Artificial intelligence and machine learning are an increasingly integral part of many industries, including marketing.
But while we often talk about using or incorporating AI in marketing, what do we really mean by that? What does it look like in practice?
Here are 15 examples of AI and machine learning in action in the marketing industry (P.S. remember to check out Econsultancy’s Marketer’s Guide to Machine Learning and AI).
1. Product/content recommendations
The practice of clustering customer behaviours to predict future behaviours began way back in 1998, with a report on ‘digital bookshelves’ by Jussi Karlgren, a Swedish computational linguist at Columbia University. In the same year, Amazon began using “collaborative filtering” to enable recommendations for millions of customers.
Fast forward to 2019, and some of the most successful digital companies have built their product offerings around the ability to provide highly relevant and personalised product or content recommendations – including Amazon, Netflix and Spotify. As Lori Goldberg wrote in a brief history of artificial intelligence in advertising for Econsultancy, “This all comes from AI-based clustering and interpreting of consumer data paired with profile information and demographics. These AI-based systems continually adapt to your likes and dislikes and react with new recommendations tailored in real-time.”
Other major brands are now following suit with their own AI-powered recommendations. For example, Sky has implemented a machine learning model that is designed to recommend content according to the viewer’s mood. As customers become used to the level of personalised recommendations provided by services like Netflix and Spotify, they look for other brands to provide the same experience.
Publishers are also implementing AI-powered content recommendation widgets that can identify related content to surface to readers, and even personalise those recommendations based on readers’ browsing habits. We recently introduced a feature like this on Econsultancy: our ‘Recommended’ sidebar is powered by a tool called IDIO, which learns our readers’ interests as they browse and uses this to suggest articles that they might be interested in reading.
2. Data filtering & analysis
Marketing is becoming an increasingly data-driven discipline, and more effective use of data is the key to improving customer experience, personalisation, targeting and more.
However, consolidating that data at scale once it has been collected, and analysing it to determine patterns, is tedious and difficult for human beings. This is where AI comes in: one of the great strengths of AI in the workplace is its ability to take on complex organisational and analysis tasks that would be difficult or impossible for humans to carry out, freeing up humans to do the more intuitive, creative work that they are better suited to.
For example, AI can be used to improve account selection in account-based marketing when ABM is carried out at scale. Targeting and personalisation company DemandBase found that it could use AI to filter out companies from its list of prospects that would ultimately lose the company money in the long run, as Jessica Fewless, the company’s VP of ABM Strategy, told the B2B Marketing Conference:
“We’re a stats-based company and if they churn from us in less than a year we lose money on them. So, we took the elements that made those customers churn and removed them from our model,” she said.
AI proved to be most useful for DemandBase in identifying ‘timely intent’: highlighting the accounts where there is a window of opportunity to approach before a commitment to a competitor is made.
3. Search engines
AI has had a profound impact on the way that we search, and the quality of the search experience, that we often tend to take for granted in 2019.
Google first started innovating with AI in search in 2015 with the introduction of RankBrain, its machine learning-based algorithm. Since then, many ecommerce websites (including Amazon) have followed in Google’s footsteps and incorporated AI into their search engines to make product searching smarter.
With innovations like natural language processing and semantic search, search engines can determine the links between products and suggest similar items, find relevant search results, and auto-correct mistakes, helping consumers to discover products even if they don’t know exactly what they’re looking for.
4. Visual search & image recognition
Similarly, advances in AI image recognition and analysis are making it possible to achieve amazing things with visual search.
While it’s still early days for the technology, visual search – the act of using search to find results that are visually similar to one another, in the same way that “traditional” text-based search finds results of a similar topic – is becoming more commonplace thanks to platforms like Pinterest, and technology like Google Lens.
Visual search has a number of useful applications in marketing and retail. For example, it can be used to improve merchandising and personalise the shopping experience: instead of recommending products based on a shopper’s past behaviour or purchases, visual search technology can recommend relevant products based on how they look, helping shoppers to find items of a similar or complementary style.
Target and Asos are two retailers that have made a big commitment to visual search as part of their ecommerce experience. Target launched a partnership with Pinterest in 2017 that integrated Pinterest Lens, Pinterest’s visual search tool for the physical world, into Target’s app, allowing shoppers to snap a photo of a product while out and about and find similar items on Target’s website.
