Asset-led Marketing: The Rise of Machine Learning in Digital Advertising.
Introduction
Digital media buying as we know it today has passed through several stages in its bid to show the perfect marketing messages to a target audience at the right time. Many of these are powered by an advancement in technology and computational processing power, but more recently — a series of algorithms — enabled by highly functional machine learning models. In this rather short essay, I want to share my thoughts on the not-so-recent trend in the digital marketing domain as it relates to ad delivery and campaign performance.
Setting a Context
Since the first clickable ad banner appeared on the internet in 1990, we’ve witnessed multiple highly sophisticated ways of serving media assets on the web — far away from just buying direct inventory from publishers, which is then served to all website visitors.
What has now become basic is targeting users by demographic — age, gender, job title, location etc — across a host of publisher websites and mobile apps that are part of an ad network.
Interest and behavioural targeting take it a step further by enabling advertisers to reach their target audience based on a visible pattern in their web activities (search, video and content views) and the actions they take on social media or advertisers-owned platforms. Remarketing (including its dynamic form) fits here.
Custom audience targeting now empowers marketers to show their marketing messages to specific highly valuable users by uploading their personal information such as email and phone numbers on an ad serving platform. Layered on top of that is the lookalike audience that targets users who portray similarities — look like — with the custom audience.
Diving Deeper
While these targeting options focus on reaching the right user, the strength of what I have termed asset-led marketing lies in delivering the right content at scale by automatically combining campaign assets in a way that is most relevant to each unique user and available ad slot, based on signals from machine learning infrastructures.
Asset led marketing is best described as a situation where an advertiser provides multiple variations of campaign assets; including texts, key visuals, videos, CTA and other brand elements and the algorithms take on the responsibility of serving different combinations of these elements to the intended audience.
This creative mechanics saves costs while allowing advertisers to quickly and automatically test multiple combinations of campaign elements that will resonate with each user and in other cases reach more audiences by utilizing every available inventory.
Asset-led marketing is presented as a responsive display ad on the largest display network. The system allows advertisers to run with different media types including text and youtube videos all at once, unlike the classic display campaign. These assets are automatically combined to create variations that are then served across the network. Google recommends this for wider reach and better overall campaign performance.
The responsive search ads on Google follow the same logic; provide the machine with as many text assets — up to fifteen headlines and four descriptions — and it will learn and deliver the best combination whenever your keywords trigger your ad. Google announced that the traditional alternative, the expanded text ad, is going away in 2022.
Facebook’s equivalent of asset-led marketing is the Dynamic creative which accommodates various media types for ad copies and can go as far as optimising your ad destination between your website and your shop on Facebook or Instagram in order to keep the ad relevant to individual users.
A particularly interesting manifestation of this is the Google Universal App install campaign that runs across the display and search networks, search partners and YouTube. This campaign type doesn’t allow for any form of audience targeting (aside from country-wide location targeting) or placement selection. The campaign assets, budget and bid are the only input required by advertisers. Facebook also has its own optional version of this called Automated App Ads.
While some of these features are now widely used, advertisers are testing out the recent ones. Wordstream reported they saw 60% more conversions and 55% lower cost-per-conversion after testing the dynamic creative ad set against a single, static version on Facebook.
Conclusion
Perhaps the most exciting part of this for me is how much the industry has come to embrace machine learning and artificial intelligence for many of our media buying activities; from targeting to messaging, personalization and ad delivery. Maybe this isn’t surprising for an industry enabled by technology, before now we’ve witnessed the growth of programmatic media buying.
While this is broadly a welcome advancement, many of these solutions have been described as requiring “less effort” from advertisers. The concern, however, is that a few years down the line, marketers might not be able to optimize, learn and claim responsibility for campaign success or otherwise — especially when advertisers’ liberty to get their hands dirty is being taken away and it’s becoming more tedious to diagnose, attribute and report campaign activities.