Approach
Most underperforming accounts aren't badly optimized. They're optimized correctly toward the wrong number.
An account bidding hard toward a conversion action that fires on every page view is not a bidding problem. It's a measurement problem wearing a bidding problem's clothes, and adjusting targets makes it worse — the algorithm gets better at finding the wrong thing.
So the first two weeks of any engagement are spent on the least glamorous work available: auditing conversion actions, checking for double-counting across tag managers, confirming that what the platform calls a conversion matches what your finance team calls revenue. It is common to find three tags reporting the same event, or a legacy container still firing alongside its replacement.
This is also why I won't quote a performance target in a first conversation. Until I know whether your numbers are real, any number I promise is theatre.
Most accounts optimize toward a form fill or a phone call, because that's what the platform can see. But leads are not equally valuable, and the campaigns that produce the most of them are frequently the ones producing the least revenue.
The fix is offline conversion import: passing the click identifier into your CRM, attaching it to the closed deal, and sending the actual value back to Google and Microsoft. Bidding then optimizes toward money instead of activity.
It takes real work — usually a few weeks of coordination with whoever owns your CRM — and it is the single most durable advantage available in paid search, because a competitor with a bigger budget still can't outbid better information.
Campaign structure should reflect how the business actually makes money, not what's convenient to report. If your repair jobs average $700 and your replacements average $7,000, those don't belong in the same campaign with the same target — the algorithm will happily buy you volume in whichever one is cheaper, and cheaper is rarely more profitable.
The same applies to geography, seasonality, and capacity. An account that overbuys demand its operations team can't service converts new customers into one-star reviews, which costs more than the wasted spend did.
Smart bidding and AI-driven matching genuinely outperform manual management in most accounts — I've migrated large accounts to automated bidding and the results were unambiguous. But they outperform on the objective you give them, using the signals you supply.
Handing a platform a broad match keyword list, weak negatives, and a lead-count target and then blaming the automation for irrelevant traffic is a category error. The work moved; it didn't disappear. It's now in feeding the system correctly and constraining where it's allowed to go.
The AI Max migration is exactly this problem at scale. Accounts that already ran campaign-level broad match get keywordless expansion added on top of an all-broad keyword list, with no tighter keyword base anchoring it — the profile independent data treats worst. Negative mining and brand exclusions matter more this quarter than they have in years.
You own the accounts. Documentation is written as we go, not assembled at the end. Reporting explains reasoning, so you can evaluate the thinking rather than take the numbers on faith.
Consultants who make themselves hard to replace are managing their own risk, not yours. The version where you could hand this to someone else next month without losing anything is the version where you keep me because the work is good.