Machine Learning in Marketing: Trends Shaping 2018
Google's Smart Bidding rollout, dynamic creative optimization and useful marketing chatbots: what machine learning changed this year, versus what's still hype.
“Machine learning” has been a marketing buzzword for a couple of years now, but 2018 is the year it started showing up in tools marketers use daily, rather than in vendor slide decks about the future. Three shifts stand out.
Bidding got automated, whether agencies liked it or not
Google pushed Smart Bidding hard through 2017 and into this year, strategies like Target CPA and Target ROAS that use machine learning to set bids per auction based on far more signals than a human managing bids manually could ever weigh in real time: device, location, time of day, past behavior and dozens of other factors scored simultaneously.
The reaction from performance marketers has been mixed. Some have handed over significant budget and seen real efficiency gains, especially on accounts with enough conversion volume for the algorithm to learn from. Others, particularly on smaller accounts with thin conversion data, have found automated bidding performs worse than a skilled human, because the algorithm doesn’t have enough signal to work with yet. The emerging rule of thumb: automated bidding needs volume to earn its keep, and accounts below a certain conversion threshold should stay manual a while longer.
Creative started adjusting itself
Dynamic creative optimization tools, which auto-assemble ad variations from a set of images, headlines and calls to action, then let machine learning find the best-performing combinations, have moved from a Facebook-only novelty to something available across most major ad platforms this year. The appeal is obvious: instead of a marketer guessing which of twelve possible ad combinations will work best, the system tests all of them and shifts spend toward winners automatically.
The catch we keep seeing in practice: DCO is only as good as the raw assets fed into it. A tool that can recombine five mediocre headlines and three mediocre images will still produce, at best, the least-bad combination of mediocre parts. It doesn’t create good creative. It finds the best arrangement of what you give it.
Chatbots got, for the first time, decent
Marketing chatbots have existed for years and mostly annoyed people, rigid decision trees dressed up as conversation, breaking the moment a customer phrased a question slightly differently than expected. The natural language processing behind this year’s chatbot tools is meaningfully better at handling variation in phrasing, and a few brands running them for straightforward support questions, order status, simple product questions, are reporting real deflection of routine tickets rather than just frustrated customers hunting for a human.
The honest caveat: this only works for narrow, well-defined use cases. Anything requiring judgment or handling an unhappy customer still needs a person, and businesses deploying chatbots without a fast, visible path to a human are getting complaints, not efficiency.
What we’d tell a marketer prioritizing where to start
Automated bidding is worth testing now if you have conversion volume to support it. Dynamic creative optimization is worth it only after you’ve invested in good creative assets, not as a substitute for that investment. Chatbots are worth deploying for a narrow set of routine questions, with an easy escalation to a human built in from day one, not as a general customer service replacement.
More on AI Marketing, or browse everything from 2018.
