Most Smart Bidding issues are not bidding issues. They are problems with the data the algorithm learns from, the budget it works within, or the number of times someone changes the settings. If your Google Ads results dropped after switching to Target ROAS, Target CPA or Maximise Conversion Value, start by auditing your conversion actions, then your budget and targets, then your patience. This guide walks through each in that order, with the checks we run on eCommerce accounts.
TL;DR
- Smart Bidding optimises for conversion volume or conversion value, so it chases whatever you tell it counts as a conversion. Wrong signals produce wrong results.
- Audit primary and secondary conversion actions first. A page view or add-to-cart set as a primary conversion is a common way to poison the account.
- Revenue-based ROAS can look healthy while profit does not. Feed margin data back to Google where you can.
- Judge performance over a longer window, and allow one to two full conversion cycles after major changes before touching targets again.
- Budget that is too tight, targets that never move with the market, and constant tinkering are the three most common self-inflicted problems.
Table of Contents
- What Smart Bidding actually optimises
- Issue 1: Polluted conversion signals
- Issue 2: Revenue, ROAS and profit tell different stories
- Issue 3: Budgets and targets that choke the algorithm
- Issue 4: Judging results too early, or changing too often
- Issue 5: Campaign types that amplify the problem
- A diagnostic order that saves time
- Our perspective
- Want a second pair of eyes on your account?
- FAQ
- Sources
What Smart Bidding actually optimises
Smart Bidding uses machine learning to set bids in real time for each individual auction, aiming at conversion volume or conversion value. To predict the likelihood of a conversion, it uses auction-time signals such as device, location, time of day, browser, operating system and language settings.
That sounds like a perfect fit for an online store. The catch is that the algorithm does not know your business. It knows the numbers you feed it. It does not know that a $12 accessory and a $900 lounge have very different margins, that returns eat into a category, or that your best customers buy three times a year. It optimises toward a proxy for your goal, and when the proxy is weak, the results fall short of what you wanted.
That is why we rarely tell a client to abandon Smart Bidding. We tell them to fix what it is learning from.
Issue 1: Polluted conversion signals
The most common cause of poor Smart Bidding performance is a messy set of conversion actions. Google’s own display guidance notes that Smart Bidding adjusts and reacts to inaccurate conversion data, which can cause swings in volume, CPA or ROAS. Search Engine Land makes the same point with a concrete example: accidentally setting a page view as a primary conversion can derail automated bidding.
What a clean setup looks like
For a typical eCommerce store, the primary conversion should be the purchase, with a transaction value passed through dynamically. Everything else is context. Add to cart, begin checkout, newsletter sign-ups and product views are useful for reporting, but if they sit in the same pool as purchases, the algorithm treats a cheap sign-up as a win and bids for more of them.
A quick audit you can do today
- Open Goals, then Conversions, then Summary, and list every action marked as primary.
- Ask of each one: would I pay the same amount for this as for a sale? If not, move it to secondary.
- Check for duplicates. A purchase counted by both a Google tag and an imported GA4 event inflates conversions and teaches the algorithm the wrong CPA.
- Confirm the purchase value is dynamic, not a fixed number, and that currency is correct.
- Look at the date of the last recorded conversion for each action. A tag that broke during a theme update on Shopify or WordPress will quietly starve the algorithm.
If you want a fuller walkthrough of reading account data after the audit, see our guide on how to measure Google Ads performance.
Low volume and micro-conversions
Stores with few monthly sales give the algorithm little to learn from. Search Engine Land notes that accounts with low conversion volume may need a carefully built micro-conversion strategy. The word carefully matters. A micro-conversion is only worth using if it genuinely correlates with purchase, such as reaching the payment step, and it should be introduced deliberately rather than by promoting every available event.
Leads and offline sales
If you run a lead generation funnel, for example quote requests for furniture or wholesale enquiries, the problem is different. Lead accounts often lack offline conversion data, so Google cannot tell a high-value sale from a low-quality lead. Without a captured click ID, offline conversion imports are not possible and CRM outcomes cannot be tied to ad clicks. Understanding how pixel tracking identifies leads is a good starting point for closing that gap, because once qualified leads flow back into Google Ads, the algorithm can finally prefer the enquiries that turn into revenue.
