Back in March this year (26th March 2019), I gave an invited talk at the Toronto Machine Learning Society Micro-summit on Retail and Advertising. The talk was titled “Autonomic Marketing: How far are we?”. What follows is loosely based on that talk. I had to redact some of the proprietary information in the case studies but other than that the content is mostly the same.
Good Evening! I am Vijay Dakshinamoorthy and I am a director of data science at Omnicom Media Group. Omnicom Media Group is a leading global media network where we create end to end solutions for our clients with data at the core. It is comprised of full service media networks including OMD, Hears & Science and PHD with Annalect providing the data and technology solutions.
Having a background in computer engineering and business, I was intrigued by the concept of autonomic marketing. I would like to explore the following four questions in this talk. What is it and where is it headed? Is true autonomic marketing possible? What are some of the applications in this area and what is OMG’s point of view?
Here is a brief overview for my talk. We will first look at what autonomic marketing is. Then briefly we will look at the evolution of the retail industry as a context in which to explore autonomic marketing. I will then talk about some of the applications of AI in marketing and advertising. We will then consider what is possible in the future along with OMG’s point of view and where we think we are in the journey. I will then wrap up with some case studies in chatbots and visual listening with some lessons learned.
What is Autonomic Marketing?
So what is autonomic marketing? The term autonomic marketing comes from autonomic computing in the computer science discipline. It was first used by a few UK based researchers in a conference paper. It was also used by Jim Sterne in his book AI for marketing. Ilya Katsov uses a similar term, algorithmic marketing in his book introduction to algorithmic marketing. What it essentially means is that there is an AI enabled system that automates the marketing process so that audiences are reached, campaigns are executed and evaluated. As shown in this picture from Wikipedia, an autonomic system is one which has knowledge of its purpose and know how to operate by itself. It is also able to sense its context and environment and use logic to operate on its inputs to produce outputs. I will illustrate with a couple of examples. Let’s say that you are a frequent train traveler or you are on a vacation trip to Europe. Your train is delayed. The system automatically sends you a message saying that your train is delayed and since you are a frequent traveler and you like gyms it has booked a free day pass at a nearby gym and says it will see you back at the train station for the 3 pm departure time. It could also include a reebok bundle for sports clothes and sports shoes. Or it might involve a visit to the nearby museum or art gallery if you are on a vacation trip. All this is handled by the autonomic system. In the second example, let’s say you have uploaded several pictures of yourself with your significant other or family member to social media. You are then shown messages and advertisements related to wedding planning, weekend getaway, ski trips or something similar based on your preferences and lifestyle choices. Again all of this is handled by an autonomic system with minimal human input. What I am trying to explore is if this is the ideal end state then how much of the marketing process is automated and AI enabled and what might be possible in the future. So let’s take a look.
If we consider the retail industry we have come a long way from the cash registers that appeared in 1960 to the digital self serve that we are seeing today. Most of the developments around digital self serve have appeared in the last 12 years or so. This is an example of the evolution from traditional staff operated super market check out to the self serve checkout we are seeing today. Digital self serve does not necessarily negate human input. It might involve the machine flagging an offer that you have missed and having a store agent pick it up for you. As you can see Pinterest and snapchat appeared in 2010 and the chatbot platforms appeared around 2015 and we will see some of the applications of AI in voice and visual search in a later slide.
Applications of AI in Promotions
When we look at the applications of AI in promotions, there are some traditional applications of machine learning including response modeling, lifetime value modeling and targeting which have been around for a while but what makes this an autonomous system is the tight coupling of the targeting server with the data management platform and analytics platform. You might imagine the request response being handled by the targeting server while the goals are set in the analytics platform. The are also some tools which enable promotion optimization and design of promotional creatives. For instance the tool from Symphony retail AI uses previous year’s promotion data and promotion plans as a starting point to forecast the current years best promotional events. It also makes recommendations on improving sales by either a BOGO offer or dollar amount or percentage off. The tool can also be used by CPGs to get strategic about their investment in creatives for advertisement placement. It automates the allocation of promotions to the circular design process for testing creatives for maximized lift and store traffic.
Applications of AI in Search
When we look at the applications of AI in search there are several applications of machine learning including the optimization of sponsored search through the use of reinforcement learning, voice search, visual search, local search. We are also seeing a significant increase in the number of product searches on Amazon and searches on the Apple platforms. According to Gartner, brands that redesign their websites by incorporating visual and voice search are likely to see a 30% improvement in digital commerce revenue by 2021. Currently the technology is able to identify specific items and retrieve similar items based on color, shape and style. Even so we are seeing different applications that enable image led experiences. For instance, Pinterest has introduced a combined visual and text search that improves intent understanding so that users are able to find the information that they are looking for faster. Alibaba has also been trialing a Fashion AI concept store which uses several AI technologies including image recognition capabilities to improve highly personalized instore experience.
