Smart Bidding is a signals problem: what the algorithm actually runs on

Which Smart Bidding strategy you pick matters less than the conversion data you send it. Get the four strategies in one table, then the five inputs that decide whether any of them work: conversion completeness, value variance, freshness, consent-modeled data, and volume. Audit the inputs before you blame the algorithm.

Kay Vink
Kay Vink

You switch a campaign to tROAS hoping for better numbers, and the numbers get worse. The reflex is to blame the algorithm or your target. Usually the problem is upstream of both. Smart Bidding is Google Ads' auction-time automated bidding (target CPA, target ROAS, Maximize conversions, Maximize conversion value), each strategy setting a bid for every individual auction from signals like device, location, time, and audience (About Smart Bidding). Every one of them optimizes toward the same thing: the conversion data your account sends it. So the strategy you pick matters less than whether that data is complete, valued, and fresh. This page is about the data, because that's the part the strategy menus never mention.

#The strategy menu in one table

Choosing a strategy is a one-table decision. Everything after the table is about the inputs.

StrategyOptimizes towardUse whenThe catch
Maximize conversionsMost conversions within budgetEvery conversion roughly equal; you want volumeWill happily spend the full budget on cheap conversions
Target CPA (tCPA)Conversions at your average cost target (About Target CPA)Equal-value conversions with a known acceptable costInherits CPA's blindness: cost of a conversion, not value of the conversion
Maximize conversion valueHighest total conversion value within budget (doc)Conversions carry real, varying valuesWithout values it silently degrades to counting
Target ROAS (tROAS)Conversion value at your return target (About Target ROAS)Valued conversions + enough volume (Google's floor: ~15 conversions/30 days for most campaign types)The most input-hungry strategy on the menu

Two structural notes. In today's Google Ads UI, tCPA and tROAS are implemented as optional targets on the Maximize strategies rather than standalone entries: same optimization, one menu level down. And Enhanced CPC, the halfway house between manual and Smart Bidding, was retired in March 2025; the migration path Google leaves you is the table above.

The count-vs-value column split is the real fork, and it's an argument about your data rather than the algorithm. Value-based bidding needs values you can trust makes the case for the value side and for when you're not ready.

#The five inputs Smart Bidding actually runs on

Here's the part the strategy docs skip: signals in, spend out. You won't find this list in Google's guidance, because "just turn it on" is a cleaner story without it. But five inputs decide whether Smart Bidding works, and every one of them is your responsibility, not Google's:

  1. Conversion completeness. The model only learns from conversions it sees. Missing tags, consent losses, and broken imports don't just distort reporting; they distort the training set.
  2. Value variance. Value strategies need value differences. A feed where every conversion is $100 gives Maximize conversion value nothing to rank.
  3. Freshness. Smart Bidding weights recent data. Conversions that arrive late (CRM imports on a lazy cadence, long lead-to-close lags) mean the model is always bidding on a delayed picture of the account.
  4. Consent-modeled data. In consented markets, part of your conversion feed is modeled rather than observed. That's a feature, not a bug, but only if your consent mode setup actually enables modeling (Consent Mode v2 without losing your signal).
  5. Volume thresholds. Below minimum conversion volume the model can't converge: bids swing, learning periods stretch, and tROAS has too little to optimize toward.

#What breaks when each input is wrong

The failure table is the audit. Each row is runnable against your own account this week:

InputFailure modeWhat bidding doesCheck it with
CompletenessTag gaps, consent losses uncompensatedUnderbids segments that actually convert (it never saw them convert)The conversion tracking QA checklist: test it like you'd test code
ValuesFlat or wrong valuesBuys cheap conversions dressed as value optimizationValue-variance report; Value-based bidding needs values you can trust
FreshnessOffline imports days lateBids on stale truth; every import rewrites recent performanceImport recency; Offline conversion tracking: send CRM deals back to Google Ads
Consent-modeledBasic consent mode (no modeling)EEA conversions vanish instead of being modeled, so the model undervalues whole geographiesConsent Mode v2 without losing your signal
VolumeBelow learning thresholdsExtended learning, volatile bids, tROAS with too little to work fromConversions/30 days vs the strategy's floor

The freshness row deserves the concrete version. An account whose CRM import runs every few days hands Smart Bidding a moving target: conversions land against past click dates, so the model's most recent window, the one it weights hardest, is systematically the emptiest. The shape it produces (illustrative, but every late-import account draws it): imports landing ~5 days after close, tCPA reading the empty window as a performance collapse, bids ratcheting down mid-week, then "recovering" when the file lands. The result is a sawtooth that tracks the upload schedule, not the market.

#How to fix each input

Every input has a fix, and each one is its own discipline:

#Strategy misbehaving? That's a different page

This page owns the inputs. If a running strategy is acting up (tCPA creeping past target, tROAS spending nothing, Maximize conversions blowing budget on junk), that's an operations problem with its own diagnostic tree, covered in our bid-strategy troubleshooting guide. The order of operations still holds, though: before you touch the strategy, audit the inputs. Most "the algorithm is broken" tickets are one of the five rows above in disguise.

#It was never really the algorithm

Keep this once the current fire is out: almost none of it was about Smart Bidding specifically. Any system that bids on your behalf at auction time, from Meta's Advantage+ to whatever launches next, optimizes toward the conversion data you send it and inherits every gap in that data. The five inputs are the same wherever the bidding happens. You didn't tune an algorithm. You learned to audit the signal it runs on.

The catch is that inputs drift silently. A tag breaks in a checkout redesign, a CRM import slips a day, a consent banner change quietly stops modeling, and none of it throws an error. It surfaces weeks later as budget pointed the wrong way. Auditing the five inputs once is easy; the hard part is remembering to re-check them every week, on every account, which is exactly the part worth handing off. Connect your accounts to Buron and it watches the inputs Smart Bidding depends on (conversion coverage, import freshness, value sanity, consent-modeled share) continuously, so a broken input lands as a dated finding in your inbox before it becomes a month of misdirected spend. *[Connect your accounts →]

Frequently asked questions

What is Smart Bidding in Google Ads?

Smart Bidding is Google's auction-time automated bidding: strategies like target CPA, target ROAS, Maximize conversions, and Maximize conversion value that set a bid for every individual auction using signals such as device, location, time, and audience, all optimizing toward the conversion data your account sends them.

Which Smart Bidding strategy should I use?

Maximize conversions (optionally with a target CPA) when every conversion is roughly equal in value; Maximize conversion value (optionally with a target ROAS) when conversions carry real, varying values. The stronger determinant isn't the menu choice; it's whether your conversion data is complete, valued, and fresh.

Related