Back to Blog
Advanced Techniques

How Google Smart Bidding Works: tCPA vs tROAS, and How to Optimise So the AI Performs at Its Best

However smart Google's algorithm is, it's useless if the initial settings are wrong

June 22, 2026 8 min read

Google Ads today is driven almost entirely by machine learning, from choosing search terms and placing ads to adjusting bids in real time through a system called Smart Bidding. But many businesses still mistakenly believe that switching on Smart Bidding will automatically improve everything, when in fact the algorithm needs the right data and settings to work at full capacity.

How does tCPA (Target Cost Per Action) work?

tCPA is a bidding strategy that lets the system automatically adjust bids to get as many conversions as possible while keeping the average cost per conversion close to your target. It bids higher when a conversion is more likely and lower when it is less likely. It suits businesses focused on lead volume or orders at a controlled cost.

How is tROAS (Target Return On Ad Spend) different from tCPA?

tROAS suits e-commerce businesses where order values vary. The system tries to grow total revenue to hit your target ratio. For example, a tROAS of 400% means you want at least ฿4 in revenue for every ฿1 spent on ads. The system bids more aggressively on audiences likely to make high-value purchases, rather than just chasing the number of conversions.

Google Smart Bidding AI Algorithm

Why does Smart Bidding perform poorly in some accounts?

The main reason Smart Bidding underperforms is insufficient conversion data. Google recommends at least 30 conversions per campaign per month for the system to learn accurately. If your tCPA or tROAS target doesn't match historical data, the system will bid too low and your ads may not show at all. Changing targets too often also forces the system to re-enter the learning period again and again, making it unstable.

How to optimise Smart Bidding for full performance

Start with targets close to your account's historical averages, then adjust gradually by 10–15% every 1–2 weeks, avoiding sudden changes. Send quality signals through accurate conversion values, such as real order values instead of counting every conversion the same, and give the system at least 2 weeks to learn before judging whether the strategy works.

  • Don't change your tCPA or tROAS targets too often, as it resets the learning period
  • You need at least 30 conversions per month per campaign for Smart Bidding to work accurately
  • Send accurate conversion values, not just a count where every conversion is treated the same
  • Give the system at least 1–2 weeks to learn before evaluating results after every adjustment

Summary

Smart Bidding isn't a magic button you press once and forget. It's a tool that needs quality data and enough time to learn. The Markettung team manages bidding settings and strategy for every client account so Google's AI can perform at its full potential.

Want to start advertising without taking the risk yourself?

Starter package from just 6,900 THB 4,900/month, with a professional team to plan your budget

View packages