The future of AI in marketing
This is an edited version of the AI section in Econsultancy’s Future of Marketing report. It covers:
- the nature of narrow AI
- use cases in marketing (baked-in to martech vs. bespoke)
- future use cases – creative AI, NLP and customer service, the marketer’s assistant
- limitations of AI in marketing
The nature of AI
The future of AI, whatever industry you want to focus on, is a strange discussion for the layman because the technology in its current state of maturity seems a mix of stunning triumph and inarguable work-in-progress.
I can speak to my phone, say “Okay, Google, show me photos of Ted” and my Pixel will quickly display images of my oldest son that it has accurately categorised in my Google Photos app.
There are cars that can drive themselves, to a large extent.
And yet, current AI systems have difficulty with causality and don’t seem to demonstrate reasoning.
Gary Marcus is a professor in the Department of Psychology at New York University and was previously founder and CEO of Geometric Intelligence, a machine learning company later acquired by Uber. In a paper titled The Next Decade of AI, he contrasts robust intelligence with what he calls “pointillistic intelligence, intelligence that works in many cases but fails in many other cases, ostensibly quite similar, in somewhat unpredictable fashion.” Memorably, Marcus illustrates his point by demonstrating the limitations of the neural network GPT-2, the text-generating model developed by OpenAI. Marcus shares some tests of GPT-2, giving the system sentence fragments with which to generate a continuation. They are enjoyably absurd. For example:
“If you break a glass bottle that holds toy soldiers, the toy soldiers will probably… follow you in there.”
“Even with massive amounts of data, and new architectures,” Marcus argues, “the knowledge gathered by contemporary neural networks remains spotty and pointillistic”.
As Brian Bergstein puts it in an article for MIT Tech Review titled What AI still can’t do, “It’s as if you knew that the presence of clouds made rain likelier, but you didn’t know clouds caused rain.”
The right way to apply AI in marketing
It’s important for marketers to understand the nature of narrow AI as a tool. AI may not yet be robust enough to reason with a customer, but it is very good at solving well-defined problems.
“It should be taken as a given that brands and marketers will be applying AI technologies in their marketing efforts as the future unfolds. The big question about an AI-Powered marketing future is how best to apply it,” Parry Malm, founder of AI Copywriting tech company Phrasee, tells Econsultancy.
He continues, “Adding AI to your marketing stack makes sense only when you have a specific problem that AI is well-suited to solve.” Even when we’re talking fairly narrow AI, Malm notes that “Finding just the right way to apply AI to the problem at hand is key.”
In this way, AI is already being widely employed.
AI use cases – the future is already here
There’s a stat from a recent Econsultancy survey (in collaboration with Tealium) that shows the future is already here. Of 529 executives responsible for their organisation’s marketing data and technology in Europe, 49% say that AI or machine learning is either in practice today for their data and marketing efforts (24%) or is currently in testing (25%). This proportion would arguably be higher if these respondents consider the machine learning that has been part of marketing tech for a number of years.
Machine learning is baked in to much adtech and martech and performs all sorts of optimisation and analysis. The most obvious examples are within Google, Amazon and Facebook, where at its most basic, content and ads are shown to people whom are judged to be most likely to click. Algorithms assess content and user behaviour in order to optimise for ad/content quality and advertiser outcome.
In a similar vein, machine learning may be used for merchandising in retail (eg. product recommendations, category sorting) or content recommendations, and predictive analytics (e.g. understanding who is likely to buy or to churn).
It’s helpful to understand the types of machine learning model that can be employed, to give an idea of their uses in marketing.
Machine learning algorithms often tackle the following tasks:
- Classification – e.g. keyword categorisation or product categorisation. These algorithms are trained with labelled datasets.
- Clustering – Algorithms look for patterns in data without being given the ‘right answers’ to match for. This is used to find new segments of customers.
- Regression – Results in a numeric value, an estimated optimal value e.g. for dynamic pricing or sales forecasting.
Even visual search, a technology that may seem more complex than some of the examples mentioned so far, is available as a plugin for Shopify or as part of social listening tools. And though I won’t suggest that implementing all of these solutions is quick and easy, it’s clear that there are now many tools and platforms that use machine learning and allow customer experience or communications to be optimised, without the need for a whole room of data scientists. This tech can be grouped together broadly with marketing automation.
Bringing bespoke AI in house (to some degree)
Beyond the baked-in capabilities offered by martech, there are marketers and ecommerce professionals working with their own AI in house.