Issue 2: Revenue, ROAS and profit tell different stories
A platform ROAS of 5.0 sounds excellent. Whether it is depends on what sits in the numerator. Conversion value is whatever you tell Google a conversion is worth. If you send revenue, ROAS is revenue divided by ad cost. If you send profit, you get a very different number for the same order. One illustrative example from a conversion value glossary: an order of 100 with 40 in profit and 25 in ad spend is a ROAS of 4 on revenue but 1.6 on profit.
For an eCommerce brand, this matters in three ways.
- Margin differences between products. Revenue-based bidding favours whatever sells most dollars, not whatever earns most. A high-ticket, low-margin line can soak up the budget.
- Returns and cancellations. The tag records the sale at checkout. The refund comes weeks later and never reaches the algorithm unless you adjust for it.
- New versus returning customers. A first purchase that costs more than its margin can still be a good acquisition if lifetime value is high. The default setup cannot see that either.
Google offers a route to closer alignment. Conversions with cart data let you send the products sold in each transaction and, combined with cost of goods sold from your Merchant Center feed, unlock profit reporting alongside revenue. Even if you do not change your bidding goal, that reporting shows you where revenue ROAS and real profit part ways.
If you want to see the gap for your own numbers, our Paid Advertising Profit Calculator is a simple place to start. For related levers, read how to lower cost per acquisition in digital ads.
Issue 3: Budgets and targets that choke the algorithm
Smart Bidding needs room to bid. When a budget is too low, the system can only enter a narrow band of auctions, and the volume it gathers is thin and skewed. Google’s own guidance for Demand Gen campaigns states that the budget should be at least 10 times the target CPA and that a low daily budget is usually the top reason a new campaign underperforms. That figure is for Demand Gen, so do not apply it blindly to every campaign type, but the principle holds: if your daily budget barely covers one or two conversions at your target CPA, the algorithm has almost nothing to learn from.
Targets that are too aggressive
A target ROAS set far above what the account has achieved historically does not make the algorithm find better customers. It makes it bid less, enter fewer auctions and spend less. The symptom is falling impressions and an unspent budget, with a ROAS that looks fine but a store that is quieter. Set the first target near the account’s recent actual performance, then move it in modest steps.
Targets that never move
The opposite problem is a target set once and forgotten. Seasonal peaks such as Black Friday, Boxing Day and end of financial year in Australia change conversion rates and competition. A target that suited a quiet August can be too loose or too tight in November. Review targets against the season, not just against last month.
Use the simulators
Google’s bid simulators forecast how changes to CPA, ROAS or budget targets might affect conversions and spend. They are estimates, not promises, but they are a better basis for a decision than a hunch.
Issue 4: Judging results too early, or changing too often
Smart Bidding performance varies from day to day. Google says this variance is normal because of auction dynamics, and that sustained spikes or drops may point to other causes such as conversion delays or reporting outages. For a fair view, it suggests longer evaluation periods, for example at least 50 conversions or a full month of data, rather than reacting to daily swings.
Learning periods are real
Campaigns often need a learning period to understand business goals. Changing settings such as conversion goals or targets during that phase can force the algorithm to restart its learning. A common recommendation is to allow one to two full conversion cycles after a major adjustment before changing targets or bidding settings again.
A conversion cycle is how long it takes your customers to go from click to purchase. A store selling phone cases might have a cycle of a day. A store selling furniture or high-consideration items might have one of several weeks. Match your patience to your product.
What tinkering looks like
- Changing the target ROAS twice in one week because Tuesday looked bad.
- Swapping between Maximise Conversions and Target CPA every fortnight.
- Editing conversion goals mid-test, so you cannot tell what caused any change.
- Pausing and relaunching campaigns, which throws away accumulated learning.
If you must change something, change one thing, write down the date and the reason, and wait. The same logic applies on social: our guide to Facebook ads stuck in learning limited explains how to preserve learning there.
Use the Bid Strategy Report
Before switching strategy, open the Bid Strategy Report. It shows average target performance and flags issues such as conversion delays. If conversions are reported late, a campaign can look like it is failing when the sales simply have not been counted yet.
