Home / Notes

Smart Bidding on incomplete signal: an anonymized breakdown

Smart Bidding doesn't grade its inputs — it optimizes toward whatever conversion signal it's given, accurately and relentlessly, wrong or not. Value-based strategies make a bad signal worse, not better, because they chase incorrect dollar amounts, not just incorrect volume. The fix is verification before automation, not a bidding setting.

"The bidding strategy isn't working" is the complaint I hear. The account is usually right that something isn't working — it's rarely the bidding strategy. Here's what an anonymized breakdown of a real signal-verification project shows, and the sequence that actually fixes it.

What "incomplete signal" actually looks like

Not usually a dramatic failure — a slow accumulation of small ones. A form-fill conversion action that fires on every submission, spam included. A "phone call" conversion counted the moment someone clicks a call button, whether or not the call connects. Default conversion values applied uniformly, so a $200 consult and a $20,000 procedure register as the same signal. None of these show up in the Google Ads interface as errors — the account looks like it's converting normally. They only surface when you reconcile against what the CRM actually closed.

An anonymized breakdown: before and after

VERIFIED EXAMPLE · When I inherited a wasteful account at a luxury US healthcare clinic, it was optimizing toward platform conversions. Rebuilt around admitted patients imported from the CRM, cost per admitted patient fell ~78% while spend scaled ~2× and approved patients rose ~9×.

The sequence that produced that outcome wasn't a bidding change first. It was signal verification first: every conversion action audited against CRM outcomes, duplicates and junk actions removed, the remaining actions reconciled to what "admitted patient" actually means in that CRM. Only after that reconciliation did bidding strategy and account structure change. Structure and bidding sitting on top of a bad signal just make the wrong number move faster — the order matters as much as the individual fixes.

Why value-based bidding amplifies the problem

Standard Smart Bidding chases volume against a target; value-based bidding chases the dollar amounts you assign per conversion. That's a meaningful upgrade when the values are real — the algorithm learns to prioritize your highest-value procedures over routine consults automatically. It's a meaningful downgrade when the values are defaults or guesses, because now the algorithm optimizes toward fictional revenue with total confidence.

RAW SIGNALmixed quality, unweightedSMART BIDDINGlearns exactly what it's fedREAL PATIENTSclean signal, verifiedOPTIMIZED NOISEdirty signal, no judgmentTHE ALGORITHM DOESN'T GRADE ITS INPUTS — YOU DO, BEFORE YOU TURN IT ON
Smart Bidding doesn't grade its inputs. Verification has to happen before automation, not after.

Google's own walkthrough of value-based bidding is worth watching with this in mind — the mechanism is sound; what it depends on is the values you feed it being true.

From the source: Google Ads Tutorials: Introduction to value based bidding in Search.

The verification sequence

In order: define what a real conversion is in business terms — booked and showed, not just booked; admitted, not just consulted. Reconcile a full month of platform-reported conversions against CRM records for that same definition; expect a gap, the question is its size and cause. Remove or fix conversion actions that don't map cleanly to the business definition. Only then adjust bidding strategy — and only then trust what the account tells you about performance. The $1,500 Starter tier of the rebuild service is exactly this sequence, delivered as a plan your team can execute.

The common failure mode

The failure I see most: an account switches to Target CPA or Target ROAS on unverified data, performance looks great for two to three weeks as the algorithm exploits the signal it's been given, then plateaus or craters as it exhausts the audience that matched the wrong pattern. The postmortem usually blames the bidding strategy. The bidding strategy did exactly what it was told.

VERIFIED CONTEXT · Value-based Smart Bidding is only as honest as the values behind it — the discipline above is what turns "the algorithm isn't working" into a signal problem you can actually fix.

What clients say

"Precisely the kind of results-oriented partner you want managing significant investment."

Brandon MitchellBrandon Mitchell ↗ · Mid-Market Partnerships, Google

Common signal defects and what they teach the algorithm
DefectWhat it looks likeWhat Smart Bidding learns
Spam counted as conversionEvery form submission fires the actionChase more of whatever produces submissions
Call-click counted as connectNo duration or outcome filterChase more clicks, not more calls
Uniform default valuesEvery conversion worth the sameNo preference for your highest-value patients

Questions owners ask

How do I know if my conversion signal is "incomplete"?

Reconcile one full month: total conversions Google Ads reports against records in your CRM that meet your real definition of a conversion. A gap of 10–20% is typical software noise. A gap of two to five times is a signal problem, not noise.

Should we turn off Smart Bidding while we fix tracking?

Not necessarily switch it fully off, but stepping down to a simpler strategy — maximize clicks, manual CPC — while tracking is being corrected avoids compounding a bad signal. Move back to Smart Bidding once the signal is verified.

Does this apply to Performance Max too?

More so. PMax leans on Smart Bidding across every channel it touches, so a bad signal there gets amplified across search, display, and placements simultaneously, not just one campaign type.

Not sure if your signal is trustworthy?

Send me a month of Google Ads conversion data next to the matching CRM export. Thirty minutes and you'll know whether Smart Bidding is learning from truth or from noise.

$500 audit + 90-day roadmap · direct with me, no account managers.

Book a 30-minute call