In short: a Smart Bidding bid is, in essence, conversion probability × target CPA — which is why the same keyword is worth a different amount in every auction. It is fed by your conversions (volume and cleanliness) plus Google's aggregated data. And it has three blind spots: it doesn't see what happens after the conversion, it doesn't see your business, and it doesn't know what you don't want.
At Basic level we saw that Smart Bidding estimates the conversion probability of each auction and bids in proportion. The Intermediate level starts by opening the box a little further: which signals it uses, how it turns them into money and, above all, what it cannot see. Because managing Smart Bidding well is not letting it get on with it: it is feeding it what it needs and covering its blind spots.
How does Smart Bidding turn signals into a bid?
In every auction the system has dozens of signals: the exact query, device, operating system, browser, physical location and location intent, time and day, language, the user's history with your website (lists), audience membership, the ad and the landing page it will show, the match type that triggered it, the position it could occupy... The model combines them all into an estimate: p(conversion) and, if it bids on value, expected value.
The bid comes out of that:
Bid ≈ p(conversion) × target CPA (target CPA)
Bid ≈ expected value ÷ target ROAS (target ROAS)
With a target CPA of €40 and an auction with an 8% probability, it bids around €3.20; with 0.5%, €0.20. That is why the same keyword can cost €4 at 11 in the morning on desktop to a returning visitor and €0.15 at midnight to a stranger on mobile. Manual bidding could never replicate that variation.
With no target (maximise conversions/value), the system spreads the budget across the auctions with the best p(conversion) per euro until it runs out: the budget is the brake.
What data feeds Smart Bidding?
Three sources, in order of weight:
- Your conversions: each one says "this combination of signals ended well". The more there are, the better it predicts; the cleaner they are, the better it learns. Hence the minimum volume (30-50 a month per strategy) and the obsession with measurement.
- Google's aggregated data: conversion patterns from millions of similar advertisers. It lets a new campaign avoid starting from zero — and it also lets Smart Bidding "know" things about your sector that you do not.
- What you tell it: target, budget, remarketing and customer lists, seasonality adjustments, data exclusions, conversion values.
What does the bid strategy report show?
In Tools → Bidding strategies → the report for each strategy, Google shows the top signals: the dimensions in which the strategy is differentiating its bids the most (device, location, time, day, specific queries, lists). It is the only window into "what it has learned": if you see it bidding hard on a remarketing list and on certain queries, that is where your customer is. Use it to confirm or correct your hypotheses, not to adjust bids (you cannot, and you should not).
What doesn't Smart Bidding see?
Smart Bidding sees the auction and the conversion. It does not see:
1. What happens after the conversion
A spam form and a €10,000 customer weigh the same if they have the same conversion value. The algorithm optimises towards "forms" until you give it back the truth: offline conversions with values (the Measurement II module). Until you do, the blind spot is yours.
2. Margin and the business
It sees revenue if you send it; it does not see the margin, the stock, your team's capacity or the seasonality of your demand (unless you warn it). A target ROAS of 4 is profitable for one product and ruinous for another: deciding that is your job.
3. What you do not want
Negatives, location exclusions, brand exclusions, audience exclusions: the algorithm optimises within whatever you allow it. If you allow it everything, it will spend on everything that converts, including your own brand and the searches you would rather not pay for.
Covering those three points is, essentially, all the management work left in 2026.
💡 Ninja trick: the three blind spots are the Suite's three axes: Lead Rating gives Google back the real value of each lead (point 1); Smart Bidding (SBNS) judges each campaign against the business target, not against the strategy target (point 2); and negatives, the white list and Ninja Shield close the doors (point 3). Google's algorithm does its part very well; ours is to give it truth and limits.
What you should remember
- Bid ≈ p(conversion) × target CPA (or value ÷ ROAS): the same keyword is worth something different in every auction.
- It is fed by your conversions (volume and cleanliness), Google's aggregated data and whatever you tell it.
- The top signals show what it has learned: use them to understand, not to meddle.
- Blind spots: after the conversion, the business, what you do not want. That is where your work is.