What $1M a month in ad spend teaches you that R10,000 doesn't

I've run accounts spending more than a million dollars a month, and accounts spending five thousand rand. The tactics overlap far less than most people assume.

Here's what only becomes visible at volume, and what it means if you're not there.

Statistical significance stops being theoretical

At R10,000 a month you might get 40 conversions. Two ad sets, 20 conversions each, one at R180 CPA and one at R310. That looks decisive. It isn't. At those volumes, the difference could reverse next month on noise alone.

At high spend you get thousands of conversions a week, and differences become real fast. You can call a test in four days and be right.

What this means for smaller budgets: stop making decisions on 20 conversions. Either extend the test window until you have meaningful volume, or test bigger differences. Two headlines that differ slightly need enormous data to separate. Two completely different angles might separate on a fraction of it.

Test big things when you're small. Test small things when you're big.

Wasted spend hides in plain sight

At a million a month, a 3% inefficiency is R500,000 a year. That justifies someone going through search terms line by line every week, auditing placements, and checking for overlap between audiences.

At R10,000 a month, 3% is R300. Nobody's building a process for that.

The trap is that the percentage is often much worse in small accounts, not better. Broad match with no negatives, Display left switched on inside a Search campaign, one campaign competing against another for the same query — these waste a far larger share of a small budget, and nobody's looking because the absolute number seems trivial.

Check the percentage, not the rand value.

Attribution error compounds

Every platform over-reports. Meta claims conversions Google also claims. Both claim conversions that would have happened anyway.

At high spend you can afford to test this properly — geo holdouts, incrementality tests, running a region dark for two weeks and watching what actually happens to revenue.

At low spend you can't. So the honest move is to stop treating platform-reported ROAS as truth and start reconciling against actual revenue in your own system. If Meta says it drove R80,000 and your bank says you did R45,000 in total, the number that matters is R45,000.

I wrote more on this in why your conversion tracking is probably wrong.

Creative volume is the real constraint

The single biggest predictor of Meta performance at scale is how many genuinely different creative concepts you can put into the account each month. Not variations — concepts.

Big accounts have a production line. Small accounts have whatever the founder filmed on their phone.

This is the most transferable lesson. Creative volume beats targeting sophistication at every budget level. If you can only fix one thing on a small account, fix how many new angles you ship per month, not your audience settings.

What genuinely doesn't transfer

Some things only work at scale, and copying them will hurt you:

• Heavy campaign segmentation. Splitting into fifteen campaigns starves each one of data. At low budget, consolidate.
• Value-based bidding. Needs conversion volume to model against. Below that threshold it's guessing.
• Aggressive automated bidding. Smart Bidding needs data. On thin volume, manual or maximise-clicks with a tight structure often outperforms.

The thing I'd tell my earlier self

Managing a lot of money doesn't make you smarter. It makes you faster at finding out you were wrong.

The advantage isn't insight, it's iteration speed. If you're working with a small budget, your job is to manufacture that iteration speed some other way: bigger tests, longer windows, and brutal honesty about what your data can and can't tell you yet.