Automated advertising on a small budget

What the platforms now sell

Google’s Performance Max, Meta’s Advantage+ and their equivalents take budget, creative and a goal, and make the targeting, bidding and placement decisions themselves. You supply assets and a conversion signal; the system decides who sees what.

For large advertisers with steady conversion volume this works well and is the sensible default. For a small business it introduces a specific problem that nobody selling it will lead with.

The learning problem

These systems learn from conversions. Until enough have happened, decisions are close to guesses, and your budget pays for the guessing.

The commonly cited figure is around fifty conversions in a rolling month before automated bidding becomes reliable. A business getting eight leads a month never reaches that threshold, so the system stays in its learning phase permanently. You are funding an education that never finishes.

There is a second effect that is easy to miss. Automation looks for the cheapest conversions, and the cheapest conversions are often the least valuable: the enquiry from outside your delivery area, the person wanting the thing you do not do. Volume goes up, quality goes down, and the dashboard reports success while you spend more time on worse leads.

What works at small scale

Intent, not interest. Somebody searching for what you sell, in your area, right now. That is search advertising, and it works on small budgets because you are not persuading anyone: you are appearing where a decision is already being made.

Negative keywords. The most valuable half hour in small-budget advertising. Read the actual search terms your ads matched and exclude the irrelevant ones: “free”, “jobs”, “salary”, “DIY”, the town you do not serve, the competitor’s brand. This is where AI helps: paste the search-term report and ask for candidate negatives grouped by reason. Then read them, because it will suggest excluding some terms you actually want.

Manual or capped bidding while volume is low. Less clever and it does not spend money on discovering things.

A tight geographic radius. Genuinely tight. Most local businesses set it far wider than they would travel.

Where AI helps in advertising

Writing variants. Fifteen headlines within a character limit, in your voice, using your facts. A dull job done well.

Reading the reports. Paste a month of data and ask what changed, which terms cost the most for nothing, and where the spend went. This is analysis nobody in a small business has time for, and the data is in front of it.

Landing pages. Matching the page to the ad. A specific ad pointing at a general homepage wastes most of the click, and this is a common, expensive and easily fixed error.

Estimating what a lead is worth. Not with a model: in a spreadsheet, with your own numbers. Enquiries, conversion rate, average job value, margin. Without this you cannot say whether any of the advertising works, and most small businesses cannot say.

The arithmetic before the campaign

Do this on paper first.

If your average job is 8,000 with a 30% margin, a job is worth 2,400 to you. If one in four enquiries becomes a job, an enquiry is worth 600. So paying 400 per enquiry is marginal and paying 150 is good.

Now you have a number to judge every platform report against, and the dashboards become legible. Without it, “cost per lead: 380” is neither good nor bad and you will believe whichever story the interface tells.

The summary

Advertising is where AI’s helpfulness and your interests diverge most sharply. The automation optimises for the metric it was given, on a budget too small for it to learn from, and reports its own success.

Keep the budget small, keep the targeting tight, read the search terms yourself, and know what a lead is worth before you spend a currency unit.

Start by not advertising

For many small businesses the first sensible answer is that paid advertising is not the constraint.

If enquiries already arrive and half are never followed up, or the local listing is incomplete, or the price is not published anywhere, money spent on ads is buying more of a problem you have not fixed. Advertising multiplies whatever your conversion process already does, including doing it badly.

The order that works: fix the follow-up, complete the listing, publish the prices, answer the real questions. Then advertise, with a known value per lead and a budget you would not miss.

Watch it weekly for a month, then monthly

Automated campaigns drift. A term that was fine in week one starts consuming the budget in week five, and nothing tells you.

Twenty minutes a week for the first month: read the search terms, add negatives, check the cost per enquiry against your number. After that, monthly is enough unless something changes. Put it in the diary, because the failure mode here is not a bad decision but an unattended one.