Your attribution is lying to you.
Platform-reported conversions are a marketing document, not a measurement. What to build instead, and roughly what it costs.
Add up the conversions each of your ad platforms reports. Compare that total to the revenue in your accounting system. In almost every business we've looked at, the first number is meaningfully larger — often by a multiple.
This isn't fraud. Each platform is answering a slightly different question, using a slightly different attribution window, with visibility limited to the traffic it can see. Then every platform claims the same conversion. The sum of several partial truths is not a truth.
The three failures underneath it.
First, visibility. Browser privacy controls, ad blockers, and consent gates mean a meaningful share of your traffic is invisible to client-side tags. The events aren't wrong; they're missing, and missing data is invisible in a dashboard.
Second, ownership. If your click tracking runs on a vendor's domain, your measurement continuity depends on that vendor's terms, uptime, and continued existence. That's a lot of business risk parked in a subdomain nobody thinks about.
Third, granularity. Site-level ROAS tells you almost nothing you can act on. The decisions that matter — which page, which product, which creative, which placement — need attribution at that level, and most stacks simply cannot produce it.
What to build instead.
A click spine on your own domain. Every outbound link and every entry point routed through infrastructure you control, with a durable click identifier, bot filtering, and a permanent record of the event.
Server-to-server postbacks. Conversions reported from your server to every platform and network that needs them, so attribution survives when the browser can't help.
A warehouse and a model. Raw events landed somewhere queryable, joined to spend and revenue, and rolled into earnings per click, earnings per visit, and revenue per page — reconciled against the finance number until the two agree.
Integrity checks. Automated monitoring for dead links, broken postbacks, and silent tracking failures. Measurement systems fail quietly, and a quiet failure looks exactly like a bad week.
What it costs, honestly.
For a single business with a handful of platforms, expect four to eight weeks of implementation after roughly a week of audit. Ongoing cost is warehouse compute and a small amount of maintenance.
The return is rarely subtle. Most operators find at least one channel they've been overfunding and at least one they've been starving, and the correction usually pays for the build inside a quarter.
The part nobody mentions.
Better attribution will make some numbers look worse. The campaign everyone is proud of may turn out to be riding on branded search. That is the system working, and it's worth deciding in advance that you want the real number rather than the comfortable one.
Every budget argument in a marketing organisation is downstream of measurement. Fix the measurement and most of the arguments simply stop.
- What an AI development company should actually deliverMost AI engagements die between the demo and the deploy. Here's the checklist that separates a vendor who ships from one who presents.
- Build, operate, or partner: choosing the right engagementThree structures, three different distributions of risk. Picking the wrong one is the most common reason good work fails.
- Publishing AI content without wrecking your rankingsThe difference between scale that works and scale that gets you deindexed is not the model. It's the gates.