I've seen campaigns double conversions while revenue dropped. I've watched winning campaigns fill CRMs with leads that never close. I've hit target CPA and still missed business goals.

The problem is simple: ad platforms don't know what a good customer looks like. They only know what you tell them. Tell Google a form fill is success, and it'll find more form fills. If half those leads are junk, that's not Google's problem-it did exactly what you asked.

Scariest part? An account can look healthier every month while the business gets worse. More conversions. Lower CPA. Higher conversion rate. Less revenue.

At some point you realize you're not managing campaigns-you're training algorithms. And the output quality depends entirely on the signal you feed in.

So here's my question: What's the most misleading KPI you've ever optimized for?

For me, it's Cost Per Lead (CPL) in B2B. A campaign can generate hundreds of "cheap" leads and look fantastic in Google Ads or LinkedIn reports, while sales quietly complains none of them are decision-makers, budget holders, or serious buyers. I've seen campaigns with 40% lower CPL, 2x more leads, higher conversion rates-yet pipeline value dropped because the leads were students, competitors, job seekers, or outside ICP.

In B2B, the KPI that matters isn't CPL-it's Cost Per Qualified Opportunity or even Cost Per Closed-Won. The mindset shift is closing the loop between marketing and sales by feeding CRM outcomes back into the ad platforms. Once Google or LinkedIn optimizes for SQLs, opportunities, and closed deals instead of form fills, the algorithm gets dramatically smarter.

I'd rather have 50 leads, 20 SQLs, 5 closed deals than 500 leads, 10 SQLs, 1 closed deal. Every day.

But here's the part that bites specifically in B2B: the accurate signal (revenue) is too slow and too sparse for the algorithm to learn from. Your deal closes 3-9 months after the click, and you might get 10-20 closed-won a month. Google's optimization needs volume and speed. Feed it only closed-won revenue, and you starve it-too few events, too delayed-and it can't optimize at all. The accurate signal (revenue) is unlearnable. The fast signal (form fill) is noise. That's the real bind.

So the actual craft isn't "feed revenue." It's engineering a proxy-a mid-funnel event that's fast and frequent enough to train on, but qualified enough to correlate with revenue. SQL, qualified meeting booked, lead score above a threshold. You're deliberately picking the best available trade-off between accuracy and learnability. That's the whole game in B2B PPC, and almost nobody frames it that way.

The garbage-in problem starts before the ad platform. Most companies optimize conversions because they never actually defined what a good customer looks like precisely enough to feed anything better. The platform optimizes the only thing you gave it-because you didn't give it anything better. You can't feed a good signal until you've done the upstream work of defining "good."

The biggest lie under your biggest lie is "we'll let the data tell us," when the data only knows what you already decided to call success.