For each mistake, a label says where my claim comes from: platform rule (Meta has it in writing), Rafa measurement (I recorded it) or experience (I have seen it, without being able to put a figure on it).
1. Stopping the ad after two days
Platform rule. Meta describes a learning phase: until the ad has had enough conversions, delivery remains approximate. Judging after 48 hours means judging on very little.
Rafa measurement. My landing page ads receive a median of 8.9 applications per week: volume takes time to build. But be careful: in Épernay, with a form, 20 of the 31 applications arrived in the first two days. The start can also be immediate.
2. Changing the ad all the time
Platform rule. A significant change (text, image, area) partly restarts the learning phase. Change one thing at a time and give it time to have an effect.
3. A page that takes too long to open
Rafa measurement. From 2023 to 2025, only 70% of link clicks became a page visit: three in ten were lost before the page displayed. Part of this loss comes from the tracker, which does not see everyone. I do not know the share of each, but a lost click is still a paid click.
4. A form that is too long
Rafa measurement, partial. On my 2026 pages, about a third of the people who start the form submit it (from 31% to 38% depending on the client). I have not tested the effect of length: I only measure that two in three give up. Experience: every extra field probably plays a part, especially a long text to type on a phone.
5. Choosing the wrong labour pool
Rafa measurement. Across twelve ads from the same network, launched on the same day in different towns, for similar jobs (workshop, sales), the cost per application ranges from €5.91 to €44.18. The two most expensive are also the ones that received the fewest applications (3 and 6): one candidate more or less moves the price a lot. So I cannot say that “the labour pool explains everything”; I say that the gap exists and that it is large.
6. Choosing the wrong format for the job
Rafa measurement. The Facebook form gave from €0.53 to €5.83 per application on three ads; the landing page €9.86 on average across 17 ads. The details are here. Putting a technician on a form with no questions means receiving contacts you will not be able to sort; putting a crew member through ten questions means paying more for applications you did not need. Experience.
7. Believing the budget works one way or the other
This is the only point where my two sets of figures do not say the same thing:
| Data set | Budget per day | Ads | Cost per application |
|---|---|---|---|
| 2023-2025, leads counted by Meta | €10 or less | 8 | €16.12 |
| 2023-2025, leads counted by Meta | more than €20 | 8 | €17.85 |
| 2026, applications received | less than €10 on average | 8 | median €15.24 |
| 2026, applications received | €10 or more | 9 | median €8.32 |
In 2023-2025, I see no effect from the budget. In 2026, the ads with the smallest budgets cost more per application at the median, but they also each receive very few applications. I draw no conclusion: with 8 and 9 ads per cell and mixed jobs, it is only a signal. Neither “a big budget is a must” nor “a small budget changes nothing”: I do not have the answer yet.
In short
- Do not stop an ad after two days, and do not keep changing it: these are platform rules.
- A slow page loses clicks (only 70% become a visit); a long form loses candidates (a third get to the end).
- The cost per application varies a lot from one ad to the next, even for the same client on the same day.
- Budget: my two sets of figures disagree. I do not decide.
