Build both, split it evenly, let the data decide

Picture a physio clinic in a growth corridor on the edge of a major city. The owner wants new patients. The obvious instinct is to pick the campaign that feels right, fund it, and hope. Location based, because everyone Googles physio near me. Or condition based, because the clinic loves treating knees. Pick one, commit, and find out twelve months later whether the hunch was any good.

That is a slow, expensive way to learn something you could have known in a fortnight. The opinion in the room is almost never the cheapest path to a booking. The clinic owner has a hunch, the last agency had a hunch, and the account quietly burns money proving whichever hunch got funded first.

I was guilty of this too, early on. I'd back the campaign I personally liked the look of. Then I noticed the campaigns I'd written off were often the ones converting cheapest. So I stopped trusting my taste and started trusting the test. The question is no longer which one will win. It's how fast can I find out.

Here is what I actually do. We'd always go a location based campaign, and then we'd probably go a condition based campaign as well. We'd split test the two, and at the end of the day, we just let the data make the decisions. Two candidates, funded equally, racing under the same conditions.

Location based is the safe bet. No matter where you are in Australia, location based keywords are a strong campaign, because the intent is sharp and you show up in the Maps pack right when someone is ready to book. Condition based is the assumption, the more interesting bet: physio for low back pain, knee pain, shoulder pain. It might convert beautifully or it might starve. I genuinely don't know in advance, and pretending I do would be the lie.

The same logic works for any fork a clinic faces. Adults versus paediatrics for an OT practice. Assessments versus ongoing therapy for a psychology clinic. Two plausible lanes, build both, split the budget evenly, and find out what works. Now the only honest question left is how to fund two campaigns without spreading the money so thin that neither can breathe.

Splitting evenly only works if each side has enough oxygen to produce a real signal. A campaign on starvation rations doesn't lose the test honestly, it just never gets the data to compete. So there's a floor: roughly $500 a campaign per month is the minimum I'll run a theme on. Below that, a campaign can't gather enough bookings to tell you anything you can trust.

That floor is why budget, not ambition, decides how many campaigns you run. At $1,000 a month you can run two themes at $500 each, a clean 50/50 test. At $600 a month you can't, so you run one theme properly rather than two badly. I won't set up heaps of campaigns for the sake of it just to underspend on all of them. Better to do one thing really well first.

Split evenly, fund both above the floor, and you've built a fair race. But a campaign is just the container. The actual contest, the thing that compounds week after week, happens one level down, inside the ad group.

Every ad group runs two ads against each other. Not one ad you fell in love with. Two, competing, with budget split between them and the only judge being which one earns cheaper bookings. There's no opinion in it. There's just a fair fight and a scoreboard.

Every 14 days I take the one that's losing, cut it, and create a new split test to try and beat the other one. The winner stays and defends its title. A fresh challenger is built specifically to beat it. The winner usually becomes clear inside 10 to 14 days, which is why the fortnightly cycle works, it's long enough to gather signal and short enough to keep moving.

The beautiful part is when nothing can beat the champion. You might have one ad that's running for a year, because every split test we make loses to it. That isn't laziness, it's the opposite. It's an ad that has survived two dozen deliberate attempts to kill it. That's an asset, and you only know it's an asset because you kept trying to dethrone it. Which raises the obvious question: how do you know any of this is actually working at the level that pays the bills, the patient in the chair?

Zoom out from the ad group and the entire build is one big hypothesis I'm willing to write down before a dollar goes out. Pick a new patient goal. Estimate a realistic cost per booking for the area. Multiply it out to a monthly budget. Build to that. Then check at month end whether the bet held.

A worked example: $1,000 a month at roughly $100 a booking should produce about 10 new patients. That's the bet, stated out loud, with a number attached. We'd make that hypothesis, we'd build the ads around that, we track that at the end of the month, we go, were we correct in this or not, and then adjust from there. If it held, the winning lane earns more budget. If it missed, no story, no excuses, you reset the hypothesis with what the month actually showed and build again.

I'm honest with owners about what those numbers look like in 2026. For generalised physio, $70 to $90 a new patient is reasonable to aim for over time. Starting out in a fresh area I'd model closer to $90 to $100 while the account figures the market out. I can't promise you $50 patients consistently, and chasing $30 to $40 patients in a metro market is a fantasy unless you're genuinely regional. Stating the bet honestly is the whole point, because a bet you can't lose isn't a bet, it's a sales pitch.

This is where the build both philosophy stops being a tactic and becomes a way of working with someone's money. I can't promise results, but we can promise transparency. I will not tell a clinic owner I know which campaign wins, because I don't, and anyone who says they do is guessing with your budget and a straight face.

What I can promise is a disciplined experiment run in the open. Two campaigns funded fairly, two ads fighting in every ad group, a loser cut every fortnight, and a month end checkpoint where the hypothesis is either confirmed or killed in front of you. There are no locking contracts, because it's either you see value or you don't. If you don't, you'd leave, and you should.

It's not a money printer where cheap patients fall out of the sky. If you do it well, you just have to do it better than the people around you, connect with the searcher more honestly, and track it properly. After that, the numbers either stack up or they don't. I like data. I think data doesn't lie. That intellectual honesty is the entire pitch, and it's the bit a generalist agency built to defend last quarter's guess can never quite bring itself to say.