Most of what gets written about Facebook advertising comes from people who never open Ads Manager. It produces advice that sounds right and does not survive a real budget: target your ideal audience, test several creatives, watch your cost per click. Nothing false in there. Nothing useful either.
This page explains the mechanics. What campaign objectives actually optimize for, why a conversion campaign needs an event volume most SMEs will never reach, what to do below that threshold, why almost every account I open sends Meta a single type of event, and how to read a report without confusing what the platform credits itself with and what your business actually earned.
I am Christian Gómez, MarketingSTRAT, a one-person studio in Quebec City. I run the accounts myself, I shoot the videos myself, and so I watch the same mistakes repeat from one account to the next. If you are reading this to run your own Facebook advertising rather than hire anyone, good: the content is the same either way.
Meta does not show your ad to people who are interested in your product. It shows it to the people most likely to take the precise action you asked for, at the lowest cost. The auction combines your bid, the estimated probability that the person takes that action, and the perceived quality of the ad. Meta documents this, and it is the whole logic of the system.
Hence the trap of the Traffic objective. You ask for link clicks, the algorithm finds you the people who click links. That population exists: it clicks a lot, everywhere, and buys almost nothing. Your reports will look beautiful. Low cost per click, high click-through rate, impressive volume. And the phone does not ring. A classic tell is the gap between link clicks and landing page views. When half your clicks never reach the page, you are paying for reflexes, not for attention.
| Objective | What the algorithm looks for | When it is defensible |
|---|---|---|
| Awareness | Reach and impressions at the lowest cost | Covering a local market, supporting a launch |
| Traffic | Link clicks or landing page views | Rarely alone, mostly to feed retargeting |
| Engagement | Reactions, comments, video views, messages | Social content, filling audiences |
| Leads | Forms submitted, on your site or inside Meta | Clear offer, short form, fast follow-up |
| Sales | Purchases or configured conversion events | Online commerce, or a well-instrumented conversion |
The rule is simple: pick the objective that matches the action you actually want, not the one that produces the prettiest numbers.
Meta documents a learning phase: an ad set needs roughly 50 optimization events per week for delivery to stabilize. Below that, the ad set stays in Learning limited. That is not a cosmetic label. It means the delivery model does not have enough signal to find a pattern, that it keeps exploring, and that your cost per result will swing from one week to the next without you changing anything.
Do the math before you launch. If your cost per quote request runs around $60, as an arithmetic example and not a market benchmark, 50 conversions a week works out to $3,000 a week, per ad set. An industrial SME can sell six-figure contracts on a nine-month cycle with a dozen good clients a year. Those companies will never hit 50 weekly conversions, and that is fine. What is not fine is building the strategy as if they would.
Second consequence, less well known: every significant edit to an ad set restarts the learning phase. Changing the budget substantially, changing the optimization event, changing the audience, adding ads. The nervous advertiser who adjusts the account three times a week keeps his own account in a permanent state of instability, then concludes that Facebook does not work.
This is the normal situation for an SME. Here is what works, in the order I apply it.
Consolidate. One ad set receiving the whole budget beats five ad sets splitting crumbs and never learning. The reflex to segment by region, by age and by interest divides your signal by the number of segments. Leave the budget at the campaign level and let the auction work.
Move up one step in the funnel. Optimize on a more frequent event that is still correlated to the sale: a contact, a form start, a pricing page view, a spec sheet download. The higher you move, the more volume you get and the weaker the correlation becomes. That is a trade-off to make consciously, not a magic trick.
Lengthen the window. Meta's default is 7 days after a click and 1 day after a view. Using the 7-day window instead of one day mechanically gives the model more events.
Change your unit of time. A low-volume business does not judge a campaign over three days. It judges it over four weeks of stable spend, with a denominator big enough to mean something. A cost per lead calculated on seven conversions is not data, it is an impression.
And sometimes, accept the channel's real function. In several Quebec B2B markets, Meta is not going to generate quote requests directly. It is going to make a company familiar to a few thousand specific people, and the result shows up elsewhere: searches for your name, replies to your emails, people who already know you when the rep calls. If that is the case, say so and measure it accordingly, rather than forcing a conversion objective that will never fill.
I regularly open accounts where the pixel has been installed for years and has never sent anything but PageView. That is the default event, the one that fires on page load. Everything else, the form submission, the booking, the call, the purchase, has to be placed deliberately. Nobody did it, because the installation stopped the moment the indicator turned green.
An account in that state cannot optimize for a conversion. It does not know what a conversion is. That is often the real reason someone fell back on the Traffic objective: it was the only one that seemed to work.
The minimum serious setup: verify the domain in Business Manager, place the standard events that match your reality (Lead, Contact, Schedule, Purchase), prioritize the events in Events Manager, since Meta only processes a limited number per domain following Apple's tracking changes, and validate with the Test Events tool before spending a dollar.
Then, the Conversions API. The browser blocks, ad blockers block, iOS blocks. Sending the event from your server, with an event ID to avoid duplicates with the pixel, recovers part of the lost signal. It is not an agency luxury, it is what lets the algorithm keep learning.
A word on the legal framework, because it applies here: in Quebec, Law 25 requires you to inform people when you use profiling technologies and to give them a way to turn those off. A working consent banner is part of the installation, not part of a later phase.
