“Machine learning allows us to build software solutions that exceed human understanding and shows us how AI can innervate every industry.” – Steve Jurvetson, co author of Hacking Growth
In the Saas world, the most common reasons for users to churn are due to a lack of engagement, poor product market fit, bugs in the system, a lack of usability, or lack of support. What this all boils down to is a disconnect between what the customer was promised before purchasing, and what they received in the post-purchase environment.
On the surface, this sounds like an easy fix, but simply changing your pre-purchase communication to better match expectations with reality, or improving customer service may not always be enough. This, in part, is due to the fact that users don’t always read the material you provide to them, or reach out when they are disconnecting from the service. The online landscape is changing and machine learning has made a strong presence in the effort to close the gap between expectation and reality.
Later in this post, I will go into the ways in which we can use machine learning to better match these expectations, and offer more to our users without burying them in information, questionnaires, and onboarding flows.
Starting at the buying process, let’s deep-dive into what users are currently going through to better understand what is going wrong and why they are dropping out down the line. Below, I have mapped out my version of the full buying process to help visualise where the disconnect is taking place.
The mismatch of expectation starts to crop up from the Comprehension phase up until the Mature adoption phase. This is simply because, regardless of user preferences, the product or service will often promote the best case scenario in its communication. This is not necessarily a bad thing, but in order to get to that best-case scenario users need to align perfectly with the product or service. If that alignment comes down to how their experience is set up from the beginning of the buying process, then we know we can apply machine learning at an early stage.
Another common issue can arise after the user signs up for a free trial or demo. During this trial period, while useful for a large portion of prospects, we still tend to see poor scrutinisation of the product or service by the user, as they are not properly comparing key metrics such as cost to value, ease of use, and adaptability for their organisation. Unfortunately, free trials are often initially gauged as a success or failure by the overall impression they give the user. In my experience, it is very common to see users highlighting real issues relating to the product or service only once the full membership as been purchased, as suddenly they are invested and there are consequences to their actions.
With an understanding of where the problem is taking place, more or less, we can now move on to why and how we should use machine learning to better match the user’s expectations.
In a Cambridge University article on machine learning within the realm of choice and preferences, it has been shown that users often cannot state preferences in advance, but construct their preferences as they see the available options. This may not seem like a big deal, however, when the list of available options can be seemingly endless, as with some complex services, it becomes critical to get these preferences right in the shortest amount of time. Getting this right in the free trial or demo set up phase is crucial to long term success and overall conversion rate.
To apply this to a real world situation, let’s take a look at skateboarders. Skateboarders do not typically write down all of the tricks that they can perform. If we were to ask them to choose from a list of all the tricks they were capable of doing within their skill level, it would be hard for them to complete the task, mainly due to the fact that choosing from an exhaustive list of options with subtle differences would take too long. It would be far quicker to show them different real examples of tricks being performed, which require very particular skills, and then map those skills to lists of similar styled tricks. After showing the skaters a variety of different tricks and asking them if they could land them or not, it would quickly provide us with a long list of potential tricks that each skateboarder could most likely perform. The real difficulty for machine learning lies in how closely you can map those similar tricks to the most relevant key skills in order to get the most accurate result.
While some services have started using machine learning to help guide their users, they often fall short at basic hurdles, such as failing to recognise one-off purchases like a toilet seat, spamming their users with in-service ads pushing them to buy more toilet seats as if they were a fashion accessory. As stated by Pearl Pu and Li Chen in AI Magazine, “many tools used today do not satisfactorily assist users to establish this kind of machine learning because they do not adequately focus on fundamental decision objectives, help them reveal hidden preferences, revise conflicting preferences, or explicitly reason about trade-offs. As a result, users fail to find the outcomes that best satisfy their needs and preferences”, which ultimately starts them down the path of dissatisfaction and eventually churn.
Machine learning is teaching us to take a step back from how we view our product or service to understand the layers of commonality behind what the consumer actually wants and what we can actually provide them with.
Using machine learning to reduce churn at the beginning of the buyer journey is very useful and the same attention should be applied to your long term user base. Machine learning can also be applied to long term users simply by looking for patterns in their behaviour on your platform or service. According to Jonathan Tarud from Koombea, machine learning can help to predict user preferences or behavior, which can then trigger alerts or actions when it appears the user is disengaging from the product or service.
For growth hacking, UX, and marketing in general, we have come full circle in the sense that the mantra has always been “less is more”, but this no longer necessarily stands true. There used to be an immense push back against adding too many functions, features, or customisations, fearing that too much choice would cram the user interface and add complexity for the user. As we enter into 2019, however, the trend is shifting to “more is more”, as increased customisation is gaining momentum with users demanding more flexibility and more options. Thankfully, we can allow these needs to be met knowing that machine learning will do a far better job than us at making personalisation and customisation work for both sides.
“R.I.P. analysis paralysis” – Everyone