Google Ads without the waste: fences for an automated auction.

Where budgets actually leak

The classic wasted dollar isn't fraud and isn't mystery — it's structure. A campaign left near its defaults will happily buy searches that merely resemble what you sell: the wrong intent (people looking for a course, a job or a free template), the wrong customer (people who want to do it themselves), the wrong place (clicks from regions you can't serve or ship to), and the work you'd decline anyway. Each of these is a leak with a named, closable cause, and closing them is most of what competent campaign management is. Automation tends to widen these gaps rather than close them: the more the platform is trusted to find "similar" searches on its own, the more it matters that someone has told it what similar must never include. The spend that remains after the fences is smaller — and it's the only spend that was ever going to produce a customer.

Match types and negatives: the fence line

A keyword isn't a fence, it's a suggestion — match types decide how loosely Google interprets it, Google has been interpreting more loosely for years, and loose interpretation is where budgets go to die. Tight match types cost you some volume and buy you intent; if you do hand matching to the platform's broadest settings and automated bidding, do it on purpose, with the fences already up. The other half of the fence is the negative keyword list: the standing record of everything you don't do — "course", "salary", "DIY", the services adjacent to yours, the brands you don't sell or support. It's built by reading the search terms report — as much of it as the platform shows you — line by line, on a schedule; every term that makes you wince becomes a negative. An account nobody reads the search terms report for isn't being managed, whatever the invoice says.

Count the outcome that's real

Automated bidding optimises toward whatever you tell it counts, so the conversion definition is now an instruction, not just a report. Graded on the wrong event, a campaign will under-report reality and then get optimised in the wrong direction. Where customers phone, the call is the conversion: call tracking on the number the ad traffic sees, call assets on the ads themselves, and conversions counted only when a call lasts long enough to have been a real conversation. Where they sign up or buy, the event fires on the confirmed outcome, not the button press — and where you can, send qualified outcomes back from your CRM, so the bidding learns from customers rather than clicks. From there, let the landing page do its half: the ad answered "who", the page answers "why you" and takes the enquiry — sending paid clicks to a slow homepage is paying twice to lose once.

Honest budget logic

We won't print typical costs-per-click here, because they vary by industry, location and season, and any number would be a lie by the time you read it. The logic, though, is durable: your click price is set by auction against your competitors; your enquiry cost is the click price divided by how well the fences and the landing page do their jobs; and whether an enquiry cost is acceptable is a question about your margins, not about advertising. Work that arithmetic with your own numbers — measured against your own sales and customer records, not the platform's dashboard — and the "is it working?" argument settles itself.

The agent-native lens: fences that hold when software is on both sides

When an assistant checks the ad's promise. An assistant comparing options for someone reads the landing page, not the ad's charm. If the ad promises "same-week delivery" or "a fixed monthly price" and the page buries or contradicts it, an agent that cross-checks claims can quietly drop you from the shortlist. So write every ad promise so it can be verified on the page it links to, and keep prices, availability and terms consistent with any product feed or schema.org markup you publish. Some paid clicks may also come from software acting for people rather than from people themselves. That is one more reason to count only confirmed outcomes — a qualified enquiry, a paid order — so the bidding isn't trained on visits no person made.

When the account runs on agents. Reading the search terms report on a schedule is the chore most accounts skip, and it suits an agent well. Give it read access to the account and the CRM. It flags terms outside the fence, proposes negatives with a reason for each, compares the conversions the platform reports with the outcomes your records confirm, and drafts ad variants for review. What it shouldn't have is a free hand with budgets, bid strategy or new keywords: those changes, and the negatives it proposes, go into a queue that a person approves, and the account's change history plus the agent's notes become the audit trail.

When ads earn their keep — and when they stop

Ads are the right tool when you need demand now, when you're launching a product or entering a market where nobody knows you, or when a seasonal window is open and the slower channels haven't caught up yet. They're rented visibility: the day the spend stops, the visibility stops. So the honest strategy for most growing businesses is a crossing pattern — ads carry the load early while profile, reviews and site earn their standing, then the paid budget narrows to the products, services and regions where it still beats the unpaid click. And if you can't take on more customers right now, the bravest optimisation is pausing the spend. A campaign that can't survive that question wasn't working; it was just billing. As the platform takes over more of the bidding, remember that it isn't built to ask that question on your behalf.