By Ben Davis @ Econsultancy
Artificial intelligence (see the Wikipedia definition), specifically machine learning, is an increasingly integral part of many industries, including marketing.
Here are a whole bunch of case studies and use cases, as a complete primer for AI in our industry.
1. Recommendations/content curation
Predictive analytics allows Netflix to surface and finesse recommendations. This kind of clustering algorithm is continually improving suggestions, allowing users to make the most of their subscription.
Uniting information from diverse datasets is a common use of AI.
Under Armour is one of the many companies to have worked with IBM’s Watson. The sports apparel company combines user data from its Record app with third-party data and research on fitness, nutrition etc.
The result is the ability for the brand to offer up relevant (personalized) training and lifecycle advice based on aggregated wisdom.
IBM explained as follows in a press release:
A 32-year-old woman who is training for a 5km race could use the app to create a personalized training and meal plan based on her size, goals, lifestyle.
The app could map routes near her home/office, taking into account the weather and time of day. It can watch what she eats and offer suggestions on how to improve her diet to improve performance.
2. Search engines
In late October 2015, Google admitted it was using RankBrain, an AI system, to interpret a ‘very large fraction’ of search queries.
RankBrain should mean better natural language processing (NLP) to help find relevance in content and queries, as well as better interpretation of voice search and user context (e.g. Google Now).
Andrew Howlett, CTO of Rain Agency, told AdWeek:
Customers providing reviews use real language and say it in ways that people might also ask a question through Google. For example, in a review, someone might say, ‘This place has the best chips and salsa anywhere that doesn’t cost a fortune.’
That sentence will help now with someone searching for something like, ‘I’m on a budget, where’s a good restaurant with awesome chips and salsa?’
Machine learning is, of course, nothing new at Google, already used in search, advertising and YouTube recommendations.
3. Preventing fraud and data breaches
Analysing credit/debit card usage patterns and device access allows security specialists to identify points of compromise.
The relevance of AI is not just for card issuers, though.
Retailers, for example, have been subject to high profile data breaches (e.g. Neiman Marcus) as a result of a system based solely on usernames and passwords (without any stronger type of authentication).
This area of security analytics has been around for years but is becoming more sophisticated. Solutions have to react quickly to new fraudster tactics and analyse unstructured data, too.
Natural language processing (NLP) can be used to look at text within transactions, for example, transforming it into structured data.
Newer AI implementations, such as that used by the United Services Automobile Association (USAA, which provides financial services for ex-military), will identify anomalies in behaviour even on the first instance.
Could Neiman Marcus have benefitted from AI to prevent a data breach?
4. Social semantics
Looking at Microsoft’s AI chatbot ‘Tay’ (and its regrettable tweets), one might not immediately think Twitter and AI go together like salt and vinegar.
However, deep learning (a term used often to refer to machine learning on large datasets – a neural network recognising abstract patterns) has plenty to get its teeth into on social.
Sentiment analysis, product recommendations, image and voice recognition – there are many areas where AI has the potential to allow social networks to improve at scale.
It’s worth having a trawl through Facebook’s AI research to see the many possibilities, if you’ve got a head for scientific whitepapers, that is.
Wired magazine covers a particularly novel use (outside of Facebook’s social network) – the tech giant analysing overhead images of topography to find evidence of human life.
Such technology may allow Facebook to target its internet-providing drones at communities that need them.
5. Website design
The Grid is an ‘AI’ website design platform.
Intelligent image recognition and cropping, algorithmic pallette and typography selection – The Grid is using AI in certain areas to effectively automate web design.
6. Product pricing
With thousands of products and many factors that impact sales, an estimate of the price to sales ratio or price elasticity is difficult.
Dynamic price optimisation using machine learning can help in this regard – correlating pricing trends with sales trends by using an algorithm, then aligning with other factors such as category management and inventory levels.
7. Predictive customer service
Knowing how a customer might get in touch and for what reason is obviously valuable information.
Not only does it allow for planning of resource (do we have enough people on the phones?) but also allows personalisation of communications.
Another project being tested at USAA uses this technique. It involves an AI technology built by Saffron, now a division of Intel.