This might be using plug-and-play machine learning (ML) models, such as those offered by Google Cloud, or more bespoke work but with similar aims to that found in some platform AI. For example, Mark Douthwaite of consultancy Peak AI told Econsultancy about his work with Footasylum, a UK-based sportswear retailer, which wanted to invest in personalised marketing communications.
Peak “worked to analyse [Footasylum’s] customer data so it could have a view of the people most likely to be both engaged with the brand and in the market for its products. From these customer profiles [Peak] created bespoke algorithms that distributed hyper-personalised product recommendations. [The] work achieved a 28% uplift in revenue per email sent by Footasylum, and when [Peak] implemented the technology into its social media advertising campaigns. In another solution for Footasylum, [Peak and Footasylum] achieved a massive 10x higher than industry average return on ad spend (ROAS) across the campaign.
In another example from back in 2017, Ben Chamberlain, then senior data scientist at Asos, told Econsultancy about his first project at the retailer, looking “at how we could predict the customer lifetime value and how we could use that information to improve shopping frequency and average basket value.”
In a paper from that year, Chamberlain et al. describe the CLTV prediction system deployed at Asos.com, which “provides daily estimates of the future value of every customer and is one of the cornerstones of the personalised shopping experience.” The authors discuss the challenges of forecasting value, predicting churn and evaluating loyalty, including “the unusual distribution of the target variable,” given that a “large percentage of customers have a CLTV of zero.”
Clearly, programming these systems in house can lead to more accurate models that may offer competitive advantage and reduce losses.
But away from bespoke models, even looking again at Google Cloud and its literature for the aforementioned plug-and-play ML models, you get some appreciation of what is available for those organisations with enough internal data science expertise. According to Google, “data analysts can perform predictive analytics right out of the box and with high accuracy.” There’s documentation to help predict CLV by using AI Platform on Google Cloud, which requires data that contains at least a customer ID, purchase amount per customer and a purchase date. Results can be improved by adding more types of data to the deep neural network, such as clickstream events, user profiles, or product features.
Case studies showcased on the Google Cloud website include 20th Century Fox using AI and ML to predict box office performance, and Hearst Newspapers using a natural language API for content classification (“allowing content creators to see patterns” ) as well as integrating ML into a CDP to segment users by reading habits (to “forecast content performance to increase ad revenue”).
A decade into the future
There is still some sci-fi-like future-gazing to be done. Despite all the current uses of ML, some ideas just out of reach still fascinate people (and marketers are people).
The three in particular that come to my mind are:
- Creative AI
- NLP, allowing for automation of customer service
- AI as a marketer’s assistant (to make suggestions and improve processes)
There have been plenty of experiments with creative AI, such as IBM Watson’s 2016 work to create a movie trailer by analysing the visuals, sound and composition of hundreds of existing horror film trailers, and selecting scenes from the completed movie (Morgan). You can watch the trailer on YouTube. John Smith, IBM Fellow and Manager of Multimedia and Vision at IBM Research, does make the assessment that AI only performs an assistant role here, and one could argue that the assertion on IBM’s website that this work was successful in “ultimately reducing what could be a weeks-long process to one day” is specious (particularly given the how narrow is this sort of AI).
Smith makes a very salient point, though – “We still have to define what creativity means. We know some of the attributes have to do with finding something novel, unexpected, and yet useful.”
Perhaps then for marketers, the logical application of creativity is within a fairly rigid design framework, where automation will offer more profound time savings. The example I am thinking of is from Alibaba. The company reportedly used its LuBan AI platform during the 2017 Singles’ Day shopping festival to design 400 million banners to advertise a range of products. In a blog post about the platform, user experience designer Ke Xu estimates this output to be equivalent to the work of a human graphic designer working non-stop for 150 years. Designers do provide feedback to the algorithm as well as evaluating final products, but the machine ‘learns’ general design frameworks and is trained on thousands of examples. This is clearly a step beyond dynamic creative optimisation, the technology used to create ‘personalised’ ads when retargeting customers by inserting product shots and names. Indeed, the Luban AI is selecting product shot, background, logo and decorating artefacts.
Not all marketers require this scale of automation (few retailers have as many SKUs as Alibaba), but it’s easy to dream of a future ecommerce platform which optimises design and is given little more than a product feed.
NLP and customer service
We’ve already briefly touched on the limitations of natural language processing and generation. But despite a convincing conversational AI seeming some way off, there has been obvious increase in the use of automated chat in the last five years, even if decision trees still play a large part in these experiences.
