Two accounts, opened up.

One person, part time, in a trade where every competitor advertises on price. Work came from word of mouth and the occasional boosted post — which meant good months and quiet months with no way to tell which was coming.

The temptation in a business like this is to buy volume: bid on “cheap aircon clean”, win the click, do the job for what it costs you. That’s a treadmill, not a business.

Facebook message ads pointed at homeowners in specific suburbs who were ready to book, with suburb landing pages behind them and Google Ads carrying the urgency offers.

April is the month an aircon business is supposed to go quiet. That’s the point of showing this one rather than a January figure — anyone can fill a summer. The off-peak number is the one that tells you whether the marketing is doing the work or the weather is.

Read the third column, not the first. Jobs went up 7.5%. Revenue went up 27%. Almost all of that gap is average job value moving $50. Volume is easy to buy — you just bid higher. A better class of job is the part that takes an account built on purpose, and it’s the part that’s still there when you stop spending.

Everything above is paid. This next part costs nothing, and over a full year it grew faster than anything I paid for. Google Business Profile, July 2026 against July 2025.

The searches that surfaced the profile are the honest version of a keyword report:

Note what isn’t there: nobody typed the business name. Every one of these is someone who needed the job done and didn’t know who to call yet.

A small hire company running Facebook traffic ads at Australian holidaymakers. Spend was going out, clicks were coming in, and the enquiries reaching the inbox kept arriving half-finished — a name and an email, no dates, no vehicle. It looked like customers being lazy.

They weren’t. The site was doing it.

1. The spam trap was eating real customers. A hidden anti-bot field on the enquiry form was silently discarding legitimate submissions. No error, no message — the visitor pressed the button and the page sat there. Every one of those people assumed the business was broken and went to a competitor.
2. Pressing “check availability” did nothing visible. When the page’s script didn’t run — old phone, patchy hotel wi-fi, a cached file — the form fell back to a plain page reload. The customer landed back where they started with no confirmation anything had happened.
3. Completed bookings were destroying their own source data. A booking that arrived from a Facebook ad was recorded correctly at first touch, then overwritten with “Direct / unknown” when the customer finished the form. Which meant the channel report — the entire basis for deciding where the budget goes — was fiction.
4. One enquiry, two records. No deduplication and no disabled submit button, so a double tap on a phone created two customers who didn’t exist.
5. Pricing was typed into 64 pages by hand. Change the daily rate and you’re editing sixty-odd pages, or you miss some and the site quotes two different prices depending where the customer lands.

• Removed the trap killing submissions, and gave the form a real destination so it works even with the scripts dead
• Built a reconciliation layer: an abandoned half-enquiry and the completed booking that follows it now merge into one record instead of two
• First-touch attribution written once and never overwritten — so the channel report is something you can actually spend money on
• Deduplication on identical submissions, and a submit button that can’t be double-fired
• Every price replaced with a single reference. One change, sixty-four pages, no stragglers
• A booking call-to-action that follows the visitor down the page, instead of one button in a corner that scrolls away

Only once the plumbing held did the advertising get touched. I pulled their strongest competitor’s ads out of Meta’s public library and read the campaign off it: Australia only, running continuously since October, and crucially a lead form that stays inside Facebook rather than a click through to a website.

Ten months of uninterrupted spend is worth paying attention to — it’s a strong hint that something in there is working, though plenty of businesses do keep funding a campaign that isn’t. Treat it as a lead worth testing, not proof of profitability. So we built a matched campaign to test that mechanic against their existing one, holding the brand voice steady so the only variable is the mechanic itself.