Intelligence With Impact: Marketing and Machines in an Artificially Intelligent World
The ancient Greek Heraclitus is perhaps most famous for saying, “The only constant is change.”
While Heraclitus was not a digital marketer, he sure spoke like one.
When paid search began almost 18 years ago, marketers had to do everything manually to achieve a set ROI. They had to handwrite hundreds of ad copy variation that drove relevant consumers to the site, and they had to manage thousands of keyword-level bids to minimize the cost of every click and conversion.
In recent years, Google has invested heavily in machine learning, and automated capabilities began assisting (or in some cases, replacing) some of paid search’s most labor-intensive activities that influence ROI. For example, Google’s Target Cost-Per-Action (CPA) capability uses advanced machine learning to automatically tailor a paid search bid to every unique search. This process maximizes conversions against a marketer-designated CPA. With over 40 queries per second within Google alone, it would be impossible for a marketer to manually achieve a similar bid frequency.
Fast forward to last July’s Google Marketing Live 2018 keynote, where the company introduced a number of new advertising capabilities and opportunities. Machine learning was baked into every major announcement.
Google’s new Smart Campaigns arguably leverages machine learning the most. When implementing a Smart Campaign, the marketer provides one ad, a description of the product or service they want to promote, a budget and their key performance indicator (such as ROI) related to their business objective. Google then takes that information and automatically displays ads to searchers the system believes will serve the marketer’s objective (e.g., a purchase, store visit or phone call). With today’s technologies, the marketer is live within paid search in minutes, without having to implement an account structure, keywords or settings.
Even with these benefits, machine learning is reshaping, not replacing, the role of the paid search marketer. Paid search marketers wear lots of hats. They must be great media managers to allocate assets appropriately, creatives to write text ads that resonate with searchers, and data scientists to interpret what performance numbers mean. The role of a paid search marketer is constantly changing, and it needs to further evolve to improve performance.
In response to the evolving search landscape, we look at three specific attributes of a paid search marketer to maximize performance in a world influenced by machine learning:
Paid search marketers should no longer think of themselves as “doers.” Instead, as machine learning moves to replace common tasks, paid search marketers should think of themselves as strategists first and foremost.
With our paid search team, we’ve focused our saved bandwidth from “doer” tasks to develop proprietary partnerships and actions that allow us to be more strategic while improving our client’s performance. We fill gaps that publisher-provided machine learning tools do not take into account. For example, the engine platforms do not natively adjust bids based on external factors like weather or when television commercials are airing. If an event occurs in the real world that the publisher platforms cannot recognize, we have the ability to automatically adjust our positions within an engine’s search result page to take advantage of consumer behaviors temporarily influenced by an external factor.
We applied this for one of our clients, a Florida government agency charged with the marketing, research and regulation of the Florida citrus industry, with exciting results. We aggressively served Florida orange juice ads to geographic areas that were about to have a flu outbreak. This ability to adapt to real-world changes had tangible results: Our program generated $300 million in incremental revenue and 1,400 new jobs for the state of Florida.
Automation has also afforded us the opportunity to test new publisher tools. For example, we took advantage of Google’s new Responsive Search Ads as those capabilities became available. Responsive Search Ads add more characters to ads and take up more real estate on the page. They also dynamically test 15 different headlines and four different description lines. Testing has shown this ad unit can improve click-through rates up to 15%, which helps ROI by driving more relevant traffic to the site with improved Quality Scores and lower costs.
Automation is extremely valuable, but it isn’t a silver bullet. A paid search marketer needs to be skeptical of what machine learning can accomplish. Machine learning has not reached a point where it can understand why something happened, and this can create undesired results. While it may identify incredible performance on a certain day for a marketer promoting a tax service, machine learning will not recognize that day as April 15th, the last day for procrastinators to file taxes. As a result, the algorithms it generates may incorrectly maintain elevated spending levels when conversion behavior is less favorable on April 16th.
Understanding these limitations, our team does not place blind faith in the systems. We monitor intraday performance and settings daily to ensure the algorithms are not taking inappropriate actions that could negatively impact ROI. We test and learn new capabilities before rolling them out extensively. If we identify something that isn’t doing what it’s supposed to be doing, we put in place manual overrides to course-correct an action, to teach the system the appropriate action.
A great example of this is Google’s loosening of what constitutes certain Match Types. Match Types tell the engines how far a marketer is willing to distribute their ads across search queries. A marketer who uses Broad Match on a keyword is telling the system to allow their ads to display against the keyword phrase, similar phrases, singular or plural forms, misspellings, synonyms, stemmings (such as “run” and “running”) and other relevant variations. However, a marketer may determine Broad Match’s reach is too broad, as ads could match queries that do not serve their ROI, and these marketers can use more restrictive match types, such as Exact Match.
When Google introduced Exact Match in the early 2000s, it meant the keyword would only place an ad exactly against the searcher’s query. A keyword (e.g., men’s pants) would only serve an ad against that identical query. However, Google’s definition of “exact” has morphed over time. In 2014, Google allowed close plural and misspelling variants to trigger ads from Exact Match terms, and in 2017, Google allowed different word order and function words to trigger ads. This year, Google announced another change to close variants. Exact Match keywords can now match to queries that share the same meaning as the keyword, including implied words and paraphrases as determined through Google’s machine learning capabilities.
As these changes evolve over time across keywords, we have maintained an appropriate skepticism that the system is not going to be right 100% of the time. Search Query Reports have been analyzed to determine how well Google is matching Exact Match ads to queries, and we have improved Exact Match ROIs by implementing new negative terms that improve query targeting.
A paid marketer needs to understand programming. Machine learning will become further ingrained into paid search, and having a firm understanding of how these algorithms work will be crucial for managing them in a way that maximizes revenue per dollar. Programming can be used to make a paid search marketer stand out, bring unique value, and improve ROI in a world where most ads will be managed by the same algorithms. We have developed proprietary scripts that can improve our targeting in real time, which today’s machine learning capabilities don’t offer natively. We can automatically adjust bids based on how a granular geographic area performs. We can also apply negative keywords to individual ad groups or entire accounts to restrict our appearance on poor-ROI queries.
Using methods like these allowed one of our retail clients to see a 28.2% week over week improvement in ROI while concurrently using machine learning to assist with bids.
While machine learning will no doubt continue to expand its capabilities and grow more sophisticated, these capabilities are only one component of our marketing process. It does not replace human thought and instinct. We believe that the consumer should be at the heart of everything that we do within paid search. When a query is made, the searcher is looking for a want or need to be fulfilled. It’s our job to make sure we offer the right solution, and while machines will help, there will be limitations that prevent maximizing performance.
Or, to put it another way: “Much learning does not teach understanding.” – Heraclitus
Want to know more?