In an MIT Tech Review article, Karen Hao reflects on the loss of jobs in the call centre industry during the Covid-19 crisis and the resulting investment in chatbots. Hao cites Otsego County, New York, as an example – the state governor reduced government staff by 50%, call centre employees were cut, but calls were rising. The county worked with Watson Assistant to automate the offer of guidance and medical information. The article reveals IBM saw a 40% increase in traffic to Watson Assistant from February to April of this year. Google has reacted to demand by launching the Rapid Response Virtual Agent, a special version of its Contact Center AI.
The most commonly cited example of chatbot tech is perhaps KLM, which has been using Messenger as a service channel since March 2016. Tech created originally by Digital Genius was used to provide confidence levels to agents, who can choose to personalise or approve automated messages. The automation threshold can be set at 90% confidence level, for example. This human-assisted bot system can create significant efficiencies.
NLP is also used in sentiment analysis, which may be employed in the triage of customer support queries. Emotion recognition, in the voice and face, as well as text, is a big area of research and arguably another potential candidate for service or advertising. Though it might be easy to think of ‘Black Mirror’ scenarios here (“sir/madam, I am detecting anxiety/mistrust/aggression in your behaviour”), as usual, employment of this type of AI wouldn’t necessarily be helpful in many business scenarios – does a telco need to understand 20 shades of slightly cheesed-off in order to run a successful call centre?
The marketer’s assistant
Aimee Falk, Global Content Strategist at 3M, was one of Econsultancy’s interviewees for the Future of Marketing report. Falk said of AI, “It could also be used to automate the marketing for marketers, automate their marketing process to make suggestions to marketers, making recommendations for marketing nets, pulling in the data collected…”
Everyone who has struggled with an analytics interface or asked their data team to create custom dashboards can likely relate. Google Analytics has its Analytics Intelligence insights, which tell the user, for example, “Your ecommerce conversion rate decreased on some landing pages.”
Acoustic (formerly IBM Watson Marketing), which offers products such as pricing, personalization and analytics, hints on its Story page at a future for marketing free of the drudgery of process – “At Acoustic, we’re reimagining marketing technology. It’s too important not to. When the complexities are hidden and the right tools are in place, you’ll free your inner marketer and indulge your wildest strategies.
That leads us neatly on to the first limitation of AI in marketing…
Limitations of AI in marketing
Hiding complexity (as Acoustic refers to), is one of the issues with AI. That hidden complexity could include bias, derived from an unrepresentative dataset. “Data used to train AI is often skewed toward a single demographic, white men,” James Vincent writes for The Verge in an article about facial recognition, “and when a program sees data not in that demographic it performs poorly. Not coincidentally, it’s white men who dominate AI research.”
Hidden complexity could also simply include inefficiency or error, but without data and machine learning expertise across a marketing organisation, this may remain ‘hidden’ until it’s too late. Is data clean? Is it useful? Complexity could also include where data comes from, and whether its use complies with privacy regulations in different markets. Data sharing is one method of increasing the accuracy of AI, but such sharing is rightly harder to do now than ever.
Data strategy as a whole is the big elephant in the room and the pre-cursor to successfully implementing AI, even a solitary platform plugin. Integrating a CRM with a chatbot, for example, is no mean feat.
Eddie O’Brien of insight consultancy Destination 5.0> writes that in the wake of Covid-19, “Smaller corporates that didn’t have a data and digital transformation strategy in place will be scrambling to become fitter for the future.” He adds “The Insight Industry has become fragmented… We’ve seen the rise of user research, UX, service design, behavioural economics, digital analytics, big data solutions, along with new and innovative methods such as AI, all complimenting, or sometimes even competing directly with traditional insight areas, such as market research.”
Arguably, those businesses already using machine learning to look for patterns in data are in a better position to respond to changing customer behaviours during and after a pandemic, even if those changes can also skew the results of some algorithms in the short term.
For all the great benefits of AI and ML in marketing, it’s clear the technology is still simply a tool, albeit a powerful one. The good or bad news for our industry (depending on how you look at it) is that AI is not going to turn bad marketers into great ones anytime soon.
Download the Future of Marketing report, or for more on AI and marketing:
- How machine learning is transforming retail both online and offline
- An overview of AI in copywriting
- Digital innovation in customer service
- A marketer’s guide to AI and machine learning
Source: Customer Experience