In my post a couple weeks ago, I talked about systems and their benefits. And I also talked a bit about automations.
Wanted to give a snapshot of all the automated systems I have online right now and what I’m working on.
1. Lead Generation (online)#
I’ve already talked about this last time, but will explain more here. It is a system of 3 parts:
- Scrapes YouTube for emails.
- I use a script to gather unique emails, and personalize the cold email outreach messages. And this exports everything in a CSV.
- I feed the CSV into Instantly.
The process is nice and automated and I do get some interest. However, of all the responders I might be actually only interested in 3-5% of them to take a meeting. I’d like to expand on the lead generation a bit quite a bit more. If I can, what are some ideas to 100X this, so I get more relavant meetings? A few ideas.
Right now, the process for scraping emails is I get a keyword list from ChatGPT and I modify/paste in keywords one by one to my script, and I generate a CSV per YouTube keyword. I sometimes forget to do this, and can be automated with a cron job. Instead of me pasting in keywords to my script and calling the script one keyword at a time, I can just simply have a supervisor script that calls ChatGPT for a list of keywords we haven’t seen before, and then everyday, scrape 10 keywords. Shove it in a cron job.
Instantly would then be the bottleneck. I am cheap and only pay $37/mo, but I can upgrade it so I can get more than 1k contacts in there at a time.
If I do that, then the bottleneck is my inboxes. I only have 8 which makes cold outreach slow. I can setup 80 more and make all inboxes forward replies to my main inbox.
Finally, I’m only sending emails on Tues-Thursday as ChatGPT advised me to. I can get a free output increase of 40% (though conversion might be lower) if I also do Mon/Fri. I can technically get even more if I send on weekends, but I don’t think that is so good and I think it is better to buy more emails in that case.
So if I 1) 100% automate the keyword gathering + scraping, 2) upgrade Instantly, 3) pay and setup 80 email accounts, and 4) expand sending to all weekdays (possibly lowering conversions), I can go about 14X faster.
It’s a lot of work, so it’s good, not great. Would be worth if it this was all I was doing (as 14X the lead flow would mean a lot more time spent in the CRM and following up). So maybe I’ll take a look at this improvement a little bit later.
2. Content Generation (online)#
There are certain formats on Tik Tok that is not extremely difficult to automate.
But a format that’s easily automated and proven to go viral is somewhat limited. Here are 2 I can think of:
- Video narrations that is basically a talking slideshow. A voice talks and with each half-sentence, you switch slides. This channel does this.
- Carousels. These are just a series of images.
I have video narrations working but not carousels yet. The process currently is as follows:
- I type in the idea I want
- I give it a screenshot or a 2-word description of the image style I want
- It will generate the script/panels with the images. I edit it as I like.
- It will spit out a video just like the TikTok channel referenced above.
- I just have claude write a description for it, and add subtitles in Da Vinci Resolve. Publish it on a burner.
Before this automation, all the cutting and image generation would take about 60-90 mins per video. Now it takes up to 10 mins, max.
This is a 6-9X increase and the quality doesn’t degrade. In fact, it is a bit better because alot of the animations I would be lazy to do in a manual setting; however, in this automation the sliding/dynamic zoom animations are randomized and consistent.
So I can pump out 90 pieces of content in the same time it might take me to pump out 10 in the past. Or 9X faster of knowing what goes viral so I can iterate on it.
9X doesn’t sound that fast but over time it is. 9 days compressed to 1 day is “whatever.” 90 months (7.5 years) compressed to 10 months is very good.
Some improvements is:
- Having the carousels automated too. However, I don’t think this is necessary because if one format can go viral, I think that is good enough to double down on. Adding this would add optionality but not necessarily speed anything up.
- Have my UI auto-post to Tik Tok. There’s a lot of hoops Tik Tok forces you to go through. And I already spent 1 week doing the work to apply for automatic posting permissions. 5 weeks later, they rejected me with notes. While this can speed things up a tiny bit (the 30 seconds of air-dropping to my burner and then uploading it), I don’t think it is worth the Herculean effort to do this. Hard to justify spending another few hours to save max 30 seconds per video. Not to mention, automated API postings won’t have subtitles, which I’d need to add anyways.
- Automating subtitles. This can be worth it, but I prefer to add subtitles in DaVinci resolve as I can ALL CAPS and stylize the fonts how I like. And I don’t think I want to automate this until I am 100% sure a very specific caption style is the best. And this also won’t save much time since auto-captions take about 2-3 mins max.
That is, most of these improvements will require a lot of engineering work (weeks) to save a very small amount of time (minutes). And thus my automation already gets me most of the way there (90 mins -> 10 mins). Adding a few weeks of engineering to go from 10 mins -> 8 mins per video makes no sense, as assuming 40 hours of work, I’d need to do about 1200 videos for me to breakeven.
Hope this example helps in terms of how I think about what to automate and what not to.
3. Automating Trading (in progress)#
Trading is hard because a lot of times you’re really into the tape and emotionally involved.
A quantitative way of trading allows you to test a system over 40+ years of past data, and have high conviction that it works. Instead of trading with emotions, you trade with logic and math.
After the post I made 2 weeks ago, I was trying very hard and couldn’t beat buy and hold. 49X returns in TQQQ vs 226X buy and hold, with a huge drawdown.
Last night I had a breakthrough though and found something that is working. There are quite a few ideas on improvements but so far, so good (this strat is walkforward and simple). The drawdown curve orange is my strategy, and dotted blue line is the buy and hold baseline drawdown. So it returns a bit higher (1232X vs 215X over 15y), and has a lower drawdown.
I hope to improve this a bit more and come up with 5-7 more strategies so I can have automated systems that is diversified (so I don’t blow up my net worth).
Conclusion#
I hope to automate more things in the future. But in the age of LLMs, I think the only temporary edge you get is really building these automated systems to 10-100X your manual input.
Because soon enough, everyone will 10-100X their inputs and 100X’ing your input is just the baseline.
Again I want to emphasize you should use discretion when automating and only put in the work to do so if it is worth it (see the discussion in the content generation automation).