How I'm Using AI to Run Meta Ads for Trades Businesses

Last month I ran 14 ad variations for a single solar installer in South East Queensland. Different headlines, different images, different calls to action, different audience segments. Within two weeks, I had statistically significant data on which combination drove the lowest cost per lead.

A traditional marketing agency would have picked one creative, run it for a month, sent a report, and billed $2,000. I tested 14 variations in the time they'd have tested one.

That's the core of what I'm building: an AI-powered marketing performance service for Australian trades businesses, starting with solar installers.

Why solar? Because the economics are perfect for this model. Solar installations are high-ticket — $8,000 to $15,000 per job. That means a business can afford to spend $100-200 to acquire a lead and still be profitable. The lifetime value of a customer who refers others or comes back for battery storage is even higher. And the market is growing — Australian solar installations hit record numbers in 2025.

Most solar installers I've spoken to are doing one of two things with their marketing: nothing (relying on word of mouth and existing customers) or paying an agency $1,500-3,000 per month for generic digital marketing. The agencies aren't bad, but they're running the same playbook for every client. Same targeting. Same creative formula. Same reporting cadence.

The problem with the agency model is that it treats marketing as a service, not an experiment. You hire them, they set up campaigns based on "best practices," and you hope for the best. If it works, great. If it doesn't, they tweak some settings and try again next month.

My approach is fundamentally different. I treat every ad campaign as a set of hypotheses to be tested.

The experiment loop works like this. Step one: I generate multiple ad variations using AI — different angles on the same offer. Maybe one headline leads with price ("Solar from $5,999"), another leads with urgency ("Last Month for the Government Rebate"), and another leads with social proof ("Trusted by 200+ Gold Coast Families"). Each gets paired with different creative and different audience targeting.

Step two: I launch all variations simultaneously with small budgets. Enough to generate statistically meaningful data within a week, but not enough to waste money on a loser.

Step three: AI analyses the results and identifies which combinations are working. Not just click-through rate — I'm looking at cost per lead, lead quality (do they actually pick up the phone?), and downstream conversion signals when available.

Step four: Kill the losers, double down on the winners, and generate a new batch of variations that iterate on what's working. The second round of tests starts from a higher baseline because we're building on proven winners, not guessing again from scratch.

This loop repeats every one to two weeks. Over the course of a month, I might test 40-50 ad variations. A traditional agency tests maybe three or four.

The AI component isn't just about generating ad copy — although that's part of it. It's about the analysis loop. Spotting patterns in performance data, generating hypotheses about why certain variations outperform others, and suggesting the next set of tests. That analytical layer is what turns a spray-and-pray approach into a systematic optimisation engine.

What most agencies get wrong about Meta ads for local trades businesses:

They target too broadly. A solar installer in Southport doesn't need to reach all of South East Queensland. They need to reach homeowners within a 30-kilometre radius who own their property and have the right roof orientation. Meta's targeting is powerful enough for that level of specificity, but most agencies default to broad targeting because it's easier.

They use the wrong objective. Lead generation ads and traffic ads are not the same thing. For a tradie who needs phone calls and quote requests, optimising for clicks is a waste. You optimise for conversions — actual leads — and you let Meta's algorithm figure out who's most likely to convert.

They don't test creative aggressively enough. The difference between a good ad and a great ad for trades businesses can be 3-5x in cost per lead. But you'll never find the great ad if you only test two variations per month.

They report on vanity metrics. Impressions, reach, click-through rate — none of these pay the bills. The only metrics that matter for a tradie are: how many leads did I get, how much did each one cost, and how many turned into jobs? Everything else is noise.

What I'm learning from early campaigns: the best-performing ads for solar installers are surprisingly specific. Generic "go solar and save" messaging gets crushed by hyper-local, specific ads. "Your Neighbour on Smith Street Just Saved $1,800 on Power Bills" outperforms "Save Money with Solar" by a factor of three.

The other learning is that creative fatigue hits fast in small geographic areas. An ad that performs brilliantly in week one can tank by week three because the same 50,000 people have seen it six times. The experiment loop solves this because you're constantly rotating in fresh creative.

I'm pricing the service at a flat monthly rate — significantly less than a traditional agency — because the AI handles the work that would normally require a media buyer sitting at a computer for hours. My time goes into strategy, analysis, and client communication, not manually building ads in Ads Manager.

If you're a trades business owner spending money on marketing and not sure what you're getting for it, I'd love to chat. daine@dainereid.com.