7 tips to unlock the full power of machine learning for your campaigns

Smart bidding on programmatic platforms such as Google and the Meta apps (Facebook, Instagram and WhatsApp) has come a long way over the past decade. These platforms not only gather a wealth of data about their audiences, but their artificial intelligence and machine learning algorithms have improved in leaps and bounds. As well as you may know your customers, there’s a good chance that Google and Meta know them better.

Consider Googleand the digital intent signals it gathers from customers as they move through an ecosystem of apps that includes Gmail, YouTube, app store, search and display advertising. Few companies—not even banks or insurers—can get quite as comprehensive a view of a customer’s habits, interests, behaviours and intentions. This intelligence is at our fingertips, but to make the best use of it, marketers need to lean into the platforms’ automation capabilities.

Here are some of the best practices that will help any marketing team to unlock better results as they take a more automated approach to bidding and targeting on the major programmatic platforms:

1. Help the machine to learn

Smart bidding strategies use intent signals from audience segments to improve performance for you.As clever as the platform algorithms are, they perform better when you support their learning process. An important first step is to correctly configure your campaign or user journey with accurate conversion tracking.  Be sure to tag every event relevant to the user journey, so you can alternate between conversion goals and help the machine to learn. This will also help youoptimise the user journey and user experience based on insights into how users interact with your digital touchpoints.

2. Set goals higher up the conversion funnel

With most bid strategies,a campaign will need to generate a certain amount of conversions to exit its learning period. During this period, we recommend setting goals that are higher up in the conversion funnel. For example, we would set our conversion goal as an add-to-cart button click instead of the customer reaching “thank you page” after a successful conversion. This will increase conversion volumes on the campaign and improve performance. After the learning period, you can switch goals to feed the algorithm with the data most relevant to the overall goal for your campaign.

 3. witch goals up if it’s not working for you

On Google’s platforms, the smart bidding strategy needs to align with the campaign goal. For search, conversion-focused bidding strategies work well. These include maximise conversions, target cost per acquisition (tCPA), target return on ad spend (tROAS), and enhanced cost per click (eCPC). However, if a campaign is struggling to gain traction, shifting to a traffic-focused bid strategy—like target impression share and maximise click—will increase volumes.

 4. Give the algorithms time to learn

The machine learning algorithms won’t deliver instant results. They need time to gather sufficient data through user intent signals. The recommended time for learning is two weeks. During this time, it’s best not to change the campaign or the tags or goals associated with it. This will restart the learning period, causing a further delay in achieving the performance you want from your campaign.

 5. Go broad rather than granular

As I mentioned earlier in this article, the platforms know your customers really well. We no longer recommend granular targeting because Facebook and Google will do the granular targeting for you and with more success than any human can. On Google Search, keyword match types are now irrelevant to conversion focused smart bidding. The keywords are no longer the main drivers of performance but instead help the algorithm, which is why we recommend broad match terms.

On Facebook, we have noticed a move away from audience segments. We still use remarketing tags,but we generally see greater conversion performance from targeting strategies where no customer lists or segmentation are implemented. The algorithm plays a major role here.Domain verifications and conversion API requirements are essential in ensuring you utilise machine learning to its full extent.

 6. Use your first-party data to get an edge

With digital platforms moving away from third-party cookies and identifiers, it’s becoming hard to track customers’ behaviour across the internet. Feeding Google and Facebook with your first-party data will help their algorithms to improve performance. Relevant, accurate data from your customer relationship management systems and other platforms will help the algorithms to vastly improve targeting, leading to greater conversion volumes.

 7. There are still some useful manual interventions

Campaign optimisation is becoming increasingly automated, but there are still some useful manual steps to keep in mind. Creating negative keyword and audience exclusion lists can help steer the platforms from showing ads in the wrong places. For example, if you’re selling casual summer clothes, you can red-list terms like eveningwear or formal. With audience exclusions, an obvious use case is to exclude known existing customers from seeing a brand ad, and existing customers who are unwanted (for example, customers who have subscribed but was unable to pay their monthly subscription). You can also exclude specific data using an advanced Google Ads API. For example, if there were errors with an account’s conversion tracking during a date range, you can use data exclusions to inform smart bidding to ignore all data from those dates.

In conclusion – Focus on strategy rather than operations

With automation tools like smart bidding becoming more sophisticated, we are able to leave more and more of the day-to-day decisions to the algorithms. In our experience, these algorithms not only spare humans a lot of manual work, they also vastly improve outcomes. What’s more, they free up our time to focus on the bigger strategic business goals rather than on a small campaign optimisations.