Applications of AI in Pricing
When we look at the application of AI in pricing, there are several applications such as pace and perfect price which utilize machine learning and AI for dynamic pricing in hotel management, rental car and other industries. Tools such as incompetitor from intelligence node and wise Athena enable users to gain access to competitors catalogue and pricing. These tools allow users to input the website of the business and competitor information and they are able to automatically retrieve competitor prices from their websites. The tool is also able to automatically select product attributes and specifications to compute loss in sales volume or revenue when a company launches a new product. Introductory price optimization enables setting prices more accurately when there is not sufficient history by looking at similar products which have historical price information. Similarity scores are computed by using textual description and image characteristics and using natural language processing and image classification techniques.
In the future, it seems plausible that marketers are able to set high level goals and objectives such as number of conversions, number of sales and number of outcomes instead of specifying detailed targeting criteria.
Applications of AI in Programmatic Advertising and CRM
AI has also been used in programmatic advertising to enable in-depth behavioral customer segmentation, highly targeted media buying and high level customer personalization. For instance, Harley Davidsion NY implemented programmatic advertising by utilizing the platform called Albert. Albert is a self learning platform which enables autonomous targeting, media buying, cross channel execution, testing and optimization. Albert matched up the shared characteristics of previous high value customers with new users who had shown purchase intent and created microsegments of lookalike users. When it was given creative assets from Harley Davidson, it was able to A/B test thousands of variables and once the headline, visuals and campaign details were fine tuned it was able to automatically scale it to all digital platforms. Through this Harley Davidson NY was able to achieve 183% increase in user transactions and 25% improvement in overall ROI.
Programmatic CRM also enables high levels of customer personalization, real time triggers, dynamic audiences and cross channel reach. For instance, OYO which is the largest network of branded hotels in India implemented programmatic CRM offered by blueshift. Through multiple triggered campaigns, price drop alerts, booking and destination recommendations, OYO was able to achieve 5X increase in room bookings. Blueshift’s personalization studio uses event streams from different touchpoints including web, apps, other sources and CRM and collaborative filtering and other techniques to create highly personalized interactions.
Talk on Autonomic Marketing at the Toronto Machine Learning Society (TMLS) Micro-Summit on Retail and Advertising.
Having looked the applications of AI in these areas, lets consider for a moment what is possible in the future. I don’t hold a crystal ball so I don’t claim to know what will happen in the future. But if you look at what some of the experts have forecasted it is possible to draw some broad themes. For instance, it seems plausible that marketers are able to set high level goals and objectives such as number of conversions, number of sales and number of outcomes instead of specifying detailed targeting criteria. It also seems possible that media optimization becomes machine driven and completely AI enabled so that the right audience is reached at the right time for the right price. AI might also generate multiple creatives and messages which can be tried target different audiences. A new B2C engine might emerge which acts as the marketing hub that is the brain and nervous system which directs external systems to put the right piece of content in front of each individual customer. Finally, an open ID alternative might emerge to the walled gardens currently maintained by Google and Facebook. It could be an independent third party provider which provides identity as a service and not necessarily advertising services.
In terms of lessons learned, we believe that it is important to take a human first approach. It is also important to identify the touchpoints which can be enhanced by AI be it digital touchpoints or physical touchpoints. It is also essential to provide a consistent experience across all touchpoints since some people might use multiple touchpoints while others default to a single touchpoint such as web or phone.
Conclusion
In conclusion, if we consider the question of whether true autonomic marketing is possible we find a nuanced solution. Yes, it seems plausible in programmatic advertising and CRM. It is also possible that some data driven techniques are used for creative development and optimization. Novel interactions and experiences may be facilitated through chatbots and interactive AI. When it comes to promotions and pricing some functions may be AI enabled while other functions are heavily human dependent. However we are a little ways off when it comes to mitigating risks to brand reputation, incorporating proprietary models and methods in the autonomous system, elimination of bias, utilizing multiple optimization levers and developing a killer creative. Ai may not be able to come up with the next blockbuster advertisement or the innovative product or killer creative. That will still reside with humans. So in the end while we have come far in terms of autonomous marketing, there are still elements of marketing which will still depend entirely on humans.