Ten years ago the advantage came from targeting. You built interest audiences your neighbour had not found. That is no longer true. Detailed targeting options have been cut back, broad audiences with automatic optimization often perform just as well, and everyone has access to the same settings. What still differs from one advertiser to the next is what is inside the ad.
Concretely, for an industrial or services SME in Quebec:
Meta provides delivery diagnostics that rank your ad against ads competing for a similar audience: quality, engagement rate, conversion rate. It is not a report card, but when an ad ranks below average on all three, the problem is not the budget.
Meta places your ads in the Feed, in Reels, in Stories, in Messenger, in search results and on Audience Network. The automatic placements option lets the auction move your budget toward the cheapest inventory that produces the action you asked for.
The important point: short-video inventory is abundant, so the cost per thousand impressions there is generally lower than in the Feed. That does not mean the customer costs you less. The consumption context is not the same. Someone scrolling Reels at night is not in the same frame of mind as someone going through their Feed. A lower cost per thousand impressions with a lower conversion rate can produce a higher cost per customer. This is exactly the kind of trade-off that a dashboard displayed at the campaign level hides.
The way I work: leave placements open, but supply real templates for each format, 9:16 and 1:1 on top of 16:9, with the important elements outside the zones the interface covers. Then, once there is enough volume, look at the placement breakdown and judge on cost per result, not on cost per impression. Audience Network deserves particular attention: it produces very cheap clicks that are often worthless. Excluding it is sometimes justified, knowing that every exclusion narrows the auction's room to move.
The Results column depends on your attribution setting. Change the window and the number changes, without anything real having moved. With 1-day view enabled, a person who saw your ad go by without clicking, then filled out your form the next day, counts as a result. Sometimes the ad played a role. Sometimes the person was going to call you anyway.
Four habits that keep you honest:
Frequency deserves an eye too. A B2B audience in Quebec can be a few tens of thousands of people. With a decent budget, the same person will see your ad often, fast. When frequency climbs and cost per result climbs with it, it is not the platform running out, it is your creative.
Whether you hire someone or decide to run your Facebook advertising yourself, the starting structure is the same and it is not negotiable.
You own the ad account, the pixel, the page and the Business Manager. The agency gets access, not ownership. An agency running ads from its own ad account with its own pixel keeps your delivery history and your audiences the day you leave. That point gets settled on day one or never.
After that, an honest work order looks like this: validate the instrumentation before spending, define the event that actually matters to your business, calculate whether your volume supports a conversion campaign or not and say so plainly, build enough material to sustain several weeks of rotation, let it run without touching it, then report with the attribution window displayed and the comparison against your internal data.
What an agency should not do: promise a cost per lead before seeing a single number, bill a percentage of the budget spent (that rewards spend, not results), present reach and impressions as results, and hide report access behind its own dashboard.
At MarketingSTRAT, Meta advertising is not sold on its own. It is part of the Signal Program, at $3,500 per month, three-month minimum, which also covers the video and photo production that feeds the ads. The reason is simple: managing ads without controlling what is inside them means optimizing settings around a problem you are not allowed to fix.
It depends entirely on your cost per result, and you do the math before you launch. Meta stabilizes delivery around 50 optimization events per week per ad set. If a quote request costs you $60, as an arithmetic example and not a market benchmark, that threshold works out to $3,000 per week. Most SMEs do not spend that. It is not a reason to skip advertising, it is a reason to configure differently: one ad set, a more frequent optimization event, a 7-day window, and a judgment made over four weeks instead of a few days.
Yes, but rarely as a direct-demand channel. Industrial buyers are on Facebook and Instagram like everyone else, they are simply not shopping for an equipment supplier at that exact moment. So the channel mostly serves to become familiar to a small, precise population, to circulate visual proof, and to stay present through a long sales cycle. The companies that fail on Meta in B2B are almost always the ones that demanded immediate bottom-of-funnel performance from the channel.
The pixel alone loses a growing share of events: blockers, browser restrictions, mobile privacy settings. The Conversions API sends the event from your server and recovers part of that signal, provided you deduplicate with a shared event ID. For a low-volume account this is not a detail: every lost conversion is one less conversion for learning. And in Quebec, Law 25 applies to both: you have to inform people and offer a way to turn profiling off.
Count on at least four weeks of stable spend, with no structural change, before drawing a conclusion. Any significant edit restarts the learning phase and makes the two periods incomparable. The question to ask is not how many days, it is how many conversions have I accumulated. A decision made on seven conversions is not a decision, it is a coin toss.
In most of the accounts I run, broad audiences with automatic optimization hold up against detailed targeting, because the model has more room to find the people who take the action you asked for. Detailed targeting keeps its value when you have a real constraint to respect: a service area, a language, a strictly defined sector. The real filter today is in the first shot of the video and the first line of the copy, not in the targeting panel.
Three questions separate the serious from the rest. Who owns the ad account and the pixel when the mandate ends, you or them? Which attribution window is shown in the reports, and is it compared against your internal data? And what is the billing logic: a percentage of budget rewards spend, a flat fee rewards work. An agency that promises you a cost per lead before seeing your numbers is guessing.
If you want to talk it through, write to me. MarketingSTRAT is a one-person studio in Quebec City: I am the one who looks at the account, who shoots the footage and who answers you. The Signal Program is $3,500 per month, three-month minimum, and it includes visual production on top of ad management. If your volume does not justify an engagement yet, I will tell you, and what is on this page is enough to start on the right foot by yourself.
Book a call