This is a story about a Facebook ad testing mistake we made with our own money.

Our team put 24 ads inside one $100-a-day budget for our own local service offer. That works out to about $4.17 per ad per day. No ad can prove anything on $4.17 a day. So it was not a test. Every hook was starved before it could be judged.

We paused the campaign on August 1. The offer is paused, not dead. Here are the numbers, what we got wrong, and how we set up tests now. If you run your own ads, or pay someone to, the same mistake is easy to make.

The situation

The offer was our own service for local businesses. The ads were videos aimed at home services owners in one small local market. The goal was qualified sales conversations with owners who wanted more booked jobs.

We built one campaign with one budget of $100 a day. Inside it sat 24 ads. They tested different hooks, the opening seconds that decide whether someone keeps watching. Meta then split that one budget across all 24 ads.

We already knew small markets give thin signal. Earlier, in the same small market, we had sent traffic to our own local landing page. We spent $222. The page got 99 visitors, 2 form starts and 0 submits. The lesson we took was to use a form-first setup before a full landing page. We should have taken a second lesson too: small markets punish thin budgets.

What the numbers said

Here is the full record of the campaign:

  • Spend: $484.37 over its whole life.
  • Leads: 1, and it was not qualified.
  • Qualified conversations: 0.
  • Video length: the cuts ran 69 to 98 seconds.
  • Average watch time: 5.1 seconds.
  • Budget per ad: about $4.17 a day.

Our verdict was that the hooks were starved, not tested. A hook needs enough people to see it before you can say it works or fails. At about $4.17 a day, no single hook got there.

The watch time tells a second story. On average, a viewer gave each video 5.1 seconds. That is a short window for videos that ran 69 to 98 seconds. Whatever we said later in those videos reached far fewer people than the opening did.

What we got wrong

We made four mistakes. None of them was about the offer itself.

1. We asked one small budget to answer 24 questions. A budget of $75 to $100 a day can tell you whether a message lands. It cannot tell you whether that message is profitable. We took a budget sized to judge one idea and spread it across 24.

2. We expected single ads to give verdicts at low spend. Low-spend verdicts are mostly noise. At $30 to $75 spent per ad, against about a $20 cost per lead, each ad holds only 1 to 4 leads. In our own simulation, a rule of "$30 and no lead means kill" would cut about 22% of ads that were performing fine. Spread across 24 ads, $484.37 left the average ad short of even $30. No single ad's number meant anything.

3. We made long videos and let the opening carry everything. With 5.1 seconds of average attention, the first seconds were the whole ad for many viewers. The offer has to show up right after the hook, not a minute in.

4. The budget was too small to reach our own success line. Our planning target for this kind of offer is $150 to $200 per qualified booked call, with $350 as the cut line. Measured against that, the whole $484.37 could only ever have bought a few booked calls, even if everything had worked. A handful of calls cannot tell a good offer from a lucky one.

How we set up tests now

The fix is not "spend more". The fix is matching the test to the budget and to the question.

We separate two questions. First: does this message land? A budget of $75 to $100 a day can answer that. Second: does it pay? That takes about $2,500 to $5,000 spent and 20 to 50 conversions. We no longer expect the first budget to answer the second question.

We place new ads by budget. Under $100 a day, new ads go into the winner's ad set and get judged over a full week. At $100 a day or more, we clone the winner's ad set into a test set. We move about 10% of the budget over, so the total stays flat. At $500 a day or more, we run one ad per test set. Below about $100 a day, a separate test set starves both the test and the winner.

We judge groups, not single ads, at low spend. When each ad has only spent a little, we read the group of ads together. A single ad's cost per lead at low spend is mostly luck.

We set the targets before the money goes out. We judge ads on cost per qualified booked appointment, not cost per lead, the cost of ad views or click-through rate. For a local service offer, our planning targets are $150 to $200 per qualified booked call as good and $350 as a cut. We plan around a 70% show rate and a 30% close rate. Those are targets, not results.

Every video carries an offer line right after the hook. If most viewers only see the opening, the opening has to say what we are offering.

In a small market, we go form-first. We collect the lead before anything else, so every interested person lands in the CRM. We compare the two setups in Facebook instant forms vs landing pages.

What it means for you

You do not need to have run 24 ads to make this mistake. Any test where the budget is spread too thin has the same problem. Before you launch a test, or approve one from your agency, ask four questions:

  1. How many ads share this budget, and what does each get per day? A few dollars each means the test is starved.
  2. What is this test meant to prove? Whether a message lands, or whether it pays. Those need very different budgets.
  3. What number decides it? It should be cost per qualified booked appointment. We explain why in the math behind closed revenue, not booked calls.
  4. How long and how much spend before a verdict? Early numbers are noise. We walk through what to check and what to leave alone in the first 72 hours of a new Facebook campaign.

If you cannot get clear answers to those four, the test is a guess with a budget.

What we do not know yet

We want to be clear about the limits of this story.

We do not know whether the offer works in that market. We paused the campaign, and the offer is paused, not dead.

We do not know whether the hooks were bad or only starved. Some of them may have been good. None got enough money to show it.

We do not know whether a form-first setup would have done better there. That is the lesson we took from our landing page, not a result we have measured for this offer.

And one lead is not a sample. We cannot read anything from it, good or bad.

What we do know is that the structure could not have told us any of this. That is the mistake worth sharing.

FAQ

How many Facebook ads should you test at once?

As many as your budget can feed, and no more. Under $100 a day, we put new ads into the winner's ad set and judge them over a full week. If each ad only gets a few dollars a day, none of them is really being tested.

How much does it cost to test Facebook ads properly?

It depends on the question. A budget of $75 to $100 a day can tell you whether a message lands. Knowing whether it pays takes about $2,500 to $5,000 spent and 20 to 50 conversions.

Should you turn off a Facebook ad that spent $30 with no leads?

Not on that number alone. In our simulation, a "$30 and no lead means kill" rule would cut about 22% of ads that were performing fine. At low spend, judge groups of ads, not single ads.

What should a video ad say in the first few seconds?

The hook, then the offer right after it. In our own test, viewers gave the videos 5.1 seconds on average. If the offer comes later than that, many people never hear it.

If you want to know whether your own ad tests can actually prove anything, you can ask our team on a discovery call.