Firstly though, what is it? At its heart, machine learning is a subset of AI: one or more algorithms drilling at lightning speed through data from past events, to look for patterns and react in a useful way. For example, autonomous cars like Uber’s driverless taxi and Annheiser-Busch’s ‘driverless’ truck - both successfully trialled in 2016 - work this way using data collected from millions of past journeys to help their vehicle make good decisions on the road. Tesla’s Autopilot software has data from over 200 million miles.
All suffer from being overhyped and are likely to see more challenges from regulators, customers and other road-users, but they show the power of machine learning at its best, building up masses of data and using ever-smarter algorithms to ping back the right response in milliseconds, arguably more reliably than any human could.
AI fund managers analysing huge data on past performance of certain shares, to help predict future winners. And in the marketing world, we’ve seen the first AI Creative Director put to work.
Marketers could use this to help measure ROI and buy smarter
So what has all this got to do with marketing? Basically it has a place because good marketing has always been strongly data-driven, and as there’s ever more data available, AI and machine learning may in the future increasingly help keep marketers sane and help us make better decisions. It may also help you buy smarter and more clearly measure ROI.
Target and ROI
Ever since Target ‘’knew a teenager was pregnant before her dad did’ in 2012 marketers have seen the power of this. Target knew pregnant mums were a lucrative subgroup and might easily become more loyal long-term. People in early pregnancy do tend to alter their buying habits a little. They had years of sales data from known first-time parents and used an algorithm to start spotting the same pre-birth browsing and shopping habits in others: it successfully alerted marketers who rolled out carefully targeted vouchers and other offers.
Machine learning can clearly help with ROI as the data load grows, measuring ever-more combinations of factors in our advertising and sales: images used, copy, channel, geolocation, time of day - in theory, all advertising combinations and customer actions from initial interest through to final sale. Anything that can be put on a spreadsheet is useful input to an algorithm.
Machine learning also lies behind some sophisticated media buys, usually using a specialist platform to make millions of individual decisions a day, serve the right ad or link to the right screen at the right millisecond, and adjust how much you pay for it. The platforms may be able to add in ‘enriching’ new data you don’t have such as demographics: potentially helping with targeting but of course increasing the amount of data that needs to be crunched. We have spotted the odd conflict here though: some of the larger platforms are reputedly linked to media-selling bodies.
Not for everyone
Clearly this isn’t for everyone and commentators advise patience: off-the-shelf algorithms need loads of clean data and quite a lot of trialling to refine and teach the algorithm how to spot a lucrative sales lead - but for certain situations, perhaps larger campaigns over multiple geographies or involving many channel partners, I can see the value. As the costs of the technology fall, a smart marketer could see a strong long-term return from a coordinated approach that incorporates machine learning.
…and some things never change
Early feedback on using this shows how some things never change: to get the best from this, start with your strategy and your business objective and stay focused there: don’t be wowed just by data on impressions, click-through rates or by the technology itself. This kit is powerful enough to follow a lead right through to the final sale, so use that capability to tie marketing ever closer to the bottom line.