Machine learning and the customer journey

  • February 9, 2017

Machine learning and the customer journey by Doug Conely

As we dive into 2017, the marketing industry conversation around big data has evolved into a discussion on the growing use of machine learning and the inevitable disruption this will create among planning and campaign delivery teams.

Over the past few years, those working in manual campaign delivery have already felt the impacts of programmatic technologies. Over the next twelve months and beyond, we can also expect automation to extend its reach into manual campaign planning, thanks to the widespread adoption of machine learning. In fact, as the benefits of machine learning become more widely understood, brands are likely to demand a change of their agencies.

The reason? Machines can quantitatively evaluate the consumer journey of millions of individual customers more quickly, efficiently, and effectively than any human planning team could ever hope to replicate, producing valuable insights into consumer behavior at a depth and level of detail that was never previously possible. These insights, in turn, can then be used by the machine to automatically deliver relevant messaging to consumers at a frequency appropriate to their stage of the consumer journey.

Machine learning will displace audience planning and manual campaign optimisation thanks to its ability to automatically adjust delivery based on the consumer journey, driving higher frequency messaging to high intent users and delivering lower frequency to users showing enough interest signals to justify targeting with brand awareness messages.

For many agencies, the main barrier to the adoption of machine learning will be internal skepticism about the ability to automate campaign planning and optimization activity in a sophisticated way. Others may initially reject the introduction of machine learning simply because they don’t understand the process involved.

What’s more, my organization’s own journey with machine learning has shown us that, as the algorithmic complexity increases and results improve, it can get harder to explain why the automated decisions are being taken, further mystifying the subject for many.

For this reason, here are few conceptual steps to help explain the way that machine learning works in practice.

Machine learning can expose the breadth and depth of consumer passions and needs

Ad platforms can access a range of anonymised audience data including demographics and location but we find the most useful, or predictive, are user interests. Manual planning processes most often look at subtle differences between high-level interests, like sport or cars, but online interests can reveal the true depth of passion or need. For example, an interest in “football” is more insightful than “sport,” while interest in “Manchester United” or even “Jose Mourinho” exposes real passion that is a stronger signal for brands. This increases the range of interests available from dozens to tens of thousands, making automation necessary.

We can build an anonymous picture of a person’s interests and needs

For an individual, these interests can be associated with their anonymous device profile. At a given point in time, the profile describes their observed breadth and depth of interests and needs. A newer profile, a result of newer cookies or devices, will be relatively sparse, but older profiles will contain a range of interests where the recency and depth of interest indicate current passions and needs. New needs show up in real time while old interests fade away.

The consumer journey for each brand plays out through their changing interests

For a given brand, we have seen how these consumer interests and needs can evolve over time, as consumers move from awareness to consideration to intent, from “low lift” (or “low probability to convert relative to the average”) to “high lift” for the brand as the interest signals increase in depth.

For instance, take a last-minute booking for a hotel break. The first time a hotel brand might be alerted to this is from a search or visit to their site. However, underlying the booking is a consumer need that is identifiable (such as the desire to attend a nearby concert or exhibition) and the machine can identify such triggers and use them to include individual anonymous consumers into a “consideration” set.

Machine learning can understand the relevance of the brand to each person

At this stage, machine learning can now consider the entire observed population and map it against buckets of intent from “high” to “low” for an individual brand. This shows how likely each user is to buy from your brand in the next 30 days relative to the average.

Crucially, though, people move between the buckets in real time. So, in aggregate, people will move up into higher buckets as display new interests but also move down as interests wane. This is the aggregate picture that the machine sees for the purposes of delivery. At this point, the machine doesn’t care about causation (why the user is doing something), just correlation (the fact that statistically they are more interested and therefore the advertising is more relevant to them).

In this way, machine learning allows brands to create a responsive, personalized ad delivery strategy for literally thousands of individual customers simultaneously and in real time.

Machine learning can automatically deliver against the brand audience

By delivering at a higher frequency to your most likely prospects, you ensure that your brand’s message is prominent as the customer reaches the critical decision point and not the message of your competitors. Where these audiences are scarce, the machine will look for lower lift audiences to stay on pace and on budget. Thus, the campaign is exposed to a broader reach with lower lift to consumers with some potential to make a purchase in the future. This, of course, is the object of most branding campaigns in the first place, so that fact that machine learning can provide this automatically means that the writing is on the wall for manual planning and optimization teams.

Working through the stages outlined above, therefore, is the automated equivalent of the traditional ad campaign planning process, but with results that are faster and more cost-efficient, as well as insights that are far deeper, more complex and potential far more commercially valuable than anything human planners could ever produce.

If you are a client of a media planning and buying agencies, it’s time to challenge your agencies and vendors to embrace the benefits of machine learning, deliver on the promise of programmatic and automation, and free up their human resource to concentrate on customer experience and strategic data management.