Smart Bidding Explained: Handing Google Ads to the Algorithm

What Smart Bidding actually is

Smart Bidding is Google's name for bid strategies that set a different bid for every single auction, using signals no human can act on at that speed: device, location, time of day, browsing behaviour, audience membership and dozens more. Instead of you deciding that a click on "emergency electrician birmingham" is worth £4, the system predicts the conversion probability of this specific search by this specific person and bids accordingly.

Google has steadily removed the middle ground. Enhanced CPC, the halfway house that nudged manual bids up and down, was retired in early 2025, leaving a starker choice: bid manually with full control and no machine learning, or hand over bidding entirely. That makes understanding the automated strategies less optional than it used to be.

The four strategies and what each optimises

• Maximise conversions: spends your full daily budget chasing as many conversions as possible, with no regard for what each costs. Fine for launching; dangerous left uncapped
• Target CPA (tCPA): now set as an optional target within Maximise conversions. You name an acceptable cost per acquisition and the system seeks volume at that average
• Maximise conversion value: like Maximise conversions, but optimises for total value rather than count, so a £900 order outweighs three £100 ones
• Target ROAS (tROAS): a return target layered on Maximise conversion value. You ask for, say, 400% return on ad spend, and the system prioritises auctions likely to hit it

Count-based strategies (Maximise conversions, tCPA) suit lead generation where enquiries are roughly equal in value. Value-based strategies (Maximise conversion value, tROAS) suit ecommerce, or lead generation where you have assigned honest values to different enquiry types.

The data thresholds nobody respects

Smart Bidding is prediction, and predictions built on thin data are guesses with confidence. The practical thresholds matter more than any feature.

• Below roughly 15 conversions a month: stay on manual CPC or uncapped Maximise conversions with a modest budget; targets will just starve delivery
• Around 15 to 30 conversions a month: tCPA becomes workable, though expect volatility week to week
• tROAS wants meaningfully more, ideally 50-plus monthly conversions with genuine variation in order values, before its return predictions stabilise
• Too few conversions across several small campaigns? Portfolio bid strategies pool them so the algorithm learns from combined data

If your account converts eight times a month, the honest answer is that no target-based strategy will behave, and your effort is better spent increasing conversion volume: broader keywords, better landing pages, or counting a higher-funnel action as the conversion.

Fix your conversion tracking before touching anything

Smart Bidding optimises to whatever your conversion column contains, so errors get amplified rather than averaged out. Before migrating, audit the basics: primary conversion actions should be genuine enquiries or purchases only, with soft actions like newsletter sign-ups demoted to secondary so they inform but do not steer bidding.

Enable enhanced conversions, and configure consent mode properly, since UK consent banners mean a slice of your conversions is modelled rather than observed. For lead generation, assign proxy values to different enquiry types, and consider offline conversion imports so the system eventually learns which leads became paying customers. A fortnight spent here is worth more than a year of bid tinkering.

A migration path with guardrails

• Step 1: run 30 days on your current setup and record baseline cost per conversion, conversion volume and spend
• Step 2: switch to Maximise conversions with no target, keeping your daily budget unchanged as the only brake on spend
• Step 3: leave it alone for two weeks; the learning period is real, and every significant edit partially resets it
• Step 4: once performance settles, add a tCPA target set 10 to 20% above your observed cost per conversion, giving the algorithm headroom
• Step 5: tighten the target gradually, no more than 10 to 20% at a time and no more often than fortnightly
• Step 6: move to value-based bidding only once conversion values are tracked and trusted

Expect a wobble during learning: spend can spike and cost per conversion can rise for days before improving. Judge each step on two full weeks of data, never two bad days.

When to intervene and when to sit on your hands

The most common way UK advertisers break Smart Bidding is meddling: changing targets weekly, pausing campaigns during a wobble, then restarting the learning period they just disrupted. Set a fortnightly review rhythm and hold to it.

Do intervene when spend spikes without conversions for a sustained stretch, when a tracking fault is feeding bad data, or ahead of predictable demand shifts, where the seasonality adjustment tool tells the algorithm what is coming better than a manual target change does. And keep the exit in mind: if two months of patient operation cannot match your manual baseline, reverting is a decision, not a defeat. If you would like help auditing your tracking and planning the migration, our team at Thind Global Services can help.