Asos’ Style Match visual search tool works in a similar way, allowing shoppers to take a picture or upload an image and search Asos’ product catalogue for items (or similar items) contained within that image. These tools encourage shoppers to treat retailers as go-to destinations for items that they might see in a magazine or while out and about, helping them to shop for the perfect product even if they don’t know what it is.
Finally, image recognition is also giving marketers an edge on social media by allowing them to find uses of brand products and logos, and identify visual trends. This is called ‘visual social listening’, and it can allow brands to spot where and how customers are interacting with their brand, logo or product even when it’s not referred to by name.
5. Social listening & sentiment analysis
Advances in natural language processing have proven extremely useful for marketers wanting to analyse their brand presence, and the conversations around their brand, on social media and use those to target campaigns.
AI allows brands to perform sentiment analysis on social conversations and understand the prevailing attitude towards their brand and products. This can allow them to spot potential issues and counteract them before they become too widespread. For example, Samsung – which works with AI consumer insights company Crimson Hexagon – was able to detect and counteract customer dissatisfaction with a red tint on the screen of its newly-released S8 smartphone model thanks to social listening.
Careful with Samsung S8 as I read yesterday some south koreans have bought S8 models that have a red tint on the display & are disappointed
— Søren Pio Nissen (@Soren_Pio) April 22, 2017
— Standa Dvořák (@s_t_a_n_i_k) April 19, 2017
Social listening and sentiment analysis can also be used to spot purchase intent by analysing the ways that consumers are talking about a product – for example “In the market for a new phone. Samsung S8 anyone? How does it hold up?” or “Borrowed my boyfriend’s iPad and now I’m thinking about getting one…” – which can enable marketers to target them with advertising or potentially a strategically-placed discount.
With that said, marketers should tread lightly with this kind of targeting or risk appearing creepy.
6. Product categorisation
Some online retailers and aggregators have discovered the extent to which machine learning can make the process of tagging and categorising products more efficient. Stuart McClure, founder of LoveTheSales.com, spoke to Econsultancy editor Ben Davis to explain how and why the company uses AI for product categorisation:
“One retailer might give us amazing data, and another could give us the same set of products but with awful data. We use a text-based classification tool, training various models with both positive and negative examples.”
This means that even if different language is used by different retailers to describe the same product – for example, “trainer”, “basketball shoe” and “sneaker” – the algorithm is able to understand that the products are the same and tag them accordingly. This can be so effective as to allow the algorithm to correctly identify a product based on nothing but context:
“The really cool thing is, we’ll have examples, loads of them, where you’ll get say 100 shirts and there’ll be a piece of data that has nothing in it at all to say it’s a shirt, but the model has classified it correctly as a shirt because of the surrounding context,” McClure told Econsultancy.
7. Product pricing
Demand-based price changes are nothing new – think changes to hotel room rates depending on the season – but with AI entering the equation, prices can be determined and optimised with a whole new level of precision, taking a wide variety of data into account.
Machine learning can be used for things like dynamic pricing, which analyses a customer’s data patterns and predicts what they are likely to be willing to pay, and also their receptiveness to special offers. This allows businesses to target them with more precision and calculate the exact level of discount needed (or not needed) to pull in a sale.
Dynamic pricing can also be used to compare businesses’ product pricing with that of their competitors, to determine if their pricing is too high, about the same, or too low.
AirBnB is one brand that has built and refined an extremely sophisticated dynamic pricing system to help property owners determine the price that they should list their property at. It takes into account a wide variety of factors including geographic location, listing features, local events, photographs and reviews, as well as market demand and time to booking date.
These calculations are provided to users as ‘price tips’, which they will choose to follow or ignore; the system will then monitor whether or not the listing succeeds, and adjust its algorithm based on the results. You can read more about this intriguing system and how it was developed in this profile published by IEEE Spectrum.
8. Predictive analytics
Predictive analytics, the practice of extracting information from data sets to predict future trends, can be used to great effect in improving customer service and customer experience.
Predictive analytics are a revolutionary capability of AI because it was previously only possible to retroactively determine trends from data sets. Thanks to artificial intelligence, things that could once only be determined retroactively can now be reliably modelled, and decisions made based on those models.
Predictive analytics can be used in ecommerce to analyse customers’ purchase behaviour and determine when they might be likely to make a repeat purchase or to purchase something new. Using predictive analytics, marketers can “reverse-engineer” customers’ experiences and actions to determine which marketing strategies resulted in a positive outcome.