Issue 5: Campaign types that amplify the problem
Performance Max and Shopping campaigns depend heavily on Smart Bidding, and they hand over more control than Search. That makes weak signals more costly: there are fewer levers to pull, and the algorithm decides where your budget goes across channels. If your Performance Max results are erratic, run the conversion audit above before anything else, then look at asset quality and search themes. We cover this in Google Ads Performance Max: worth it? and Performance Max search themes.
Broad match works the same way. It is far more productive when the account has strong conversion signals, and it needs a ramp rather than a switch flip. See our notes on the 6 to 8 week broad match ramp.
A diagnostic order that saves time
When a client arrives saying Smart Bidding has stopped working, we go in this order, because each step makes the next more reliable:
| Step | What to check | Typical finding |
|---|---|---|
| 1 | Primary conversion actions and duplicates | Non-purchase events counted as primary, or double counting |
| 2 | Tag health and conversion delay | Broken tag after a site change, late-reported sales |
| 3 | Conversion value | Revenue sent where margin matters, no refund adjustment |
| 4 | Budget against target CPA or ROAS | Budget too small for the volume the target needs |
| 5 | Target level and history of changes | Target far above actual, or edited every few days |
| 6 | Landing page and offer | Traffic is fine but the page does not convert |
Step 6 is the one people skip. A bidding algorithm can buy qualified clicks, but it cannot make a weak product page convert. If traffic is healthy and conversion rate is low, look at the page itself, using our guide to Google Ads landing pages that convert.
Our perspective
We see the same pattern again and again. A brand switches to Smart Bidding, results wobble, and the diagnosis becomes “the algorithm is broken” or “automation does not work for us”. Then the audit shows a primary conversion list that nobody has reviewed in a year, a target changed every few days, and a profit picture nobody connected to the ad account.
Smart Bidding does what it is told, very efficiently. The strategy work is deciding what it should be told. That means data-driven decisions about which actions count, what a sale is worth to the business, and what the budget needs to look like for the growth target. It is also why we keep senior strategists on the account rather than outsourcing it: the judgement calls above are not tasks you can hand to a template.
Our advice to owners is simple. Fix the signals, give the system a budget it can learn on, change one thing at a time, and measure against profit, not only against the platform’s own ROAS.
— Liza
Want a second pair of eyes on your account?
If your Google Ads account is spending but not scaling, we can look at the conversion setup, targets and structure with you. Our 12 Week DOUBLE Your Revenue Challenge is a hands-on mentoring program built around exactly this kind of problem, and you can see how we run paid search for stores on our Google Performance Max™ and advertising page.
FAQ
Why did my results get worse after switching to Smart Bidding?
Usually because the algorithm is now optimising hard toward whatever your conversion data says is valuable. If that data includes low-value actions, duplicates or missing profit information, performance moves in the wrong direction. Also check that you did not change the conversion goals or targets during the learning period, which can restart learning.
How long should I wait before changing a Smart Bidding strategy?
Google’s guidance points to longer evaluation windows, such as at least 50 conversions or a full month of data, and one to two full conversion cycles after a major change. Daily volatility is normal. Change targets only when the longer trend, not a single bad day, says it is needed.
Should I go back to manual bidding?
Manual bidding can feel more controllable, but it cannot react to auction-time signals the way automated bidding does. In most cases the better move is to repair the signals and settings rather than abandon automation. If the account has very little conversion volume, a simpler strategy while you build data may be sensible.
Does a low budget cause Smart Bidding problems?
Yes. A budget that only covers a handful of conversions gives the algorithm too little to learn from. Google states that for Demand Gen the budget should be at least 10 times the target CPA, and the same logic of leaving room to learn applies to other campaign types.
Sources
- Google Ads automated bidding, Google Ads Help
- About Smart Bidding, Google Ads Help
- Tips on measuring Smart Bidding performance, Google Ads Help
- Demand Gen campaign performance guide, Google Ads Help
- About conversions with cart data, Google Ads Help
- Display Smart Bidding Guide, Google
- How to troubleshoot automated bidding in Google Ads, Search Engine Land
- The Tracking Problems That Break Paid Search, ClickTrends
- Conversion value glossary, Product Metrics
- How pixel tracking identifies leads for marketers
