7 July 2026

Smart Bidding is Google’s machine-learning bidding system, and it sets a unique bid for every single auction using dozens of contextual signals rather than one fixed bid you control manually. The short answer to “how do I control it?” is this: you don’t set bids directly any more — you control the inputs (goals, data quality, structure and guardrails) that the algorithm learns from. Get those inputs right and Smart Bidding will usually outperform manual bidding. Get them wrong, and it will confidently spend your budget in the wrong direction. This guide breaks down how the algorithms actually work, what signals they lean on, and the specific levers you still have to keep them pointed at your business goals. What is Smart Bidding, and Why Did Google Build It? Smart Bidding is the umbrella term for Google Ads’ automated bid strategies: Target CPA, Target ROAS, Maximise Conversions, Maximise Conversion Value, and Enhanced CPC. Instead of a human adjusting bids by keyword or device, Google’s models predict the probability and value of a conversion for every auction and set a bid within milliseconds, using real-time signals a person could never process manually. Google built this because the auction environment became too complex for manual rules. A single search can carry hundreds of contextual signals — device, location, time of day, browsing history, query phrasing, remarketing list membership — and the value of a click can swing wildly depending on the combination. Manual bidding treats a keyword the same at 2 pm on a Tuesday as it does at 11 pm on a Sunday. Smart Bidding doesn’t. How Do Smart Bidding Algorithms Actually Work? Each strategy shares the same core mechanism — a machine learning model trained on your account’s historic conversion data, layered with Google’s cross-account signals — but they optimise for different outcomes. Target CPA (Cost Per Acquisition) You tell Google the average amount you’re willing to pay per conversion, and the algorithm adjusts bids auction-by-auction to hit that average across the campaign. Some auctions will bid well above your target because the model predicts a high conversion probability; others will bid low or skip the auction entirely. It’s an average, not a ceiling, which trips up a lot of advertisers who expect every conversion to cost the same. Target ROAS (Return On Ad Spend) This works the same way but optimises for conversion value rather than volume. You set a target return — for example 400%, meaning £4 in revenue for every £1 spent — and the model bids higher for auctions it predicts will produce higher-value transactions. This only works well when conversion values are actually fed into Google Ads accurately, which is where a lot of ecommerce accounts fall down. Maximise Conversions / Maximise Conversion Value These strategies spend your full daily budget while trying to get the most conversions or value possible, without a fixed target. They’re useful early in an account’s life when there isn’t enough conversion history to set a realistic CPA or ROAS target, and they’re often used as a stepping stone before switching to Target CPA or Target ROAS once the algorithm has learned your account’s patterns. Enhanced CPC (Legacy) A hybrid model that adjusts your manual bids up or down based on conversion likelihood, rather than setting bids entirely automatically. Google has been quietly retiring this in favour of full Smart Bidding, and most accounts should have moved on by now. What Signals Does Smart Bidding Actually Use? Google doesn’t publish the full model, but based on its own documentation and what we see in client accounts at SEO Syrup, the signals broadly fall into a few categories: Device, location, time of day and day of week Browser and operating system Remarketing list membership and audience data Search query characteristics, including intent and specificity Language and interface settings Seasonality and real-world events, where seasonality adjustments have been set Landing page relevance and site performance The critical point for advertisers is that the model needs volume and clean data to weight these signals properly. An account with fewer than roughly 30 conversions a month per campaign gives the algorithm very little to learn from, and bidding decisions will be noisy and inconsistent until that volume builds up. Why Do Smart Bidding Campaigns Sometimes Underperform? In our experience auditing UK accounts, underperformance almost never comes down to “the algorithm being wrong.” It comes down to feeding it the wrong instructions or the wrong data. The most common causes we see are: Conversion tracking counting the wrong actions — for example, counting a phone number click as a conversion when only 1 in 20 of those calls actually becomes a customer Targets set from wishful thinking rather than historic account data, causing the algorithm to restrict spend to near zero while it tries to hit an unrealistic CPA Switching bid strategy or changing targets too frequently, which resets the learning period and keeps the campaign permanently unstable Budget constraints that prevent the algorithm from bidding competitively in the auctions it identifies as high value Mixing high-intent and low-intent conversion actions in the same “conversions” column, diluting the value signal How Can You Control Smart Bidding Without Fighting It? You can’t move individual keyword bids any more, but you have more control than most advertisers realise. Here’s where that control actually lives. 1. Fix Conversion Tracking Before Touching Bid Strategy This is the single highest-leverage lever available. Smart Bidding is only as good as the data it’s optimising towards. Before adjusting any target, audit what’s being counted as a conversion, apply conversion value rules to weight leads by quality, and where possible, feed offline conversion data back into Google Ads so the algorithm learns which leads actually turned into paying customers. 2. Set Realistic Targets Using Your Own Historic Data Pull your actual average CPA or ROAS from the last 30–90 days before setting a target, then adjust it gradually — typically no more than 10–15% at a time. A target set far below