Cater to the "market of the one" — this has always been the holy grail of marketing.
Brands and marketers have always strived to understand individual consumer necessities and tried to cater to them directly through an open dialog, at scale.
While this was long a pipe-dream, with the advent of deep neural networks, the current crop of machine learning algorithms, and advancements in artificial intelligence (AI) research, the age-old spray and pray marketing is coming to an end.
Now, with machine learning, brands have a good shot of being truly coherent in their narrative and engaging consumers with a consistent voice, tailored to individuals across omnichannel end-points.
To break it down, let's take a concrete example of advertising a kid’s video game, such as "Plants vs. Zombies — Garden Warfare 2" and compare the two marketing options.
Traditional Marketing vs. Machine Learning in Marketing
In the marketing world, the best course of action for such a game would involve defining the genre of the game, the intended audience behavior and the market segment to advertise.
It’s also worth deciding on a channel strategy along with efficient channel assets, defining the expected performance per channel based on some pre-determined assumptions, and eventually allocating the budget per channel.
The 3-Step Challenge
Step 1: Genre
What is the genre of the game? Is it the mid-core game? A hard-core game? Is it a broad category like a third-person shooter?
The gameplay has characteristics of tower defense as part of it. It also has customizations, multi-player portals, missions and quests.
So how should one template this out?
Traditional Marketing: Marketers love templates. When defining the genre of a game, they will often settle for a broad category such as third-person-shooter.
The reason for this is to avoid reporting and downstream channel complexity. In best cases, additional tags can get added to the template which also must come from a predefined taxonomy — the fewer details the better.
Machine Learning in Marketing: In the world of machine learning, the game does not necessarily need to have a genre. A detailed, written description of the game and gameplay is all that is needed to feed into the machines. In contrast with traditional marketing, the more details, the better.
Better yet, machine learning algorithms thrive on additional text from comments, articles and user descriptions of the game, and can make up its own dynamic internal knowledge representation of the game to truly capture the ‘spirit’ of the game as against a genre.
Here, machine learning techniques such as entity extraction, relationship extraction, sentiment analysis and other retrieval techniques are used to encapsulate the spirit of the game.
Step 2: Audience
Who is the intended audience for the game? What is the behavior of the audience?
Since this is on all game consoles (Xbox One, Xbox 360, PS4, etc.) should we market to all console manufacturers? Or is it more prudent to allocate budgets only for third-person shooter genres on Xbox One only? What are the general characteristics of this audience?
Traditional Marketing: Traditional marketing is heavily based around demography, such as age, race, gender and psychographics, such as personality, values, interests and lifestyles.
Kids aged 7 to 14, and predominantly male, is the first cut. Next comes additional cuts of cohorts of audience owning a specific console, such as who has purchased tower-defense or third-party-shooters in the past, who are in a specific geo-location. It is important to note that the audience cohorts should be pre-determined and bounded in advance.
Machine Learning in Marketing: With machine learning, you eliminate the cognitive-bias that only certain demographics would like this game. Instead, you do three things:
As explained in step 1, build an abstract knowledge on the true "spirit" of the game based on machine learning techniques like Word2Vec and techniques from Natural Language Processing (NLP) which abstracts knowledge from text based game descriptions, social feeds etc. This is unique to the game, yet also contain hundreds of nuanced dimensions of the game.
Once you have the spirit of the game play as an abstract knowledge, you build machine learning algorithms to allow the machines to dynamically learn what types of users have shown an affinity towards such a game in the past.
You can define affinity as a like/click/download/purchase or any other consumer actions that deems fit. Here, machine learning can learn the "essence" of the user as an abstract knowledge based on multiple dimensions. Machine learning thrives on data and dimensions about the user beyond mere demography. Dimensions such as users marketing channel affinity, past attributes, view-throughs, the frequency of visit per channel, average opportunity to see, etc. can all be computed through other machine learning techniques.
Once you have the spirit of the game and the essence of the user, you can decide if the game advertisement is a good match for a user, in real-time, without having to create a database of users from the pool. This is far more effective and conducive to audience-exchanges where you don't or can't own or access the entire user population per channel.
Step 3: Channel Strategy and Assets
A channel strategy entails choosing the right communication channel for the end-user such as mobile, web, brochure, retail outlets, TV ads etc. The assets allocated to the channel include the message to the audience, the message format, frequency of communication, follow-ups etc.
Traditional Marketing: For mobile advertising channels, you have to decide early on what the audience-cohort messaging is, and what would be relevant for the mobile channel. The message for a third-person shooter will be different than a generic demographic targeting kids aged 7 to 14.
Additionally, the ad format has to be predetermined such as full-screen ads, video ads, native ads etc. You have to pre-create all the messages and pre-fit them to the format. You do not have a choice of making up a message on-the-fly that is relevant to the end-user, environment and the utility of the channel.
Machine Learning in Marketing: Unlike traditional marketing, with machine learning you can ask the machines to prepare the messaging dynamically at run-time based on how the essence of the user matches the game spirit.
If the user is a 35-year-old tech professional who happens to love this game, not only will you advertise to the user, but the machines can create an appropriate message that can dovetail the spirit of the game to specific sensibilities of the user. For example, the machines may choose to prepare a message stating, "Tired of boardroom presentations? How about a quick backyard battle with the zombies?" where backyard-battle is the element of the gameplay.
This is channel agnostic and can fit into chatbots, virtual assistants, mobile advertising or any other channels you employ.
Similarly, the budget per channel can be altered in real-time through channel lift metrics that can be fed into the machines, which can dynamically control the bids.
Machine learning can be used to improve channel efficacy by a minimum of 80 percent and unbounded maximums if done right. Even simple deep learning and machine learning techniques on KPIs like click-through-rates, conversion rates and bid-to-win ratios can bump the efficacy by 50 percent to 200 percent.
Machine learning is dramatically changing the marketing landscape. It’s prudent for marketers to understand and employ machine learning algorithms sooner than later.