INside Performance Marketing
Q&A: Cordial Busts Myths around Email and Machine Learning

Q&A: Cordial Busts Myths around Email and Machine Learning

Cordial’s David Baker on key misconceptions around email marketing and machine learning.

As the machine learning technology develops, its impact on email marketing is growing. However, there are still many misconceptions across the industry around both email marketing and machine learning.

To find out more about them - and how the industry is fighting against those myths - PerformanceIN caught up with David Baker, chief strategy and operations officer at Cordial.

First of all, David, could you tell us what the main email marketing misconceptions are from your perspective? And what is the industry doing to change this?

David Baker: There are four main misconceptions with regards to email marketing that the industry is struggling to embrace:

  • Email is cheap
  • Email doesn’t have the scale of other media
  • Personalisation is the holy grail
  • Sending more email, more frequently, is the key to success.

First and foremost, there is still a “catalogue marketer” and “email 101” mindset that is campaign oriented: send an email on a Tuesday at 8am and to improve, send on Thursday and maybe Saturday. This is obviously different for those selling products and those that are selling eyeballs (aka publishers). Most are just having a hard time keeping up and are struggling to try new things working with the same processes and approaches we were using a decade ago.

Email is cheap, relative to other media channels, and it receives a fraction of the budget, due to the perception that email doesn’t scale. Looking at it from afar, it makes sense. An email list may be 50,000 or 5 million contacts, and when only 10-20% are engaging, the numbers don’t impress the C-suites when media can project audience exposure to ads at 10-100 times that. But there is a big conundrum in that most email marketers cite personalisation as the number one goal of their programme; the elusive 1:1 vision. In actuality, many think of this in terms of mail merge “Dear <First Name>”, not true personalisation that improves as the relationship evolves and behaviours change.

Similar to the proverbial cat with no fur on its paws looking at the fish bowl with a piranha, some are scared and can’t move out of production mode to see new ways of addressing content, personalisation, automation and optimisation. But it’s coming soon. The market is changing so fast, it will force changes in programs; some we are already seeing with the digital native companies that grew up online. The industry is addressing this as we see new companies popping up that are focusing on data sciences; answering key questions about preferences recommendations, and audience views (value of customer, churn predictions) and some are focusing just on automation and trying to broaden brand’s messaging portfolios with trigger email.

Machine learning has been a popular topic in performance marketing debates lately. What’s your take on machine learning in email marketing?

DB: Overall, it is largely misunderstood. If you buy into the scale challenge and perception that email is cheap to blast out, then machine learning is somewhat of an oxymoron. If all you are doing is blasting emails, you really don’t learn a great deal about device, timing, cadence, frequency, all of which machine learning can help you address in predictive ways.

The key to machine learning is that it helps make decisions faster, but does require fresh data and the need to continually feed the machine. This is not new. We’ve been building models for decades, it just took longer to build and the models weren’t as adaptable as behaviours and market conditions change.

Today machine learning is a buzzword, much like artificial intelligence. Ten years ago we called it modelling and self-learning models. Overall, it’s having a dramatic impact on today’s programmes. In our view, machine learning starts with recommender algorithms related to what someone purchases, might purchase and recommendations related to products and/or content. Next, there is audience-based machine learning that is related to engagement, churn and “lifecycle or audience”, purchase propensities all associated to “state” based behaviours and events. This is somewhat taking recency, frequency and monetary (RFM) analysis to an actionable level where it can be continually improved. And then lastly, there is innovation using various sampling methods and algorithms that are shifting how people think about A/B testing and applying machines to optimise in real time based on behaviours and test variables the marketer is experimenting with. We tend to call it experimentation vs testing, and machines are essential to doing this in real-time.

This is critical to digital communications. The world has morphed into time, place and device-switching consumers who behave differently in each situation. Machine learning works well when there is rapid iteration, where the value in real-time optimisation can take effect, vs learn, test, analyse and optimise. We are seeing the industry evolve with triggers, behavioural automations and where many are thinking about mobile email and web email differently in a more programmatic way. We think about consumers in terms of  “lean forward” and “lean back” moments. Gesture or mouse-based interactions are all elements that affect how people will react to advertising, promotions and content. And the shelf life of this insight is very short. The only way to keep up and be relevant is to augment marketers intuitive and creativity with ways to process things faster and the core of machine learning does just that.

What are the key challenges machine learning poses?

DB: Latency is a major challenge. Many operate too slowly to take full advantage of machine learning. Machine learning isn’t a learn, analyse and deploy method. Depending on how you deploy machine learning, in some cases you need to compile data and let the models learn and improve over time. In other cases, it’s an “explore” and “exploit” approach that allows you to optimise in the market. Many struggle to grasp the breadth of machine learning and how it can apply to their programmes. Another challenge is that data has a shelf life, so it is somewhat counter intuitive to have long cycles to make decisions. A browse event means very little a week or a month from now.

The biggest challenge is accelerating business decisioning to match the pace of the observations. You have to know what you want to know, you can’t just hope the data will reveal things. While it will, the hypothesis is important to drive direction, or you can end up in an endless loop of data and information that isn’t actionable. As confidence in the data increases, so should the risk profile (e.g. take more chances) - the amount of risk in applying decisions in market will far out accelerate any potential negative outcome. It’s similar to the theory of perfect vs on time. Rarely does the extra effort to make something perfect outweighs the return of getting it to market faster. The same principle applies. If marketers can get a handle on this, then they can apply it to seasonality, competitive conditions, consumer shifts and generational divides.and generational divides.

Lastly, there is a myth that you need a stable of data scientists to make this work. Sure, if it’s a company charter, you should build a team. But there are so many open source tools, companies specialised in this and talent in the contractor space that support this, there really is no excuse for not exploring, outside of “time” to do it.

More generally, what trends are you expecting to see take over the industry in the near future?

DB: Generally speaking, companies are more focused on interaction management today than ever. Previously, this was “triggers,” such as welcome triggers, birthday emails, cart abandonment email, etc. Today, companies are thinking about time and events that can be used to create opportunities to engage with consumers in a push/pull mentality vs a blast-and-observe approach.

At Cordial, we’ve seen a major shift from legacy platforms that can’t adapt to the real-time nature of consumer interactions to a more real-time, self-optimising interaction management that is co-mingled with the entry point (web, mobile app, PoS). It’s safe to predict that a major disruption to the creative process for email is coming. The days of long email production cycles is coming to an end. We are seeing a world where content is multi-purposed, templates with logic that can fetch content and assembles emails instantaneously based on real-time inputs. I also see a world where email is timed to the consumer patterns, devices and location.

Overall, the industry is putting more of a premium on speed as the catalyst for being more creative and better leveraging marketer's intuition. But we still have a long way to go in many industries.

Continue the conversation

Got a question or comment – post on Twitter, Facebook or LinkedIN.

Monika Komar

Monika Komar

A News and Features Reporter at PerformanceIN, Monika covers stories and developments in the fast-evolving world of performance marketing.

Monika studied Modern Languages at the University of Southampton and worked in marketing and communications before making her way over to PerformanceIN.   

Read more from Monika

You may also like…