Create Next App

The one idea

The pixel is a teacher, not a net

Most people picture the pixel as a net: cast it over your website, catch visitors, then chase them with ads until they buy. That picture is years out of date and causes most common mistakes.

The pixel is a teacher. Every time someone does something meaningful on your site, it tells Meta. Over thousands of these reports, Meta learns what your buyers look like, then goes and finds more people like them across billions of users.

So you are not mainly building a list to chase. You are training a prediction engine to recognise your ideal customer. Almost everything below follows from that.

The basics

Pixel, dataset, events

A pixel (Meta now calls it a dataset, but most people still say pixel) is the connection between your site and Meta. It carries messages about what people do. Each message is an event:

ViewContentsomeone looks at a product or key page
AddToCartsomeone adds something to their cart
InitiateCheckoutsomeone starts checkout
Leadsomeone fills in a form or enquiry
Purchasesomeone buys

Meta uses this stream for two jobs: reporting how your ads did, and deciding who to show your ads to next. The second job is where the money is.

What CAPI does

Why the server matters

For years every event was sent from the visitor's browser. That worked until it did not. Apple's privacy changes block tracking on iPhones unless people opt in (most do not), ad blockers stop the pixel firing, and browsers restrict cookies. The browser alone now misses a big share of what really happened, often a quarter or more of your conversions.

The Conversions API(CAPI) fixes this from the other side. Instead of relying on the browser, your server sends the event straight to Meta's server. No browser in the middle to block or lose it.

You run both together. The browser sends what it can, CAPI fills the gaps, and a shared event ID stops the same sale being counted twice. This is called deduplication, which is why running both is safe.

Event Match Quality

Why the details matter

When you open Events Manager you will see an Event Match Quality score out of 10 for each event. Here is what it means in plain terms.

For Meta to learn from a conversion, it has to match that event to a real Meta account. It does this using the identifying details you send with the event: email, phone, name, location, and so on (all securely hashed, so you are not handing over raw personal data). The more good detail you send, the more confidently Meta matches it, and a higher match score means better optimisation.

This is also where CAPI quietly wins. A browser often cannot access much identifying detail, but your server usually can (you already have the customer's email and details at checkout). Richer identifiers sent server-side lift your match quality, which lifts everything downstream. If your score is low, that is usually the first thing to fix.

How good data helps

It is not about one buyer

Meta's ad system is a prediction engine. When you ask for purchases, it is constantly guessing which person out of billions is most likely to buy from you right now. The quality of those guesses depends entirely on the examples it learned from.

When the browser loses a quarter of your conversions, you are hiding a quarter of your best teaching examples. The engine has fewer confirmed buyers to learn from, so its guesses get worse, and worse guesses cost you more per result. CAPI restores those examples.

Now the part that saves the most money: the value of a buyer is the pattern, not the person. The individual who just bought is the least useful person to advertise to next — you usually want to exclude them from acquisition. Their real worth is teaching Meta what a buyer looks like, so it can find new people who resemble them. These are lookalikes, and they are where growth comes from.

And you never stop training. Every fresh purchase keeps the engine's picture current as seasons, trends, and products change. Feed it the whole funnel (views and add-to-carts, not just sales), and feed it continuously. Stop, and its picture of your customer slowly goes stale.

Why broad targeting wins

Hand-picked interests are last decade

A few years ago you were taught to pick your audience by hand: choose interests like “yoga,” layer on demographics, narrow it down. That was right when the engine was weak and needed pointing.

The engine is no longer weak. Fed strong conversion signal, it finds your buyers better than you can describe them. People interested in yoga are not necessarily people who buy your yoga product; the engine, learning from real purchases, finds the ones who actually buy. This is why broad targeting (often called Advantage+) now beats hand-built interest stacks.

But notice the condition: broad targeting only works if you feed the engine good data. Broad with weak signal is blind guessing across a huge audience, and it flails. Broad with strong signal is confident picking, and it wins. That is the direct link between CAPI and modern strategy.

Shorter attribution windows

Why this makes good data matter more, not less

An attribution windowis how far back Meta looks to credit a conversion to an ad. After Apple's privacy changes the default got shorter (commonly 7-day click, 1-day view), and the old long 28-day default went away. So Meta now credits a conversion only if it happens fairly soon after the click.

