How do you know whether your attribution data is any good?
Ask for the match rate: of all jobs or sales in the period, what share were tied to an identified source? A report that will not show this number is hiding its own uncertainty. When the rate is low, the channel splits are being extrapolated from a minority of records. Track it over time, because a sudden drop almost always means a tracking break rather than a change in the market.
The single number that grades everything else
Attribution reports are usually presented as percentages of a total, which conceals how much of the total was actually identified. If half the revenue has no source, then a channel shown at thirty percent is thirty percent of the identified half, and the other half is being ignored or silently assumed to look the same.
Match rate makes that visible. Compute it as identified records divided by all records for the period, and put it on the report next to the date range. Nothing else you do to improve reporting quality has a better ratio of effort to clarity.
Compute it separately by path
- Web form leads. Should match at a high rate, since a form can carry parameters directly. A low rate here points at capture or redirect problems.
- Tracked calls. Depends on number coverage and pool sizing. Compare matched calls against total inbound calls, not against calls the platform already labeled.
- Jobs to leads. The share of completed jobs that trace back to any lead record at all. This is often the weakest link and the one nobody measures.
- Leads to clicks. The share of paid leads carrying a usable click identifier, which determines what you can upload back to ad platforms.
Treat it as a monitored metric, not a one-time audit
Match rates decay. A site redesign changes a form. A redirect rule gets added. A number pool runs short in peak season. A field service system upgrade renames a field. Each of these breaks matching quietly, and the report keeps rendering.
So chart the rate weekly and alert on a meaningful drop. A step change on a specific date is nearly always a deployment, not a market shift, and knowing the date is most of the diagnosis. This is exactly the kind of thing an automated intelligence brief should be watching so nobody has to remember to check.
What to do with the unmatched portion
Show it. Keep an explicit unknown bucket in every channel breakdown rather than distributing it proportionally across known channels. Proportional redistribution assumes the unknown looks like the known, and it usually does not: unmatched revenue skews toward repeat customers, referrals and offline influence.
If you must model the unknown for a specific decision, do it in a clearly labeled scenario view, not in the standing report. Preserving that distinction is the practical difference between revenue reporting that holds up under questioning and reporting that quietly flatters whichever channel already looked best.
Topics: match rate · data quality · monitoring · reporting
Have a version of this question about your own business?
The useful answer usually depends on which systems you run and how they're connected. That's a conversation, not a blog post.