01 — Foundation
Why the platforms want to hear from your server
A browser pixel only sees what the browser lets through, and that gets less every year. Conversion APIs reverse the direction: your server reports the conversion directly to Meta, Google, TikTok or LinkedIn, reliably and with better data. We prefer to build these pipelines on your own tagging server, as described under server-side tracking, with no third-party tools in between.
We have been doing web analytics since Google Analytics existed, back when it was still Urchin. The lesson from two decades: interfaces are quick to wire up, but only a properly thought-through setup delivers numbers you can trust. The framework for that is our overview of Tracking & Analytics.
02 — Current
What's changing in the Conversion APIs right now
Meta rolled out two big updates in April 2026. The Meta-enabled Conversions API is a free one-click setup where Meta takes over the server infrastructure. And the pixel gets an AI enrichment layer that automatically reads page content like product name, price and availability from the page; existing pixel users get a 30-day notice in Events Manager before it activates. Meta's own argument: advertisers with CAPI see on average around 18 percent lower cost per result than without it.
Our read: the one-click setup lowers the barrier to entry, but you give up control over which data flows, and industries in Special Ad Categories such as finance or health are excluded anyway. If you care about data minimization and your own event logic, you are still better off running CAPI through your own sGTM.
What that means in practice: check the notices in Events Manager before the AI enrichment switches on automatically. We take care of that as part of a free CAPI check, just reach out below.
03 — Discipline
Meta CAPI with deduplication and EMQ as a KPI
With Meta, pixel and CAPI run in parallel, and that is exactly where the most common mistake happens: the same purchase event arrives twice. The solution is a shared event ID carried by both the browser event and the server event, so Meta recognizes them as one. This is not a nice-to-have, it is the core of the setup.
The quality yardstick is event match quality: it shows per event how well Meta can match the customer data you send along. We treat EMQ as a KPI, lift it deliberately through advanced matching and keep watching it after go-live. What the better signals do inside the campaigns is covered on our Meta Ads page.
04 — Discipline
Google Enhanced Conversions
Google takes the same route with Enhanced Conversions: first-party data such as email address or phone number is transmitted SHA-256 hashed and matched against signed-in Google accounts. Conversions that would otherwise be lost to the click become attributable again. Clean normalization before hashing is critical, things like lowercasing and trimmed whitespace, otherwise nothing matches. The consent signals have to travel along correctly, as described on the Consent Mode page.
05 — Discipline
TikTok Events API and LinkedIn CAPI
TikTok and LinkedIn follow the same logic with their own pitfalls. The TikTok Events API deduplicates via event_id and needs the ttclid for good matching. LinkedIn CAPI is interesting for B2B because leads can be matched via the hashed email even when the click has long expired. We build both setups on the same sGTM pipeline: one event stream, four endpoints.
06 — Practice
Offline conversions from the CRM
Especially in lead funnels, the real revenue happens days or weeks later in the CRM. Conversion APIs can feed exactly that back: the lead becomes a qualified deal, and the platform hears about it, value included. The algorithm then optimizes for customers instead of form fillers. The prerequisite is that click IDs and consent status are passed cleanly all the way into the CRM. The measurement base for it, cleanly named events in GA4, is described on the GA4 page.
07 — Honest
Typical mistakes we remove
The classic is double counting: pixel and CAPI fire without a shared event ID, the conversion count doubles, ROAS looks fantastic and is wrong from top to bottom. Add to that events without customer data that have an EMQ near zero, test events landing in the live data stream, and setups where the same conversion is reported from the shop backend and the website in parallel. Mistakes like these rarely stand out immediately; they just quietly distort every budget decision.
A conversion counted twice is worse than one lost. It looks like success.
Do your conversion numbers add up?
Tell us briefly which platforms you run and how your conversions are currently reported. We check deduplication and match quality and tell you honestly what is missing. First call free, 30 minutes.