Media buying involves a lot of manual work on a day-to-day basis. And whenever repetitive low-value processes are present, machine learning tends to bring the most value. Making programmatic changes to Google Adwords no longer requires profound coding knowledge. New AI-driven programmatic advertising platforms are becoming more accessible cost- and implementation-wise.
As machine learning moves to the ranks of “commodity” technology, we can expect that up to 80% of PPC managers’ tasks will be effectively automated. But you should not assume that AI will replace the human professionals. On the contrary, more augmented PPC teams will emerge, with human managers being responsible for campaign ideation and creation, while AI will take over various optimisation facets on a large scale. Here are several use cases illustrating how machine learning can be applied to enhance your PPC campaigns.
1. Machine learning enables granular micro-targeting
PPC budgets go to waste when you try to pursue the wrong audiences with the wrong message. People come to your website with different intentions and at different stages of their purchase journeys. With those journeys becoming increasingly omnichannel, it’s hard to segment different users on a truly granular level eg. by their past actions on your website, on third party platforms (eg. social media) and by their responses to various sales pitches.
Algorithms, powered by machine learning, can help you detangle that hot mess of scattered actions into a set of comprehensive user profiles that include the kind of data mentioned above. Here’s just one example illustrating how machine learning can be used for advanced user profiling. The developed algorithm categorised website visitors based on their salient behavioural patterns and assigned respective roles to them. For example, people who mainly browsed entertainment websites, services and games received the ‘entertainer’ alias.
The accumulated data and various profile information was then used to run a PPC campaign. During that campaign, the agency managed to identify 42% more active users and increase the average CTR to 0.52%-0.54% (from 0.5%). By targeting prospects with a higher intent to purchase, that slight spike in CTR can actually translate into a 25%-51% profit increase, while the cost-per-visit remained the same.
2. Real-time ad spending optimisation
Most managers are handling a large portfolio of campaigns simultaneously. They do not always have the time to optimise and re-adjust the budgets based on real-time market conditions, especially if the bidding is set to take place during off-work hours.
But every professional’s goals is to pay less for more qualified clicks. Researchers from China recently tested a real-time ad bidding algorithm, powered by reinforcement learning, to see how well it could cope with optimising the allocated daily ad budget. The results were as follows:
- Manual bidding yielded 100% ROI with 99.52% of budget spent
- RL-powered bidding secured 340% ROI with 99.51% of budget spent
The best part? RL and ML algorithms are capable of self-optimisation. Over time, their state-of-art can improve even further which means that they can find even more creative ways of securing higher returns while being challenged with tighter constraints.
3. Reduced ad fraud
Fraudulent ad impressions cost businesses nearly $1.28 million (approx. £975,900) per day. A single bot can drive up to 300 million fake views for video ads per day. What’s even more problematic is that fraudsters are now hiding behind the facades of reputable publishers. Recently, the Financial Times discovered that at least six different ad exchange platforms pretended to host advertising inventory on FT.com when that wasn't the case.
Machine learning can help managers clean up their inventory and create custom risk thresholds that will refine the quality and quantity levels of media buys. IAS Insider reports that one CPG brand managed to gain the following results after adopting an ML-driven tool:
- Pre-bid optimisation resulted in a 7.9% decrease in fraud levels.
- The overall fraud levels were contained at 1.3%.
- Over 32.2 million fraudulent impressions were avoided.
What machine learning cannot do in PPC (yet)
The common tasks such as bidding, reporting, budget optimisation and campaign analysis processes can be handled with higher efficiency by algorithms. AI cannot fully replace a PPC manager, but it can make them more productive and allow them to focus on creative tasks like designing compelling ads.
Algorithms are not yet well-versed in content creation (though some progress is being made in this direction). Managers are still responsible for creating and pre-testing wording and visuals for different types of ads. While AI can pinpoint the customer’s biggest pain points and struggles, it cannot comprise those insights into an actionable strategy that matches your current business goals. Again, that’s what your PPC manager should do.
To summarise, automation and machine learning are now capable of taking over a lot of the heavy lifting in PPC and programmatic advertising. Managers, in turn, are becoming empowered to spend their time creating, strategising, and experimenting with new tactics. This combination of ML and human creativity will continue to drive innovation in the PPC space, and the role of a PPC manager will continue to evolve into a more creative role as ML capability grows.
Henry Carless is the PPC and Data Science Specialist at Vertical Leap.