The errors, roughly in order of cost
- Backfilled guesses — someone assigns a plausible campaign to an unknown lead; the guess is indistinguishable from evidence forever.
- Thread breaks at conversion — a new customer or job record loses the origin, so revenue cannot reach back.
- Last-write-wins merging — the second touch overwrites the first, quietly converting the system to last-touch.
- Context lost between landing and submission — campaign traffic is recorded as direct.
- Untracked channels — signage, wraps, word of mouth and print are absorbed into whichever tracked channel the person used last.
- Window mismatch — a 30-day rule applied to a 60-day sales cycle discards the revenue it should credit.
How to spot each one
Most of these are visible in the shape of the data rather than in any single record. An unattributed share that is suspiciously low is a sign of backfilling, not of good capture — a healthy service business has a real unattributed segment, because word of mouth exists.
A direct-traffic segment that grows in step with paid spend means the parameters are being dropped somewhere on the path. And sold jobs that cannot reach a lead record are the direct measure of thread breakage — count them, do not estimate them.
Sizing the damage honestly
The cost of an attribution error is not the inaccuracy; it is the budget that moved because of it. If a channel was under-credited by a third for two quarters and was defunded as a result, the loss is every job that channel would have produced, less the spend saved.
That framing also tells you which errors to fix first: the ones touching the channels with the largest spend. An error in a channel receiving two percent of the budget can wait.
A quarterly audit that actually finds things
- Trace five sold jobs end to end, by hand, back to the original contact record and its captured origin.
- Compare the platform's conversion count to the CRM's lead count for the same window and explain the gap.
- List every distinct value in the source field and delete or merge the near-duplicates.
- Confirm the attribution window still matches the measured sales cycle.
- Check that a campaign link still delivers its parameters into a submitted record, on mobile as well as desktop.