong and carousels not working.

Has it all been worth it? Well, I’ve learned a lot in a week, and it is definitely worth it for that. The way I’ve packaged the pieces, I think they could be reusable by others, so maybe I can make money selling these things. The igbot makes social media something more reachable (previously it took me too much energy). The growthbot MIGHT be able to help me keep on track. Long way to say, “I’m not sure yet, but I think I’m better off for having tried.”