Paid advertising Search catches demand. Social creates it.

The row you cannot afford is the one you pick because it is fashionable. A budget spread across all four buys four sets of learning data, none of them large enough to learn from — which is why most clients start with one.

• The number is agreed before any money moves

  What a customer is worth to you, what you can afford to pay for one, and what counts as a result. Written down, and used to judge every later decision instead of last month's engagement rate.

  Ask thisAsk what your target cost per booked job is. If nobody can say it out loud, nothing is being managed against it.

• Server-side tracking goes in before the first ad runs

  Meta's Conversions API, LinkedIn's conversions API and TikTok's Events API, each paired with the browser pixel and sharing a reference so one sale is never counted as two.

  Ask thisAsk to watch a real test enquiry arrive in the ad platform by both roads, browser and server, before launch.

• Campaigns are created paused

  Audiences, exclusions, budgets, creative, copy and landing pages are all readable before a dollar moves. You activate when you are ready, not when we are.

  Ask thisAsk to press go yourself. An agency that cannot hand you that button has already spent your money.

• Prospecting and retargeting are never reported together

  Warm audiences convert cheaply, so a blended cost per lead can look healthy while the prospecting half quietly fails. The two are separated in the account and in the report.

  Ask thisAsk for last month's cost per result for cold audiences only. A single blended number is the answer to a different question.

• Creative is replaced on a schedule, not after the drop

  Frequency and cost per result are read weekly, and new concepts are tested against the current winner one variable at a time, with enough spend behind each to mean something.

  Ask thisAsk when the running creative was made and what it is currently being tested against. Two answers, not one.

• Leads are routed in seconds, with the source attached

  Lead-form submissions push straight into your CRM or inbox the moment they land. On social you interrupted someone rather than answered them, so the enquiry cools faster than a search lead does.

  Ask thisSend a test enquiry at four o'clock on a Friday and time the callback. That is the number your customers experience.

• The report ends in jobs and what each one cost

  A fixed cadence, in the language you use about your own business, covering what was spent, what it produced, what changed and why — whether the news is good or bad.

  Ask thisAsk to see a real client report before you sign anything. Reach and engagement rate on the last page is a tell.

• Rule 05

  Cut what costs more than a job is worth

  The number it moves: Cost per lead

  Ads are judged on cost per booked call, and anything above your line is replaced or cut.

  How this rule works: Cut what costs more than a job is worth
• Rule 06

  Spend more only where jobs get booked

  The number it moves: Enquiries vs spend growth

  Budget moves to the ad and audience that books calls, only while enquiries outgrow the spend.

  How this rule works: Spend more only where jobs get booked

• Facebook

  Feed at 1:1 and 4:5 · Stories and Reels at 9:16 · lead form or landing page

  The ad carries the whole promise, because a lead form means there may be no landing page behind it. One offer, one audience, one reason to act now. Static images and carousels do a great deal of the work here.

• Instagram

  Reels, Stories and Explore at 9:16 · Feed at 4:5 · same Meta account as Facebook

  Show the job. The first frame is the ad, and most of it is watched on mute, so the words go on the screen. A Facebook creative pasted in here is the most common thing we replace in week one.

• LinkedIn

  Single image, document and video · conversation and lead-gen forms

  Written for the person who signs off, not the person who searches. Opens on the problem, states the constraint you solve, and lands on something that survives being forwarded to two other people.

• TikTok

  Full-screen vertical video at 9:16 · Spark Ads run a real organic post as the ad

  Shot rather than designed, and made in volume rather than in polish. The first second decides whether the rest of the budget was worth spending, so the hook is written before anything else is.

Where your own footage or photography exists we will use it. On social, real material from the job site regularly outperforms polished production — and it is the one asset a competitor cannot copy.

That judgement is easier to check on somebody else's work than on ours. We take seven Australian campaigns apart on the same five questions each time — what the opening second of a Metro Trains or Kennards Hire film is actually doing — with the advertiser named in the heading and the agency credited wherever one is publicly known.

• The learning phase never ends

  Budget split across too many ad sets means none of them gets enough results a week for delivery to leave learning. Every one of them stays expensive and none of them stabilises.

  Prevented byFewer ad sets with more budget each. When something is not moving we consolidate rather than add, and we say no to the extra audience someone wants to test.

• Edits during the learning window buy nothing

  A meaningful edit to an ad set restarts its learning. An anxious week of daily tweaks can hold an account in a permanent learning phase and it will look like the platform is the problem.

  Prevented byStructure is held steady through the first two to four weeks. Changes are queued and batched, and we tell you what is being held rather than going quiet.

• Retargeting flatters the whole account

  People who already know you convert cheaply. Blend them into one number and prospecting can fail for months behind a cost per lead that reads beautifully.

  Prevented byCold and warm are separated in the build with hard exclusions, and reported separately. Budget decisions are made on the cold number.

• Fatigue is read a month too late

  Frequency climbs, cost per result follows, and a monthly average smooths the whole thing away until the invoice does not match the leads.

  Prevented byFrequency and cost per result are read weekly, and the refresh cycle runs on a calendar rather than on a bad month.

• The campaign optimises toward a proxy

  Set a campaign to buy landing-page views or link clicks and it will buy cheap ones forever, very efficiently, without ever finding a customer.

  Prevented byEvery campaign optimises toward an event on the agreed list, with a value attached. Nothing is bought against a proxy because the real event was harder to wire.

• The same sale is counted twice

  The pixel and the server-side connection both report the conversion without a shared reference, so one job arrives as two and the cost per job halves on paper.

  Prevented byDeduplication is part of the build, and it is proved with a real test enquiry pushed end to end before launch rather than discovered in month three.

The first of those is easier to avoid with something to measure against. Commercial cleaning carries the widest spread in our whole book, and the scenario page puts the cheap end and the expensive end side by side across the accounts we publish, because the gap is the lesson rather than the average. The failure at the other end is quieter: advertising is the easy part of scaling and attending the work is not, so it is worth checking how many jobs a month your crew can physically complete before you ask four platforms to find you more.