Google Ads Bidding

Can’t trust Smart Bidding on big decisions? Outperform it with an out of the box approach.

Google is pushing hard its Smart Bidding to the Google Ads accounts. For a lot of people this is indeed a better solution than setting manual Cost-Per-Clicks (CPCs) on their own. It is also free.

If you manage big budgets within your Google Ads account, you are probably looking for more sophisticated approaches that are tailored to your business. Let’s have a look at the weaknesses of Smart Bidding and the approaches you can use to outperform it.


The following aspects are not only limitations of Google Smart Bidding – you can see them as challenges in general when it comes to a bidding solution.

Smart Bidding and other 3rd party bidding tools have to work for a lot of companies out there.

You will outperform those systems when your are able to bring your business knowledge to the bidding models.

Sparse data

If the data is not sufficient you will not be able to make good decisions. This is very often the case for long tail keywords. When numbers are not stable from a statistical standpoint you shouldn’t use it for your bid calculation.

High variance in basket sizes

At some point you will use the basket size to calculate a value per click. But which numbers should you use. A single conversion can bring up your total basket size average. Is it a reason to raise the bids for that keyword for the next months? I do not think so.

No external data sources

Google uses the data that is found in your Google Ads account. They do not know about overall transaction numbers of your e-commerce site, but it makes total sense to use them for bid estimations for low sample size product keywords. What are your competitor prices look like today. This is just an example where more data means better results for your bidding.

Short term seasonalities

If you run a lot of sales actions for a limited period of time it will take a while for Google to realize that there is change in conversion rate. The sale may have already finished when bids were adjusted. Knowing that there is a sale, and even knowing on which brands and products is a huge advantage.

Optimizing on margins

In the end you will look at the profit you’re making as a company. But how can we handle that in Google Ads when it comes to bidding? Just sending the basket size as conversion value does not tell the full story. When using the absolute margin you will also struggle when margins per products change over time.

Information of order cancellations

When you have a lot of returns in your business you should also consider this for your bidding approach. Of course you can work with global percentages to make high level adjustments. In reality cancellations will not be distributed in the same way.


When analyzing the status quo of bidding solution we look at a lot of different aspects to quantify the weaknesses of the current solution.

Let’s have look at 2 simple bidding audits you can run easily on your account.

What is the highest CPC in your account?

If you follow Google’s recommendation, you may have not set upper bounds for your bids. The trouble often begins here. When Google machine learning algorithm finds a lot of good signals you will probably observe single click prices that are higher than your CPO target. You do not understand machine learning to realize that there is something wrong. I think you will be surprised upon seeing what Google payed for a single click.

How does the 4th CPC quartile perform?

When your bidding has trouble with low sample sizes or outliers in your basket values this causes that you make wrong decisions. Historical data can not be used for predicting the future performance. If a keyword triggered randomly a big order it does not mean you should bid more in the future. For that reason you should have a close look at the top end of your most expensive clicks. What is the aggregated performance? The sneaky thing: You will not realize this when you just look at average numbers.


Yes it is effort doing this. If you are spending huge budgets on Google Ads this can be one of your best investments. I’m going to share some principles and techniques that will help you to get ideas for the components of your bidding system.

Spend your time on feature engineering

Machine learning is not a competitive advantage anymore. Kaggle challenges are won by the guys how found the smartest ways for feature creation. This will make your model superior to others. Some generic approaches to generate new features is to create n-grams on top of your keywords, and even better use an entity database for your business. The rest depends on your business – use your knowledge to make smart clusterings, reduce dimensionality for attributes or use thirds party data to create totally new data points.

Choose the right machine learning model

Of course you can use stacked models that give you some more improvements on model quality. The downside is that they are difficult to interpret and it is very difficult to explain it to other people. For that reason I really like Random Forest Regression – it is possible to explain the most important variables.