A shorter window leaves less room for error. A conversion the browser fails to report, or reports late, can fall outside the window and never get counted, never teach the engine, never appear in your reporting. When the window was long, sloppy tracking was survivable. Now every lost or delayed event is a real loss, and CAPI helps by delivering events reliably and promptly.

Reporting

Why Meta's numbers won't match Shopify or Stripe

Expect this, so it does not alarm you. Meta will often report a different number of sales than your store or payment system shows. The tool is not broken and Meta is not lying. They simply count differently.

Meta credits sales within its attribution window and to the ad it thinks influenced them, fills some gaps with modelled estimates, and loses some events to iOS. Your backend counts every order regardless of how it found you. These are two different questions (“how many orders did I get” versus “how many orders can Meta tie to an ad it showed”), so the numbers rarely line up exactly. Use your backend as the source of truth for revenue, and use Meta's numbers to compare ads against each other.

Be patient

The learning phase is real

New advertisers often judge results on day two, panic at erratic numbers, and switch everything off. That is usually a mistake.

The engine needs a stash of conversions before its guesses stabilise. Early on, results swing around while it learns; this is called the learning phase, and bumpy early numbers are normal, not a sign of failure. Give campaigns time and conversions to settle before judging them. Turning things off too early resets the learning and wastes the spend that was buying it.

Honest expectations

Data is necessary, not magic

Good signal makes a good offer and decent creative perform better. It cannot rescue a weak offer or a bad ad. Think of clean data as removing the friction between you and Meta's engine: it lets a good product find its buyers faster and cheaper, but the product and the offer still have to be worth buying. If the fundamentals are not there, better tracking will show you the truth sooner rather than hide it.

Data retention

How long Meta can keep using your data

When someone triggers an event, Meta can hold them in a custom audience for a limited time, then they age out:

What it is	How long Meta can use it
Website custom audiences (views, add to cart, leads, etc.)	up to 180 days
Purchase-event audiences (website and app)	up to 730 days
Meta native lead-form (Instant Form) engagement	up to 90 days

Purchases get the longest window because they are the highest-value signal and many buying cycles run longer than six months. Note that a Lead event sent to your pixel is on the 180-day clock, which is different from a Meta native lead form at 90 days. If you are in a region with strong privacy rules, these are the maximums Meta allows, not necessarily what your consent and privacy policy permit.

Where PixelFlow fits

Why we exist

All of the above assumes you can actually get your events, especially server-side CAPI events, flowing to Meta cleanly. That used to mean writing server code, out of reach for anyone on a no-code platform like Webflow, Framer, Squarespace, or WordPress without a developer.

That is the gap PixelFlow closes. It sends your events server-side through the Conversions API without you touching code, so you get the data-quality benefits this guide describes (more matched conversions, fewer lost events, higher match quality) that used to need an engineering team. In plain terms, it lets the prediction engine see more of what really happened on your site, which is the whole game.

Recap

Remember this if nothing else

You are not buying ads. You are training a prediction engine, then buying the quality of its predictions. Good, complete data buys good predictions, and good predictions buy cheap results.

The pixel is a teacher, not a net.
The browser loses events; CAPI sends them server to server so they arrive.
Higher match quality (more identifiers) means better optimisation.
Data teaches a pattern, not chases a person. Patterns find new buyers.
You never stop training, and you feed the whole funnel.
Broad targeting wins now, but only with strong signal behind it.
Shorter attribution windows make clean, prompt data more important.
Meta's numbers won't match your backend, and that is normal.
Give the learning phase time before you judge results.
Data is necessary, not magic; the offer still has to be good.

Reference

Quick glossary

CAPI (Conversions API): sends events from your server to Meta, bypassing the browser.
Pixel / Dataset: the connection that carries your site's events to Meta.
Event: a single reported action (ViewContent, AddToCart, Lead, Purchase).
EMQ (Event Match Quality): a 0 to 10 score for how well Meta can match your events to real accounts.
Lookalike: new people Meta finds who resemble your existing buyers.
Learning phase: the early period where the engine gathers conversions and results are unstable.
Attribution window: how far back Meta looks to credit a conversion to an ad.
ROAS: return on ad spend, revenue divided by ad cost.
CPA: cost per acquisition, what you pay for one result.
ATT: Apple's App Tracking Transparency, the iPhone prompt that lets users block tracking.

Understanding Facebook Ads and the Conversions API