The Part Of Running Campaigns That AI Cannot Hand You

Quick Takeaways

1. This month I took on someone new and started teaching them the Meta and Facebook campaign side of the business from a standing start.
2. I assumed the platform would be the difficult part. Learning where every button lives took roughly an afternoon.
3. What takes months is judgement. Reading a dip correctly, telling a broken measurement from a broken campaign, and knowing when the right move is to leave it alone.
4. Judgement is built from repetitions, which is the one thing no tool and no course can post to you.
5. AI has removed almost all of the mechanical work in this job and left the hardest part exactly where it was.
6. The real question for an owner deciding between learning this and hiring it out is whether you will get enough accounts, over enough months, to build the instinct.
7. A proper training sequence goes foundations, then tracking, then one supervised campaign, then a written brief, then approval-gated launches, then independence.

I was wrong about which bit was hard

When I decided to train someone into the campaign half of this business, I set aside a large chunk of September for the platform. Ads Manager is a genuinely horrible piece of software. Things are buried three menus deep, the same setting appears under two different names, and the interface changes shape every few months for reasons nobody has ever explained. I had a mental image of weeks spent on that.

We got through it in an afternoon. By the end of the first week the person learning could build a campaign, set an audience, load creative, and find every number that matters without me pointing. If the job were the software, the training would have finished before the month did.

What I had underestimated is everything that sits on top. The part where you look at a screen full of numbers that all appear to be saying something and have to decide what is actually true, then decide whether to do anything about it. That is the whole job, it turns out, and I had been treating it as the easy bit because I stopped noticing I was doing it about six years ago.

Why the mechanical part collapsed

Here is the context that makes this worth writing about. Ten years ago, running campaigns was largely labour. Building variations by hand, writing copy from scratch, pulling reports together on a Friday, manually splitting audiences and watching them. A big share of the skill was stamina and organisation.

Most of that has gone. Our creative testing method is five images against five pieces of text, which gives twenty five ads out of one build, and assembling that set is now quick. Copy variants arrive in seconds. Reporting assembles itself. An ideal Australian audience of one to three million people can be defined in a couple of minutes. The platform will even tell you, in a confident tone, which ad it thinks is winning.

So the part of the job you could once charge for by the hour has largely evaporated, and what is left is the part that was always underneath it. Deciding what the numbers mean. That shift is the reason this post exists, because it changes the answer to a question a lot of owners are asking themselves right now.

The three things that actually take months

Knowing when a dip is normal

Every campaign that has ever worked has had a bad three days in it. Revenue climbs into a weekend, then Sunday and Monday fall off a cliff, and the graph looks like the end of the world if you have not seen it forty times before.

Usually nothing is wrong. The campaign has scaled, it has finished with the warm audience who were always going to buy, and it has moved on to strangers. That first weekend was the easy crowd getting used up, exactly as it is meant to happen. You need to have watched that pattern complete itself several times before you can look at the same graph and feel calm about it.

I can teach the explanation in five minutes. Believing it at eleven o'clock on a Sunday night when it is someone else's money on the screen is a completely different skill, and it only arrives with repetition.

Knowing when the instrument is broken

This is the one I care most about, and it is why the person learning owns pixel and conversion tracking setup for their clients, not only the ads. If you cannot set up measurement, you cannot tell whether the measurement is lying to you.

Sales that stop appearing in the dashboard while the client's bank account keeps filling up. Conversions counted twice. An event that fires on page load and therefore reports every visitor as a purchase. A tag quietly wiped during a site update on Wednesday. In all of those cases the campaign is fine and the instrument is broken, and every single one of them looks identical to a campaign failing.

Somebody who has only learned the ads side will respond by changing the ads. They will tear up something that was working, because the only tool they have is the one they were taught. Teaching tracking first and ads second is deliberate for that reason.

Knowing when to sit on your hands

The hardest thing to teach is doing nothing on purpose.

I got this wrong myself not long ago. Three video ads had produced nothing, three tidy zeros in a column, so I turned them off after a single day. Something nagged at me straight away and I ignored it, because a chart looks like a fact and a feeling looks like an excuse. When I went back and looked underneath the chart, two of those ads had barely been shown to anybody. The zero was too young to mean anything at all. Two went straight back on.

