Helping the customer: a start to NBA platforms
Building an NBA platform for exceptional customer engagement
You’ve heard talk of Next-Best-Action (NBA) platforms and how great they are at improving customer engagement, like the airline that achieved an 800% uplift in satisfaction and a 60% reduction in churn for priority customers. You are considering the introduction of such a platform in your business. Consequently, questions like, “Where does one start?” and “What is required before starting such a project?” arise. We will address these questions in this post to provide clarity on how we view NBA platforms.
Platform context
Before discussing what an NBA platform is and does, it’s helpful to set the context in which such a platform operates. Imagine a business with customers. The business exists because of products or services that help its customers in some way. From the business point of view, it wants its customers to use its products or services to their full potential. The customers want a solution to their problem efficiently. Both goals are closely aligned and making progress in both leads to happier customers and a happier business.
As a business tries to help customers, interactions occur between customers and the business. These interactions between customers and the business occur over channels. For example, a customer can visit the business’ website, which is the channel, to look at products. Therefore, the business can avoid frustration and dissatisfaction by understanding what customers are seeking when (or before) they interact. This use case is but one example of why an NBA platform is needed.
Defining NBA platforms
An NBA platform decides the best way to interact with a customer. It does so by calculating one or more actions that are relevant to the customer. These actions can range from offering a product or service, inquiring about missing information or sending a gift. Performing the action should result in the customer having a better relationship with the business. Such a better relationship would ensure greater customer loyalty and satisfaction. In simple terms, an NBA platform needs to know about customers and their interactions to determine actions within a context with feedback to improve continually. All of these ingredients form the view shown below.
Looking at the diagram, let’s start from the inside out by looking at the block representing the NBA platform.
For an NBA platform to achieve its goals, it has to answer the questions of “how” and “what”, and possibly “when”. “How” refers to the presentation of the action to the customer. It could be answered by the type of channel, language, tone of voice, colour, images and more. “What” defines the action. It is fulfilled by selecting a relevant product, service, proposition, or request. “When” could be related to the time of day or date when the action should be executed. Addressing these three questions require input, which moves us to the outside of the NBA platform block from our previous figure.
Data driving the platform
Answering the “how”, “what” and “why”, necessitate data. This data can be divided into 4 broad categories: customers, interactions, context and feedback.
Customer data is the information typically stored in a CRM system. This data should be available in near real-time so that a change is quickly visible in the platform. It follows logically that, the less data you have on your customers, the less you will be able to do with your platform. The objective, however, is not to assimilate as much data as possible to know everything about a customer. Instead, high-quality data related to your domain on customers is required.
Interaction data is stored whenever a customer interacts with the business and vice versa. These interactions are seen by the platform as either inbound or outbound communication. Inbound communication occurs when a customer interacts with the business and outbound is the reverse. Interaction data has a real-time and historical component; new interactions should be sent to the platform as they occur for real-time NBA calculation. Whereas previous interactions help in determining which actions are correct.
Contextual data is the information that configures the platform and places boundaries on its operation. It defines what is possible, for example, types of products, discount coupons available, and the like.
The last category of input is feedback. Feedback is information that is used to continually improve the platform. This information contains the decisions made and the information that was used to make the decisions. The feedback is typically used in the creation of Machine Learning (ML) models.
Platform components
We’ve covered the high-level view of an NBA platform. What follows is a brief look into the components that constitute an NBA platform.
An NBA platform consists of the components shown in the diagram below. At the start, events flow from inbound channels to an event dispatching component. This event dispatching is similar to the reception at a hotel that directs people where to go, with people taking the place of events. The event dispatcher could do some minor processing and then trigger the processing of the received data based on the channel on which it arrived. Channel processing makes sense of the event at the technical level. This is needed because two channels are rarely equal and require processing specific to the channel. After finishing with its processing, the channel processor requests actions. Upon receiving the request, the business logic applies rules and conditions that determine the possibilities or limits of the NBAs that will be calculated. For example, consider the situation in which giving a 90% discount is a possibility. Giving such a discount to all customers would result in the happiest customers but also bankruptcy. In this situation, some boundaries and eligible actions need to be defined, which fit into the business logic.
Next, the business logic requests recommendations from the ML logic. In the ML logic, ML models make predictions and prescriptions in the form of recommended actions. These recommendations are used by the business logic to determine NBAs. Based on the chosen NBA, the business logic communicates with the channel logic to perform the necessary tasks. For example, the (inbound) web channel can be configured with a cross-selling banner (i.e. what) in a sporty style and tone (i.e. how). Another example is for an outbound-related NBA. Consider when the NBA is to send a thank you message for being a loyal customer for 5 years. In this case, the (outbound) email channel would be used to send the generated message (i.e. what and how) on the day that the customer has been with the business for 5 years (i.e. when).
Do note that while we have given a simplified description of the four components that constitute an NBA platform, there is a lot more complexity behind those simple blocks. For example, the decision logic component exists as part of an MLOps process.
In addition to the four components just discussed, there are supporting services and activities to operate the platform.
- Data storage — the secure storage of data used and created by the platform.
- Monitoring — ensures continued functioning. Some of the types of monitoring include operational monitoring of the platform, technical monitoring of the software, security monitoring of the platform, and performance monitoring of the ML models.
- Modelling — are those data science activities related to creating and deploying ML models.
- Experimentation — is related to modelling through analyzing data and experimenting with new models.
- Reporting — refers to dashboarding and the generation of reports to inform stakeholders on KPIs and metrics.
Before getting a platform
To end this post we wanted to share some considerations before plunging into an NBA project.
Know your customer
As stated earlier, an NBA platform can do more when more information on customers is available. If you have no information on your customers then it won’t help to get your NBA platform. More information can also be collected on your customers provided you comply with regulations, such as the GDPR, and ethically use customers’ data. At the least, customers’ consent is needed to start using their data in an NBA platform.
You can’t improve what you don’t measure
This principle is true in general but applied to an NBA project it means that end-to-end visibility on customer interactions and customer feedback are needed. For example, the platform will only irritate customers if it shows the same product offer because it does not know of a previous purchase of the product.
Take everyone along
An NBA platform will not make your Marketing teams or any other teams obsolete. In fact, without expert knowledge to assist in the creation and improvement of the platform, its realised impact will be less than its potential. Taking people along in an NBA project requires leaders that encourage those involved to contribute and collaborate.
Unlock data for usage
Any data-driven business should be applying this principle. It doesn’t help to store data if it gets locked away deep within the bowels of a system never to see the light of day again. What is needed is secure access to all data sources related to an NBA platform. Enabling these capabilities before starting an NBA project helps to avoid data bottlenecks during the project.
We have explained how an NBA platform recommends the best action for a customer to take. The goal of which is to improve the interactions and relationships between the business and its customers. Various business and technical practitioners work jointly on an NBA project to ensure its success. On a high level, their work centres around the development, operation and improvement of an interface, channel logic, business logic, decision logic, and support components that constitute the platform. Finally, we discussed the importance of knowing customers, exposing their data securely, measuring the end-to-end information flow, and taking everyone along on the project before it starts. With this post, we hope to have given visibility into NBA platforms to enable you to initiate such a project.
If you have any comments or like to discuss more, please contact me.
About the author
Jacobus Herman is a data engineer at BigData Republic, a data science consultancy company in the Netherlands. He specializes in data platforms on Azure and AWS.
We provide expertise in machine learning, ML ops and data engineering. If you need advice or are yourself interested in creating machine learning solutions, feel free to contact us at info@bigdatarepublic.nl.
This blog was created with help from Sven Stringer, Annieke Hoekman, and Erik Jan de Vries.