Companies like FedEx and Sprint are also using predictive analytics to pinpoint customers who are “flight risk” factors and may defect to a competitor.
In customer service, predictive analytics can be used to anticipate high or low call volumes and ensure that phone lines (and other outlets) are staffed sufficiently.
9. Audience targeting & segmentation
For marketers to reach their customers with the level of personalisation that many have come to expect, they need to target increasingly granular segments.
AI can be used to achieve this. Drawing on the data that marketers already have about their customers, machine learning algorithms can be trained against a “gold standard” training set to identify important variables and common properties, and even pick out incorrectly identified contacts.
The extent to which marketers can segment their consumers comes down to the data that they have – segments can be as simple as gender and age, or as complex as past behaviours and buying personas.
Segmentation also doesn’t have to be static. Dynamic segmentation is an application of AI that takes into account the fact that customers’ behaviours are rarely fixed or unchanging, and that people can take on different personas at different times for different reasons.
For example, if a young person browses for a gift for an older relative, dynamic segmentation will group them in with the segment most appropriate to their current buying behaviour using real-time data, presenting the most relevant offers and avoiding using outdated data for targeting.
10. Programmatic ad targeting
The introduction of artificial intelligence has made bidding on and targeting programmatic advertising vastly more efficient. Again, this is tied into predictive analytics and the ability to model things that could previously only be determined retroactively.
When applied to programmatic advertising, AI can determine things like the best time of day to serve an ad, the probability of an impression converting, or the likelihood of a user engaging with an ad that appears in the middle of an article they are reading.
AI can also be used to adjust bidding strategies based on customer lifetime value (CLV) and invest more in potentially higher-value customers.
11. Sales forecasting
Sales forecasting is another prediction-based application of AI – this time, for sales.
Using past sales data, industry-wide comparisons and economic trends, artificial intelligence can forecast sales outcomes and help companies to inform business decisions and predict short and long-term performance.
Sales forecasts can also help to estimate product demand, although sales teams should be careful to take other factors into account as well: for example, a company experiencing manufacturing issues may only sell a certain number of units due to a lack of stock, not due to a lack of demand for the product. Thus, using only sales figures to predict demand would produce an inaccurate forecast.
12. Chatbots & conversational AI
Over the past few years, the chatbot star has risen dramatically and then fallen even more dramatically, as a once-hyped application of AI was found not to work as well as many had hoped.
However, a number of companies are still successfully using chatbots. At a recent Econsultancy Digital Outlook event in Singapore, marketing AI expert Deborah Kay spoke of the popularity of Singapore’s ‘Bus Uncle’ chat platform, which uses Facebook Messenger to give information on bus arrival times, and responds to natural language queries.
Other brands have begun to build conversational voice experiences, placing a bet on voice interfaces and voice-enabled devices as the future of brand interaction. Trainline, a travel company with a number of apps that incorporate AI, recently launched a voice app for the Google Assistant. In an interview with Econsultancy editor Ben Davis, Trainline’s Director of Engineering Jonathan Midgley called the app “the UK’s most advanced rail voice AI, with 12 levels of conversation depth”. The app improves itself through machine learning, and becomes more accurate the more commuters make use of it.
13. Speech recognition
Over the past several years, voice-activated devices and their potential have become the talk (ha) of the marketing industry.
This is possible due to advances in speech recognition technology, as well as things like natural language processing. In 2017, Google’s level of speech recognition accuracy reached the coveted 95% threshold, while in the same year, Baidu claimed to have reached a 97% accuracy rate with speech recognition – and is aiming for 99%.
While speech recognition is only one component of a good voice experience, it does play an important rule in making sure that voice interfaces and voice interactions function smoothly, and that users’ requests are interpreted correctly.
14. Computer vision & augmented reality
Computer vision, much as the name implies, is a discipline that involves programming computers to ‘see’ the world around them and gain a high level of understanding from digital images and videos.
True computer vision is only achievable with AI, machine learning, and huge datasets with which to train a machine to recognise and identify objects.
Accurate computer vision is important for the sophisticated development of augmented reality (AR), and particularly for its applications in marketing. The better computers can detect and identify the physical world, the more accurately and usefully augmented reality can be overlaid on top of it.
