How machine learning is busting marketing myths and changing the game
For decades, marketing was considered a "black art," with executives struggling to see how marketing activities translated to the bottom line. When digital marketing emerged, it meant marketers could more accurately track the impact of each digital campaign dollar, bringing science firmly into the mix. Digital was a massive advance for the profession. But it was still really hard to track the impact of traditional, non-digital marketing channels until weeks or months after their execution through the use of manual statistical means.
Now, for the first time, marketers are using machine learning to measure impact right across the marketing mix, including free-to-air television, radio and billboard advertising in addition to digital and below the line. Marketers can now quickly see which elements of campaigns are working—and how they are impacting each other—in a matter of days. At the touch of a button, we can plan, track and optimise marketing strategies and investments to respond to changing market conditions and competitor behaviour.
Next Generation Marketing Mix Modelling (MMM)
Marketing mix modelling has been around for a while, using traditional statistical methods. But it can take up to six months to collect and clean audience data from broadcasters, business and sales data internally, and external data (on competition) and then build and refine a system of statistical models. The information eventually generated is interesting, but often unhelpful—especially if it doesn’t emerge until weeks or months after the campaign has finished. How many campaigns would we have cancelled, changed or upgraded if we could see their sales or retention impact a few weeks in?
Today, advances in computational power and machine learning are dramatically accelerating this process by enabling computers to perform the manual tasks previously done by humans. In an innovative machine learning development between Telstra and Accenture, the new MMM uses hundreds of machine learning models and thousands of data variables to track the ROI of TV, press, outdoor and other non-digital advertising campaigns.
Compared with traditional statistical methods, where humans update the models, this new machine learning-accelerated approach has the potential to provide critical information within days of receiving the data—not months. For the first time, we can adapt traditional campaigns in time to get ahead of evolving customer, market and competitive changes.
As technologies advance and data capture speeds up, response times will collapse, getting closer and closer to real time.
Marketing myth busters
As it delivers faster and deeper insights, MMM is tearing down the last remaining marketing myths. Previously, although digital channels came back with instant reporting, it was hard to figure out whether TV or outdoor campaigns were working. Now, MMM can tell you how successful those traditional, non-digital channels are—and it’s not necessarily what you think.
MYTH 1 Digital beats analogue – In a digital world, many people discount traditional media based on "gut feel"— and the fact that digital campaigns are easier to measure. But a single campaign rarely succeeds in isolation. A TV advert may prompt a consumer to do the digital search that eventually leads to a sale. Now, fed with data on consumer purchase behaviour broken down by demographic profiles, plus granular offline and online marketing, MMM gives marketers a common comparable measurement of marketing performance across all segments, campaigns and channels. This means marketers can now see how different channels and activities really interact with one another. It’s a big step up from traditional models, which can tell you about the impact of lots of different variables (TV, radio, press, digital) on sales, but not about their combined impact. It is not about digital versus analogue. It is about digital and analogue, working together.
MYTH 2 Offer beats brand – Marketers often assume offers are more powerful than general brand awareness. But MMM reveals clear instances where positioning and marketing messaging show good returns, with campaigns at times over-investing in tactical offer marketing. What’s more, MMM helps establish the optimal balance of brand versus offer marketing so they work best together.
MYTH 3 Traditional channels in general are dying – Traditional channels still play a very important role for particular messages. For certain campaigns, MMM shows that media channels such as cinema and magazines, that offer a captive, engaged audience and dwell time, work well.
Human-centred design puts art into the science
Bleeding edge data science may be changing the game but, as marketers, we still need to interpret information and decide how and with what to go to market. Marketers must co-create and co-design the tools MMM enables so that they can interpret the analytical insight and apply it to their day to day activities.
For example, Telstra has now moved away from static reports and instead developed an app that enables the marketing team to quickly see which elements of the campaign are working and which aren’t.
The abovementioned app visualises the insights from the model, enabling marketers to quickly grasp the implications of the data and adjust tactics accordingly. Meanwhile, executive reporting elevates the insights to leaders for strategic decision making. As a result, Telstra has taken a big step forwards on its journey to evidence-based marketing planning. By deploying MMM at both strategic and tactical levels, the marketing team is gaining rapid, actionable insights that support powerful, targeted decisions.
In the future, as machine learning evolves, computational power in the cloud grows exponentially, and marketers learn how to tap into this power, marketing will be transformed by explosive possibilities. As marketers get close to real-time information from every channel, and as we learn from new emerging unstructured data on the message and creative itself, plus consumer emotion and sentiment, marketing itself will transform into new genres of mixed-media marketing of increasing intelligence and sophistication. Potentially, we could see:
Increased ability for experimentation across all channels. Real time measurement across all channels opens up the possibility for rapid experimentation of refined creative, messages and channel mix, with the framework in place to test, learn and adapt quickly.
Real-time bidding in traditional channels. The media networks will be equipped to receive and execute real-time bookings, changes or refinements to campaigns across all channels in response to audience numbers, profiles or reactions.
Genuine hyper personalisation. Next generation of the MMM merged with rich customer-level insights and modern dynamic advertisement and content-serving capabilities across all channels will open up the possibilities of delivering and measuring highly engaging audience of one experiences.
Location data retargets billboards. Marketers will use GPS location data from phones and vehicles to recalibrate electronic bill boards to audience preferences—and even to their situations. Regular traffic jam spots will host sympathetic messages to frustrated motorists.
And this is just the tip of the iceberg. The fast, deep insights available from machine-learning will allow marketers to understand the precise collective impact of multiple campaigns across the fragmented marketing landscape. Armed with this knowledge, we will be able to subtly adjust each activity to get the best return on our marketing investment.