No matter if talking about “artificial Intelligence”, “machine learning”, “predictive analytics”, “data mining” or “big data” – all of these hot topics from the general area of artificial intelligence are predominantly concerned with one and the same thing: learning from the wealth of available data – not least for tailor-made and therefore successful email marketing. How is email turnover developing with each individual customer? How can spam be effectively identified and filtered out? What is the “next best action” to be taken? Artificial intelligence is providing exciting new potentials for email marketers.
Never before has there been so much data, so many tools and the levels of computing power available as exists today. And at the same time, artificial intelligence is no longer science fiction. Its beginnings can be traced back to as early as the 19th century – and the method of least squares, from Carl Friedrich Gauss, is still in use today in trend forecasting. Concepts of neural networks have also existed for more than 50 years. For some time already, filter programs such as SpamAssassin have been using such models to classify emails that are desired in the inbox and those that aren’t. While email marketing “works” using rules-based principles, artificial intelligence goes a step further and makes forecasts and places content based on predefined algorithms and available data.
Analysis of human language
Gmail from Google is a good example of an intelligent system. The Gmail “Priority Inbox” classifies new emails as being important or less important. Users train and personalise the algorithm by individually indicating which emails are of interest to them. Recently, Gmail introduced an additional inbox feature called “Smart Reply”, which recognises and interprets human language. It deduces the email content and intention – and then also suggests corresponding replies. This lets users answer emails, for example a request to attend a meeting, with an automatically generated reply such as “Yes, I can confirm the appointment” or “No, I cannot attend”, via a single click.
Automatic optimisation of subject lines and copy
The likelihood of an email being opened is largely based on the interplay of a known sender, an interesting subject line and the right timing. Artificial intelligence can already assist in identifying the most promising text modules and keywords. It is possible for marketers to allow the decision between “15% off jeans”, “John Doe, jeans reduced by 15%!”, or “A jeans surprise for you…” to be made automatically. The algorithm decides which option is the most promising for the particular group of subscribers. Subject line specialists such as Phrasee , Touchstone or Persado even go beyond the possibilities given by classic test processes. The latter provider is specialised in the automated analysis of emails, social posts, etc., and uses the results to create copy that features an emotional approach to effectively address the reader.
Customising the time of dispatch
Besides the subject line and content copy, the time of dispatch can also be precisely personalised. It is wise to take this into account as opening and click rates can vary significantly according to the time of day or day of the week dispatch takes place. While some users check their emails first thing in the morning, others prefer to do so during their lunch break or even just before going to bed in the evening. Once again, artificial intelligence is used for the automated optimisation of dispatch times. The actual dispatch times are based on an analysis of all historical opening and click logs for each individual user. Ideally, all subscribers will be sent emails at their personal ideal time – even if the distribution list is extremely large. The probability that the email will be opened and clicked will thereby be correspondingly high.
Intelligent tagging of user clicks
The combination of query forms and behaviour-based profiling measures is particularly important for customised content. Marketers should therefore ensure that they explicitly request to be given information on individual interests and demographic data, upon suitable occasions. At the same time, valuable know-how on the recipient level can be obtained through implicit analyses of clicks and user movements. As such, through click profiling for “interest in family holidays on the Baltic coast”, for example, the corresponding group of recipients can be “tagged” very precisely for future communication. In addition, agile templates and recommendation systems have already taken the next step forward from target groups to individual users. Intelligent algorithms learn from current and historical user reactions and automatically generate tailor-made offers. Such recommendations can furthermore be adjusted using templates and in real time up to the actual moment that each individual email is opened. This functionality is provided “out of the box” by larger specialist providers.
Anticipating interest via next best actions
Also providing potential are so-called “next best actions”. These are based on available response data combined with information on past demand and customer behaviour. Through the linkage of the customer profile and contact history, probabilities for the subsequent user behaviour can be determined for each individual user – which can then be used for specific recommendations for action. In a sense, a bet on future user interest and behaviour is thereby being made. The email content is based on forecasts regarding which incentives and what specific constellation or sequence will be particularly promising. In addition, “open” data such as weather forecasts or search trends can also be incorporated in “next best actions”.
Summary: artificial intelligence has already made inroads to the email inbox. Yet there are many useful applications that are already available on the market but have so far not been adopted on a wide scale. But in the long term, I personally would not rule out that there will be a future tectonic shift from rule-based communication to automated machine-based placement of information.