This enables AR advertising that integrates with people’s surroundings in a way that is relevant without being intrusive and can open up new possibilities for things like interactive shopping, product insights and offers, and business information.
So far, augmented reality has been deployed to great effect in a marketing context by home improvement and furniture companies like Home Depot, Lowe’s and Ikea. In 2015, Home Depot launched an augmented reality app called Project Color that allowed users to visualise how different paint colours would look on their walls.
Since 2013, it has also offered augmented reality features within its mobile app that allow shoppers to view how products like doors, patio furniture, vanity units and faucets would look in their home. Lowe’s and Ikea offer similar functionality with furnishings, and Lowe’s has also developed an app called Measured which – as the name implies – can measure an object or distance using augmented reality, a useful tool for DIY.
Beauty brands are also deploying AR to great effect to engage consumers and drive return purchases. Brands including Lancôme, L’Oreal, Estée Lauder and Sephora have created augmented reality experiences that allow people to virtually try on lipstick or nail varnish, provide personalised product recommendations, and even carry out a full makeover via AR, educating consumers about the products they need and how to apply them.
Lurking behind a good half of the articles published about AI is the implicit question: will the robots be taking all our jobs?
Most of the use cases for AI that we’ve so far covered in this article consist of artificial intelligence either augmenting a tool or performing a role which aids a human in their day-to-day work.
However, one job that AI is becoming able to perform in its own right is copywriting. At the moment, AI’s copywriting skills in the realm of marketing are largely limited to basic tasks with limited parameters like email subject lines and product descriptions.
One well-known vendor in this space is Phrasee. The company prides itself on being able to produce email subject lines that are indistinguishable from those written by humans – even using emoji correctly – and has recently moved into the realm of social ad copy on Facebook and Instagram.
Chinese tech giant Alibaba claims to have developed an AI copywriting tool that “passes the Turing test”, but it’s unclear whether the AI has actually been put through a quantitative test, or whether this is just a turn of phrase to signify that the AI can pass for human. Even then, it still isn’t being used to produce complex copy, but mostly produces ads and retail product descriptions for websites like Taobao and Tmall.
However, something much bigger may be just around the corner. Earlier this month, the non-profit company OpenAI, backed by Elon Musk, revealed that it had created an artificially intelligent text generator called GPT2 which is so accurate, the company is holding back from releasing the research until it has had more of a chance to investigate how the technology might be abused.
Journalistic outlets including Reuters and The Washington Post are already employing ‘robot reporters’ to churn out reports on company earnings and sports scores based on structured data. While these articles are still very formulaic in nature, the creation of GPT2 – which has demonstrated its ability to perfectly replicate the tone of individual columnists – suggests that something much more ‘human’ is possible.
Whether this AI will be capable of comprehending the finer points of SEO or martech and thus take over writing blog posts (like this one!) from human copywriters or editors remains to be seen. For now, it seems that GPT2 is mostly capable of mimicking, rather than devising original ideas and executing them – but it mimics very, very well.
So, is there anything AI can’t do?
We’ve found plenty of examples of ways that AI is transforming the ways that we work, shop, market and sell, allowing us to achieve things that would never have been possible without it. However, AI is by no means all-powerful.
It’s worth pointing out that AI and machine learning still need people, such as Google’s raters, to improve their accuracy and to train algorithms properly.
Crowdsourcing of a workforce (e.g. Amazon’s Mechanical Turk) will perhaps become a bigger industry as more AI brings a need for a human’s guiding hand to adjust datasets.
In addition to this, AI has yet to dramatically reshape most businesses to the extent that many expected it would. Brian Bergstein published an insightful recent write-up for the MIT Technology Review examining why this is. In it, he wrote,
“AI might eventually transform the economy—by making new products and new business models possible, by predicting things humans couldn’t have foreseen, and by relieving employees of drudgery. But that could take longer than hoped or feared, depending on where you sit. […]
“This doesn’t necessarily mean that AI is overhyped. It’s just that when it comes to reshaping how business gets done, pattern-recognition algorithms are a small part of what matters. Far more important are organizational elements that ripple from the IT department all the way to the front lines of a business. Pretty much everyone has to be attuned to how AI works and where its blind spots are, especially the people who will be expected to trust its judgments. All this requires not just money but also patience, meticulousness, and other quintessentially human skills that too often are in short supply.”