Decisions, decisions, decisions … How do digitally enabled companies cope with it?
Research by Twilio reveals that COVID-19 accelerated companies’ digital communications strategy by an average of 6 years while 97% of enterprise decision makers believe the pandemic sped up their company’s digital transformation. In this article, Rodrigo Barreto explores the role of data, analytics and Machine Learning based decisioning as enablers of end to end processes that are key to digital transformation.
Digital transformations are occurring at lightning speed. Companies are looking to change the focus from operational functions to end to end processes to ensure a more efficient, customer centric experience. The three end-to-end processes which companies delivering digital services will invariably implement are Order to Activation, Usage to Payment and Complaint to Satisfaction. These are essential for sustainable delivery of services to users and digitally enabled companies need to get them right or risk failing. Then, there are other end-to-end processes which may be addressed as part of a large transformation programme or may be addressed as independent projects, these are processes such as Campaign to Lead and Insight to Loyalty.
What do all of these processes have in common? Generation of data to gain better insight as well as further interaction with customers.
One user decides to browse for a film or a product online. Plenty of data can be collected regarding the browser behaviour, product selection, abandoning the order half-way or proceeding to confirm order. Now, take all of this information from a large universe of users – a lot of information can be taken away. This can be used by the digital platform to provide insight into new interactions which can vary from a simple change in the visuals of a webpage to very targeted promotions and suggestions of products or services.
Another user is not happy with services received from a company and decides to make a complaint. Digitally enabled companies understand well the ripple effect that dissatisfaction can have. Because they are customer centric, they frequently live and die to keep customer satisfaction metrics at healthy levels; quite a few even implement customer sentiment analysis in social media to automatically identify messages of dissent. Again, a lot of data is collected and analysed; the fight is to manage damaged reputation and turn bad situations around so that instead of having dissatisfied customers, the company actually receive praise for the way complaints are dealt with. This is only possible by the speedy identification of underlying issues causing the customer complaint and the actions that can be taken and that are known to generate customer satisfaction, in either case, decisions are taken by the digital platform based on learning from masses of data collected from the broader customer base.
A third user has been using too much or too little of a service and, sooner or later, realises it is not getting the best value for money. For transactional services, this may also be reflected in levels of discount that customers receive as part of promotions. If customer satisfaction is the number one concern, digital services companies want to make sure users perceive the best value for the services and/or products they pay for. Ensuring customers are in the right price plan or receive promotional discounts consistent with their behaviour can also boost customer satisfaction metrics. It is therefore particularly important that the digital platform be able to either automatically move customers to different tiers or strongly suggest that the customer take action to move to a better-value tier. These decisions and suggestions, once again, depend on learning that the digital platform derives from masses of data.
Now, you might be wondering, how these companies are able to do meaningful analysis on such huge amounts of data and come up, in real time, with decisions in their interaction with customers. There are a few elements which are crucial to a well-oiled digital platform.
Know thy customer
A few eyebrows will rise here, especially from those who are more mindful of data protection policies and regulations. Worry not, the digital platforms usually (but, unfortunately, not always) use a fool-proof system to replace the real identity of a customer or user with a marketing identity. This is done at the point of interaction with the user and makes it impossible (or, at least, extremely difficult) for data stored and analysed by these companies to be associated with information which would allow identifying individuals (the so called personally identifiable information).
An identity management system is positioned at the intersection between the company’s channels of communication with customers and the company’s internal systems and data stores. The channels need to know and authenticate the identity of users. The identity management system maintains the mapping between full customer identity and marketing identity. For full compliance with Data Protection regulation, the identity management platforms tend to work hand in hand with consent management, resulting that data is not stored as a result of an interaction if there isn’t a consent from the user for that to happen.
There are several types of information that can get associated with a marketing identity. Some are abstracted from details provided by the customer or authorised to be collected when joining a service. These may be the credit score, the place of residence, gender, occupation, etc. All needs to be abstracted and taken to a level which ensures that the customer is not personally identifiable. More data is collected, over time, from the direct interaction of the user with the company and the services it provides. Every order, every complaint, every request, every record of usage of a service adds up to the data associated with the marketing identity.