The instinct that said hold on was not magic. It was years of watching campaigns, running faster than I could put words to it. What instinct cannot do is prove itself, so it turns up to the argument with no paperwork and loses to a spreadsheet, including the arguments it should win. The gut tells you where to dig. The numbers tell you whether anything is down there. Somebody in their first month has one of those and is still building the other.

What AI does to this picture

AI will build your campaign, write your variants, summarise your week and give you a recommendation with a confident little arrow next to it. What it will not do is tell you that a number is too young to act on, because it has no sense of how long a thing has been running unless somebody has taught it to check. It will hand you a beautifully formatted zero and let you act on it.

Which is a fair description of what happened to me with those three video ads. Every dashboard showed me the same clean figure. Not one of them thought to mention that a single day is nowhere near long enough to have earned that conclusion.

So the technology has taken the labour and left the judgement, and judgement is built the slow way: by watching many accounts, over many months, including the ones that went badly. That is true for the person I am training and it is equally true for you.

What a real training sequence looks like

This is the order we are actually working through. It is deliberately slow at the start and it is filmed as we go, because it became obvious in the first week that training one person and building a course are the same job done once.

1. Fix the foundations before anything is spent. I build or repair the website first, then hand the campaign half over. We will not run ads for a business until an audit puts the site above 75 out of 100. Warner Pool Care in Brendale went from 42 to 96 before a dollar went in, and then thirty dollars a day on Google Shopping returned fourteen to one. You would not hand the kicking tee to a middle forward when you have a fullback landing 85 per cent. The ad budget is the shot at goal and the website is who takes the kick.
2. Then tracking, owned end to end. Pixel, conversion events, the lot, set up by the person learning for their own clients. This has to come before campaign work so that a broken instrument is something they can recognise and repair.
3. Then one campaign under supervision. A single account, built together, watched daily. One is the right number. Three at once teaches speed and stops you noticing anything.
4. Then a written brief for every client. A handover meeting and a document that states what the business does, who the customer is, what a lead is worth, what has been tried, and what would count as success. Writing it forces the thinking that separates managing an account from managing a business.
5. Then approval-gated launches. Nothing goes live without me looking at it first. This is where most of the real teaching happens, because the conversation is about a specific decision on a specific account with real money attached.
6. Then independence, account by account. The gate comes off when the judgement is demonstrably there, one client at a time, on the evidence of how they handled the last one. Our floor is an eight to one return on ad spend, with eight to fifteen the working range, and that number is the same for both of us.

So should you learn this yourself or hire it out?

If you are an owner weighing that decision, the software is a distraction. You can learn the software. Everybody can learn the software now.

The honest question is whether you will get the repetitions. One account, your own, watched for a few weeks when you have a spare evening, will not build judgement. You will see one seasonal cycle, one audience, one product set, and you will still be guessing three years from now. Somebody running fifteen accounts sees fifteen versions of every pattern in a quarter, which is why they can look at your flat Sunday and tell you without much drama whether it matters.

There is a version of this where learning it yourself makes complete sense. Steady spend, a product you understand deeply, enough volume that you are in the account most days, and the patience to be mediocre at it for a while. If your situation looks more like a few thousand dollars a month and a diary already full of actual work, you are better off buying the repetitions somebody else has already paid for.

The short version

The platform was an afternoon. The judgement is a year, and possibly longer, and none of it can be shortcut by a tool that has never watched a campaign die for a stupid reason.

We are documenting the whole sequence as it happens, including the parts that go sideways, inside The Billy Bot Blueprint, a free community on Skool at skool.com/the-billy-bot-blueprint-4837. If you would rather have the repetitions than build them, that conversation is a short one and we have it most weeks at Infinity1.

Frequently asked questions

How long before someone can run campaigns unsupervised? Months, and it varies by person and by account. A sensible marker is whether they can explain a bad week in terms of what the numbers mean before they suggest changing anything.

Why teach tracking before ads? Because a campaign that looks broken and an instrument that is broken produce identical screens. Somebody who can only change ads will always change the ads.

Do I need my website sorted before I start advertising? Yes, and it is the cheapest decision in this whole list. Traffic sent to a site that cannot answer questions or convert a visitor is money spent finding out something an audit would have told you in a day.

What is a realistic return on ad spend to aim for? We work to eight to one as a floor and fifteen to one as a strong result, though the honest answer depends on your margins and what a customer is worth to you over a year.