Why Google Ads Smart Bidding Fails (And How to Fix It)

The pattern always looks the same from the outside. Someone reads that Target ROAS beats manual bidding, flips the switch, and watches ROAS drop for a week or two. They read that as proof the algorithm can't handle their account, revert to manual, and performance drops a second time because now the campaign is relearning again from a worse starting point.

Nobody in that sequence asked whether the account had the data, the conversion setup, or the patience the algorithm actually needed. That's the audit question I ask first, before I look at a single bid.

At every single auction, Google's bidding models read dozens of realtime signals: device, time of day, location, browser behavior, whether that specific user has visited your site before. From that it calculates what it thinks is the optimal bid for that one impression, thousands of times a second across the whole network.

That process works well when there's solid historical data to calibrate against. When the data is sparse, wrong, or mixed with the wrong signals, the model makes bad decisions with confidence, because it has nothing better to go on. That's the root cause behind most of the smart bidding failures I've audited: not a broken algorithm, a starved one.

The three problems below are the ones I find over and over, roughly in order of how often each one shows up.

Google's own guidance says a campaign needs at least 30 conversions in 30 days before Target ROAS starts to perform reliably. For Target ROAS specifically, I push that number higher, to 40 or 50. Below that, the algorithm is flying on a very limited signal, and you'll see it in a week over week analysis as constant swings, good one week, bad the next, for no reason you can point to.

If you're getting 10 purchases a month, Target ROAS will almost certainly underperform. Not because the AI is bad. Because there isn't enough data for it to work with, full stop.

The right move at lower volumes is Maximize Conversion Value with no target set. It's less constrained, it functions at lower data volumes, and it still optimizes for revenue rather than volume. Once you're consistently hitting 30 to 50 purchases a month, that's when it makes sense to add a ROAS target.

This one is subtle and I see it constantly. An account shows 200 conversions a month. Looks great on paper. Dig in, and 170 of those are soft micro-events: page views, newsletter signups, phone call initiations. Only 30 are actual purchases.

The algorithm optimizes for whatever sits in the conversions column, especially the ones marked as primary. If most of what's there is soft events, it will happily send traffic that generates soft events, because they're cheap. That traffic clicks but doesn't buy. ROAS looks inflated in the dashboard while actual revenue stays flat, and you won't catch the gap unless you go looking for it.

The fix is that your primary conversion should be purchase, and only purchase. Everything else, newsletter signups, add to carts, phone calls, goes to secondary.

If your tracking itself is the issue, meaning the purchase event isn't firing cleanly in the first place, that's a separate problem from bidding, and I'd fix it before touching any strategy. I've written a full walkthrough on setting up and testing Shopify conversion tracking, and a broader Google Ads audit checklist that covers the conversion side alongside everything else worth checking.

Every time you change a bidding strategy, the campaign enters a learning period. The algorithm is recalibrating its model from scratch, and performance usually gets worse for a few days, sometimes up to two weeks. That's normal and expected, not a sign anything is broken.

The problem is what people do next. They see the drop, panic, and switch back, which starts a new learning period. Then they switch again. I've audited accounts that changed bidding strategy ten times in a single month and were essentially permanently stuck in learning mode. The campaign never got a real chance to find its footing.

My rule is a minimum of two to three weeks before evaluating any strategy change, compared against an equivalent prior period. If it's still underperforming after that, you have real information based on real data. Pull the plug before that window closes and you're reacting to noise, not results.

Strip away the specifics and smart bidding needs exactly three things to work: enough conversion volume before you push it toward a target, a clean primary conversion action that's only purchase, and enough undisturbed time to actually learn. Get those three right and the algorithm does the job it was built for. None of that requires a smarter algorithm. It requires an account that's actually ready.

Watch the full breakdown below, including the exact conversion data threshold and how I check the conversions table on a live account.

If you want to know whether your own account has these three things in place, that's exactly what I check in a real client Google Ads audit: conversion action hygiene, bidding history, and data volume against what the strategy actually needs.