Q: My traffic has a high fraud score and I don't know what's driving it. How do I break it down?

Q: My traffic has a high fraud score and I don't know what's driving it. How do I break it down?

A: A fraud score (a 0-100 risk rating from your anti-fraud tool) is an average hiding the real offender. Decompose it:

— Step 1: Export scores per lead, not the campaign average. Averages lie.
— Step 2: Group by sub-source. Usually one or two sub-IDs carry the bad weight.
— Step 3: Read the score reasons. Most tools list factors: proxy detected, geo mismatch, velocity, recycled email.
— Step 4: Rank the factors by frequency. If 'proxy' fires on 70% of bad leads, your problem is proxy traffic, not a vague 'quality' issue.
— Step 5: Pause the worst sub-source and re-measure. The score should drop fast.
— Step 6: Take the factor breakdown to that traffic source — they can often fix it upstream.

Short version: per-lead, group by sub-ID, read the factors, pause the worst, re-measure.

Still stuck? Drop your case in the comments.