But it is not only data collected by the digital services company that gets associated with a user’s marketing identity. Through so called Data Management Platforms, digital services companies acquire additional information from partners (2nd party data) and from other data exchanges (3rd party data). These data sources all leverage web cookies which identify generic attributes which can be used to form audiences. For instance, there may be a cookie for people who recently bought baby milk formula. When the attributes for the specific cookie id are known, this cookie can be identified by another company during the interaction of the user with its digital services and can be used to, for instance, select baby clothes to be displayed as a suggested product or a ‘Pilates for new mums’ video to be added to a list of suggestions. A ‘cookie sync’ process is used to map cookies’ unique id’s with the traits they represent and map these traits to a user’s marketing id.
What not many people know and is a more contentious topic, is that even when you are not logged in to your digital services, and provided you are not browsing in incognito mode, your identity can still be derived by digital platforms when you visit their online services. This is because your footprint, determined by stored cookies, locations and types of device used, allows the digital platform to derive your presumed marketing identity with a high degree of certainty.
All these data types – directly collected, augmented and indirectly derived – are available to the digital platform. A large range of very specific events can be selected for in-memory processing and that forms a timeline of events akin to a movie, a ‘Customer Movie’.
Tell me who you walk with and I will tell you who you are
Although we like to think of ourselves as unique, the truth is, we are rather predictable. Marketing professionals have been exploring, for many decades, the fact that people that fit within certain categories or segments have similar preferences. Big Data processing technology allows huge amounts of data to be continuously sifted through and rigorous statistical models can be applied to identify cohorts of market identities with similar traits and to calculate the propensity of certain behaviour or preference to happen. These are not one-off or very time-spaced data classification exercises; these are round-the-clock continuous analysis with machine learning type feedback so that at every second the ability to predict gets better and adapts to dynamic events, take the COVID-19 lockdown as an example, that may sway users preferences and behaviours.
Suppose you are a young professional, married, leaving in an affluent neighbourhood. You have been travelling but are back home. You frequently dine out and the weekend is approaching. Chances are you will invite your partner to dine out in a special place. Now, suppose you use a digital service which pushes special offers for ‘distinguished’ users. If well equipped, the digital platform will have been fed with the propensity scores for you to pick different types of offers. It can select the offers for which you have the highest propensity to accept and push these offers to your app. Should you accept it or dismiss it, the digital platform will feed back your action. It might have been the case that you are moving to a slightly different cohort or the cohort you belong to is collectively changing behaviour or preference. At your next interaction with the service, this will already had been accounted for.
But it is not only to push new sales or consumption that propensity scores are calculated and used. Propensity scores are also calculated to rank reasons for dissatisfaction and the risk that a user will simply stop using a service or break the relationship with a company. These are costly events that not only affect marketing KPI metrics but also cost money. The ability to predict the propensity of such events before they happen is very important. Early visibility gives companies the ability to try to remediate issues or offer inducements which keep dissatisfaction at bay. The cost to attract new customers is often higher than the cost to stop customers moving away.
Every decision has consequences
The digital platform is constantly listening to events. It does so by subscribing to many types of events that are fed into data lakes and other operational and enterprise data stores. Information associated with events are transmitted in real-time data streams. When the platform detects a given set of events which represent a natural opportunity of interaction with the user or a situation which is likely to require pro-active action, it needs to draw upon the history associated with the particular identity. It goes through the customer behaviours and uses cognitive techniques to infer the types of actions it can take. It then maps these actions with relevant propensity scores and decides on the best action, i.e. the action which has the highest probability to cause a positive result.
The best action can be pushing an offer of a product or service to the top of a list the user is about to browse, suggest that a customer care agent propose a certain remediation action while the user is still going through the interactive voice response menu to file a complaint, modifying a file to be sent to a printing house so that the customer monthly bill has a suggestion for changing of plans, etc. While there is some latency allowed for some types of interaction, for the majority of events via online channels the time gap between detection of an event, making a decision, and implementing an action has to be below one second. Large digitally enabled companies do just that, thousands of times a day, on a 24/7 basis.
Wrong decisions can cost dearly. It is not granted that the decisions will always be right but these platforms depend on capturing feedback about the result of an action and self-calibrating so that the next time a similar decision needs to be made, the probability to get the decision right is continually increasing.