A heating and cooling client, €80 a day, Meta Instant Forms. Over 4 weeks the account ran 8 angles: energy savings, air quality, financing, a free inspection, a buyback deal, before-and-after photos, an AI-generated set, and real job photos from the vans. Every one of them came back between €41 and €48 per lead. The client wanted to know which creative to fix next. I did not have a creative answer for him.

Short answer: Whether your problem is the offer or the ads shows up in the spread between angles. Creative problems produce winners and losers, because different promises reach different people. Offer problems fail evenly, since no promise rescues a deal nobody wants. Measure the gap between your best and worst angle before you touch anything.

The takeaways

Spread carries the diagnosis. When the creative is the weak part, the best angle usually costs 2 to 3 times less per result than the worst. The ranking barely matters.
8 angles are often one promise in 8 costumes. If none of them changes what the buyer is asked to accept, flat results say nothing about your offer.
The read has an entry ticket. Under roughly 25 to 30 conversions in the window, with no angle below 8, uniformity is what randomness looks like anyway.

What does a creative problem look like in the numbers?

A creative problem shows up as a distribution. Some angles beat the account average by a wide margin, some sit well under it, and the order holds steady enough to name a winner without squinting. In the accounts I have run, the gap between best and worst at that point lands around 2 to 3 times on cost per result, read straight off Meta Ads Manager.

That gap is good news. The market moves when you change what you say, so saying it better is a lever you still hold. You rebuild the losers around whatever the winner promised, leave the deal untouched, and improve inside a bargain that already works for somebody.

The familiar version of this diagnosis reads the funnel vertically, hunting for the step where people drop out, which I went through in high CTR, no conversions. It needs a break to find. When nothing broke and everything is mediocre, it returns a shrug.

What does an offer problem look like?

An offer problem is flat. Your angles land inside a band so tight that reordering them would be arbitrary, and the band sits above what the business can pay. Nothing separated. That uniformity is the signal: 8 different promises drawing the same shrug means the promise was never the variable.

Read it the uncomfortable way. Your ads worked. They put the deal in front of interested people, who declined at the same rate through every door you opened. The €41 to €48 band in the heating account was not 8 failures. It was 1 failure repeated 8 times, and the repeated part was a free inspection from an unknown company, in a category where 3 competitors ran the same free inspection that week.

Why do eight angles usually test only one promise?

Read the list again: energy savings, air quality, financing, free inspection, buyback, before-and-after, AI creative, real job photos. 4 of those are subjects. 2 are formats. 1 is a production method. Only financing and the buyback touch what the customer is asked to accept, and both showed up as a line in the copy rather than a term of the deal.

So the month did not run 8 tests. It ran 1 promise, we will come and look at your system for free, in 8 costumes. Flat results from that set carry no verdict on your offer, because the offer never varied. Before you read uniformity as an offer problem, check that at least two of your angles differ in what the buyer gets, not only in what the buyer sees.

How much volume do you need before flat results mean anything?

Enough that each angle has a shape. At 4 leads per angle everything looks flat, so you would be reading noise as a diagnosis. My working floor is about 25 to 30 conversions on the account inside the window, with no angle under 8. Below that, the honest answer is that the data cannot tell you which problem you have.

Under the floor, stop diagnosing and go collect volume: fewer angles, more budget each. 3 angles at €80 a day for 2 weeks give you something to read. 8 angles at that spend give you 8 anecdotes. Meta's own documentation puts the learning phase exit at roughly 50 optimization events per ad set within 7 days, and I worked through the decision-window version in how long to run Facebook ads.

What do you change when the offer is what failed?

Start with the terms, since the price is often the one thing you cannot move. Risk is usually available: who carries it if this goes wrong, and can you take that weight off the buyer. So is speed, so is named proof, and so is the shape of a fixed price, which is what financing does once it is a term instead of a caption.

If I were rebuilding that heating account, the first change would be the inspection: a fixed-price diagnostic with the fee credited against any repair, so the visit reads like work instead of a sales call. Then 3 angles that disagree about the deal itself, photography held constant. If the spread comes back, you own a creative problem again, which is a much nicer problem.

How do you find out what the market wants?

Ask it in its own words before you guess at a new deal. Read what buyers say when they are not talking to a seller: Reddit threads, Amazon reviews, the Apple App Store and Google Play, the forums where your category gets argued about. Keep the exact phrasing with the emotion attached, and rank the pains by intensity instead of frequency. The loudest complaint and the strongest one are usually different lists.

That evidence layer is what Adscalr builds: it searches those 5 sources seeded with your and your competitors' URLs, keeps 2 to 4 verbatim phrase markers per quote with emotion and register intact, and maps quotes to Eugene Schwartz's 5 awareness stages. That mapping shows whether your deal answers the blocker people keep naming. It will not design, price, or test an offer for you. Audience intelligence does the listening. The deal is still yours to make.

Is it my offer or my ads? How to tell