How to know if your ads are working

In June I had an ad bringing clicks in at three cents each. A hundred and eighty-nine clicks, one for every 40 people who saw it. Anyone looking at that dashboard would have said it was going well.

I had no way of knowing.

And it wasn't the platform's fault. I had asked it for clicks, and it brought me clicks, flawlessly. The problem is that clicks weren't my business.

Almost everyone who writes to me starts the diagnosis at the end: "is this number good?". That's question six. There are five before it, and if you skip them, whatever answer you land on is worth about as much as a coin flip.

This is the order I use. It works for Google, for Meta, for whatever you're paying for.

1. What counts as a result?

Write down, in one line, what has to happen for the month to have been worth it. Not in dashboard language: in the language of your business.

For a dental clinic it's a booked appointment that showed up. For a repair shop, an accepted quote. For me, a completed scan.

It sounds obvious and almost nobody has it written down. Until it is, the platform will choose for you: it will label as "result" the last thing it managed to see, which is almost always a click or a message someone started typing.

And it will chase that with all the efficiency in the world.

2. Can the platform see that result?

This is the one that breaks most accounts, and it isn't a setup problem: it's about where the sale happens.

If your customer messages you, asks for a price, books, shows up and pays at the counter, none of that touches your website. The platform never saw it. What it shows you as a result is the last thing it could still record before the conversation moved somewhere else.

Of 23 small-business sites we scanned in July, 12 couldn't connect a single dollar of Instagram spend to a customer who actually paid. I take it apart in what a good ROAS is in your sector.

If the answer here is no, stop. The four questions that follow don't mean anything yet, because you'd be asking them about a made-up number.

3. How much can you pay for that result?

This number isn't Google's and it isn't your agency's. It comes out of your business and two figures you already know:

What a customer leaves you. Not what they pay you: what's left after the cost of serving them. And if they buy from you several times a year, count the whole thing.

How many of the people who talk to you end up buying. If you close 2 out of every 10 conversations, each conversation is worth a fifth of a customer.

In money: if a customer leaves you $220 a year and you close 2 out of 10, every conversation that comes in is worth $44. That's your ceiling. Above it, each new conversation costs you more than it gives back.

Here's where I had to eat my own. When I launched my own campaign, each scan was costing me $7.70. Expensive or cheap? No idea, because I hadn't worked out the ceiling before launching. I spent weeks staring at a number I couldn't judge. This is literally what I do for a living.

4. What happens after the click?

This is where most of the money leaks, and it's the part nobody looks at because it isn't in the ads dashboard.

My own numbers, six days of campaign, $88 spent:

392 people landed on the page
10 ran the scan
0 started a conversation

One in every 39 who arrived did what the page asked. And of those, none took the next step.

The ads were healthy. The hole was in my page: on a phone the progress bar sat below the fold, so tapping "Scan" looked like nothing had happened, and the contact button was five screens of scrolling away.

Scaling a campaign with that open is paying to fill a leaking bucket. It took me a month of data to see it.

5. Do you have enough to believe the number?

Ten scans in six days isn't a measurement. It's an anecdote with decimals.

The platform isn't comfortable either: Meta's automated systems need on the order of 50 results a week to settle. Below that, cost per result drifts on its own: you'll see swings from one day to the next that mean nothing at all.

The rule I use so I don't switch things off too early: an ad doesn't get judged until it has spent three times what you're willing to pay for one result. If your ceiling is $44 per conversation, that ad needs $132 behind it before its number says anything.

And don't edit it in the meantime. Editing resets the count.

6. What are you comparing it against?

Only now does the question everyone starts with make an appearance.

Last month isn't a benchmark: last month may have been bad. Neither is your competitor, because you don't know their margin or what they're counting as a sale.

Two yardsticks are left standing. The first is yours: how much you have to sell for every dollar you put in just to break even, which comes out of your margin. The second is your sector, as an order of magnitude.

For reference, the 2026 search-campaign benchmarks — 13,474 campaigns — put the average cost of a new customer at $66.69 and the average click at $5.42. Those are medians, so half of all advertisers sit below them, and they're search, not social. Useful for checking you're in the right order of magnitude, not for handing you a verdict.

The uncomfortable case

In that same campaign I had two ads running.

One looked better up top: 4.6 out of every 100 people who saw it tapped on it, against 3.6 for the other one. If you'd asked me which to keep, I'd have kept that one.

It spent $11, was seen by 3,467 people and produced zero scans. Zero.

The automatic budget split figured it out before I did and put 87% of the money on the other one, the one fewer people tapped, which is the one that brought all 10 scans. It was right.

That's why this list reads from the sale upward and not from the ad downward. A high number at the top can coexist perfectly well with a zero at the bottom, and the top one is the one you see first. On how to diagnose a creative that hooks but doesn't close, I wrote this: the 3-second rule on Meta.

What to do this week

One. Write down in one line what counts as a result in your business. If it doesn't fit in one line, that's the problem right there.

Two. Work out your ceiling: what a customer leaves you, times how many of the people who talk to you end up buying. A rough number is enough.

Three. Follow one person end to end. Take a customer who came in this month and walk backwards: where did they come from? was it recorded anywhere? If at any point in that chain nobody knows, that's your blind spot, and changing the ad won't fix it.

None of this needs an agency. It needs an afternoon. And if after that afternoon you want to take the campaigns into your own hands, the step by step is in how to run your Google Ads without an agency.

Flying blind isn't bad luck. It's almost always a question nobody asked in time.

Frequently asked questions

How do I know if my advertising is working?
By answering six questions in order: what counts as a result in your business, whether the platform can see it, how much you can pay for it, what happens after the click, whether you have enough volume to believe the number, and what you're comparing it against. Most people start with the last one, which is why they decide blind.

How long should I wait before judging a campaign?
It isn't measured in days, it's measured in money and results. An ad doesn't get judged until it has spent three times what you're willing to pay for one result, and automated systems need on the order of 50 results a week to settle. Below that, cost drifts on its own. Editing an ad also resets the count.

Does a cheap click mean the campaign is working?
No. I had an ad with three-cent clicks that produced no sales at all, and another one that more people tapped than its direct competitor and that produced zero results. A cheap click only means the platform got you what you asked for, cheaply. If you asked for the wrong thing, it will get it very efficiently.

How do I measure sales that close over chat or on the phone?
Start simple: ask every person where they came from and write it down, even in a spreadsheet. Then connect your site's tracking to the moment someone opens the chat, so the platform knows that ad produced a conversation. Without one of the two, the sale happens out of the platform's reach and the number you're looking at is calculated on something else.

How much should a new customer cost me?
Less than that customer leaves you. As an external reference, the 2026 US search benchmarks put the average at $66.69 per new customer, with some sectors running triple that. Use it as an order of magnitude: the number that decides is your ceiling, not the average.

If you'd rather skip the afternoon of auditing, scan your site free: I'll tell you in plain terms what your marketing is measuring today, what's leaking after the click and how your operation looks next to your sector's. No agency and no airplane-cockpit dashboards.