Analysing thousands of factors allows the matching of broad patterns of customer behavior to those of individual members.
The AI has so far helped USAA improve its guess rate from 50% to 88%, increasingly knowing how users will next contact and for what products.
A graphic from the Saffron website.
8. Ad targeting
As Andrew Ng, Chief Scientist at Baidu Research, tells Wired, “Deep learning [is] able to handle more signal for better detection of trends in user behavior. Serving ads is basically running a recommendation engine, which deep learning does well.”
Optimising bids for advertisers, algorithms will achieve the best cost per acquisition (CPA) from the available inventory.
When it comes to targeting of programmatic ads, machine learning helps to increase the likelihood a user will click. This might be optimising what product mix to display when retargeting, or what ad copy to use for what demographics.
9. Speech recognition
Skype Translator now supports Arabic, English, French, German, Italian, Mandarin, Brazilian Portuguese, and Spanish.
One reviewer of this product using the recently added Arabic speech translation said that mistakes occurred but ‘with enough patience, I usually got the message’.
Translation of speech has come so far due to progress in neural networks over the past five years.
I can’t pretend to understand the logic and computing involved, but if you want some background info try this blog post from Andrew Gibiansky.
Siri and Cortana and other personal assistants also use speech recognition of course, so doubtless most savvy consumers are quite aware of how accurate they are.
Speech recognition is particularly important in the Chinese market, where using a keyboard to type small and intricate characters can be laborious. Baidu is making big strides on this front with voice search.
10. Language recognition
Behind speech recognition sits the challenge of language recognition. Not what you’re saying, but what it means (in relation to other things and concepts).
This is another use of machine learning that consumers are familiar with, given its use in search.
However, language recognition may be increasingly used by brands to digest unstructured information from customers and prospects.
WayBlazer is a so-called ‘cognitive travel platform’, a B2B service using IBM’s Watson AI to power consumer applications from third parties such as hotel chains and airlines.
So, for example, images, recommendations and travel insight are personalised depending on customer data.
This might be unstructured text, e.g. ‘We want a family beach holiday with plenty of kids activities but also culture’.
11. Customer Segmentation
Plugging first- and third-party data into a clustering algorithm, then using the results in a CRM or custom experience system is a burgeoning use of machine learning.
Companies such as AgilOne are allowing marketers to optimise email and website comms, continually learning from user behaviour.
12. Sales forecasting
Conversion management again, but this time using inbound communication.
Much like predictive customer service, inbound emails can be analysed and appropriate action taken based on past behaviours and conversions.
Should a responce be sent, a meeting invite, an alert created, or the lead disqualified altogether? Machine learning can help with this filtering process.
13. Image recognition
Google Photos allows you to search your photos for ‘cats’. Facebook recognises faces, as does Snapchat Face Swap.
AR relies on sophisticated recognition of the landscape in front of you, include other people, to accurately overlay holograms.
Perhaps the coolest implementation of image recognition is DuLight from Baidu. Designed for the visually impaired, this early prototype recognises what is in front of the wearer and then describes it back to them.
Of course, for marketers the uses could be manyfold, from content searching to innovative customer experiences.
14. Content generation
At the moment, content generation is chiefly done using structured data. Wordsmith is a platform that alllows the automatic generation of news articles, for example, from financial reports.
This relies on the reports being fed into a CSV in the right way – it’s essentially automation.
However, in the not-too-distant future, the plan is to do this sort of content generation with unstructured data.
15. Bots, PAs and messengers
Leaving probably the most notorious example to last, chatbots are thought by many to be the future of user input on mobile, replacing apps.
Simply talking or typing to a chatbot will allow a service to be delivered through the analysis of natural language combined with understanding of a brand’s datasets.
As Techcrunch points out, Facebook’s platform, previewed at F8, could conservatively soon lead to chatbots replacing ‘1-800 numbers, offering more comfortable customer support experiences without the hassle of synchronous phone conversations, hold times and annoying phone trees.’
So, is there anything AI can’t do?
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.
However, if you’re doing a job that could conceivably be automated (and there are many listed above), AI could be more and more of a pressing issue.
Ben Davis is a senior writer at Econsultancy.