Rebuilding 6-Fig Agency Case Study 2

Then - how to make an extra £60,000/year in 140 hours.

Last month I wrote about the start of the re-building the 6-fig case study.

This month there’s a major update(s) as well as some big opportunities in other fields.

The agency part

• Organic traffic value is up from $4.6k to 5.0k - did peak a bit higher but dropped down nearer the end of the month.

• Search console clicks look really good, trending up massively,, although it is in waves because it’s B2B.

• Affiliate earnings are around $500.

• Client earnings around $1000

• 1 Course built, one other in motion (live case study course).

• YouTube videos being created.

• Importantly YouTube SEO (Proper not the BS version) working extremely well and processed.

• Link building & content marketing takes time so not a massive amount of progress there but 5 new posts, 10-15 fully re-written updated ones.

The other part - The opportunities

In the process of rebuilding this branch of the agency I mentioned about the below points in the last post that would make up the core ideas of the plan;

In October I’ve gone super-deep on a few of these.

Even the ones I’m not going to cover like content marketing, link building, improving UX, rebuilding posts, those were all covered effectively, but weren’t as noteworthy.

The approach to implementing and improving on anything I do in business or otherwise has to always follow the marketing mentality I have around growth marketing or growth hacking, organic marketing whatever you want to call it.

But in the past 3 weeks I think I’ve made about 8 months worth of progress with 2 core principals/processes/skills.

#1 - YouTube Rankings

When I say YouTube rankings, I don’t mean the type of youtube SEO tips you’ll see online if you search “how to rank on youtube” or variations of this approach.

I mean re-building the model that we had years ago that enabled us to rank for medium competition keywords on-demand.

Obviously we can’t rank for keywords generating millions of searches a month on youtube, but medium competition terms, generating thousands to tens of thousands of searches a month can be ranked pretty much at will now.

This also importantly has been productised.

After building the process manually and trying to semi-automate this with software (which I thought would work perfectly, actually ended up not working at all).

But with some accounts, proxies, and a trained up VA, the “YT Division” costs about $800/month to run and can pretty much rank any video.

This is a skill that doesn’t have a massive amount of utility right now, it’s early days and our video content sucks, but it follows the key principal of my edge marketing mentality.

Implement the work/system once, profit forever.

Now every video that ranks should equate to anywhere between $10-$100/month in value to the company, whether that’s leads, sign ups or more likely just affiliate commission.

So although the “YT division” costs $800 ish a month to run. Once 10-15 videos are up and ranking it’s breakeven, after that every new video is pure profit with no marketing-unit-cost effort (obviously still need to build good video content).

But the real earnings won’t actually come from just ranking affiliate style content or videos to get a few more Amazon SEO leads, the real value will be utilising this in a future business with the process already fully set-up and integrated.

#2 - AI Coding Prediction Models

I had been using chatGPT and more recently Claude a bit, maybe once or twice a week until September 2024.

In late 2024 I started to use Claude for our content marketing, writing posts, researching outlines etc.

It’s incredibly good.

We even built out 3 fully AI niche sites. Which I’ve left to age for a bit (SEO nerd speak for being lazy and doing nothing with them).

Once of which recently woke up….

But that’s a post for another time.

So then I started to get into Claude 3.5 Sonnet for the content side, really looking deeply into how I can optimise my projects around this.

Even using it to help the growth marketing of the KDP brand (the other case study that’s running, again a post for another time).

But on 22nd October a new Claude model came out, big news in the AI circles, basically nothing in the broader marketing/business circles I’m in.

But absolute game changer for me personally.

For anyone who doesn’t know I’ve been building out simplified sports betting prediction models for years, varying degrees of success, for the sheer hours and resources put in, probably not worth it in general, but this last Claude update has basically enabled me to build sports betting models (in the proper way)

Almost like a convergence of 10 years of working with overseas team members, 5 years of researching sports, stats, betting and predictions and then 2 weeks of having the realisation of how effective these AI models can be when they have a very specific use case.

The past 11 days or so has been a massive grind, everyday has been at least 10-11 hours building these models, arguing with machines, breaking things and basically learning how to code in that 11 day period because the use-case for it was there.

Although it’s felt like a grind, in reality, zooming out it’s 11 days and 110 hours later to have something that previously would have taken months to build, and likely thousands.

But now we have a structure that any data can be fed through to get back-tested results in seconds and minutes instead of weeks.

And although its definitely not perfect yet, the concept of work once, profit forever is always so clear in these situations.

The process of the betting models is currently being tested with rugby. The idea behind this is;

• Build a model that can run off structured data of players, matches, betting info etc.

• Test/train off one season for one league/tournament.

• Verify with another season from that same tournament type, if results hold up, bet moving forward.

And after hours of fixing bugs and ensuring results and predictions are correct, initial signs are promising.

That would be great if that were the case on its own but this process itself has more leverage, meaning any new rugby tournament that has those same data points that can be acquired, can be fed into the same model.

And although different tournaments have different nuances, performance should be similar and tweaking takes hours not weeks or months.

All meaning if you can generate an 8% edge on a bet, with decent liquidity, results become fairly exponential pretty quickly.

• 8% edge per bet.

• No liquidity issues on £2,000 per unit.

• Average of 70 bets per tournament per year.

• 5 tournaments per year (Top 14, Premiership, United Rugby Championship, Super Rugby, Champions Cup, potentially internationals eventually but they are probably sharper).

• 8% of £2k x 70 - £11,200.

• £11.2k x 5 = £50-60k/year with obvious easy upside potential (increase edge, increase unit sizing, branch out into other tournaments, rugby league etc etc).

And the ancillary skills learnt through building this in this way obviously cross back over into other sports we bet; cricket, hockey etc.

Of course building out and tweaking the other models will take another 20-30 hours, but all the data can be outsourced to a VA or scraped (if I can work out how to do that) and the feeding in of this takes seconds…