Estimating competitor spend from auction insights alone is a nightmare, totally with you on that. The reverse engineering approach using your own CTR, CPC, and quality score is something I've been digging into too-it's one of those angles that should be more obvious but everyone seems to skip.
The real trick is isolating the variables correctly. If you know your actual spend, your CTR, and your impression share, you can backfill theirs with basic algebra. But you absolutely have to normalise for quality score or your numbers drift off into fantasy land. Seasonality messes with it as well-competitors bid differently by hour and day, so your estimate is only solid if you're comparing the same slot.
The most reliable method I've found is running a tiny test campaign with max CPC bids, then plotting the impression share curve against what auction insights shows. That gives you a much clearer picture.
I came across a tool recently that's trying to automate a lot of this kind of analysis-interesting concept, but early days.
how close have your estimates been when you've actually been able to verify them against real spend?
