Using Digital Twins to Manage Customer Relationships

“Digital twins” are commonly associated with Industrial Internet of Things installations, in which they are widely used. They generally work as follows: the sensors from a machine are mapped to a digital abstraction, or “twin,” making it easier to monitor the machine and know when it needs maintenance. The digital twin can also be used to model the lifecycle of the machine, predicting when, for example, individual parts are likely to fail. In fact, any physical asset that is receptive to monitoring and prediction, such as a city or even the human body, could benefit from a digital twin.

But the “digital twin” as a concept can also be extended in a different direction—to non-physical modeling. One particular area where it can be effective is in managing customer relationships, i.e. tracking and accurately predicting a customer’s needs. Basically, it can enable businesses to offer “the right product at the right time.”

Most businesses deal with customers statically: a customer’s needs are predicted based on historical data, and products are offered using risk analysis and profitability assessments. By the time all of this manual research is finished and an offer is made, however, there is a good chance that the market has moved in another direction, leading to markedly decreased profits, even if the customer accepts the offer.

A more effective approach is to make a digital twin for each customer that dynamically tracks, for example:

• the products they consume
• their levels of satisfaction
• other behavioral elements

By analyzing these data points together and in a streaming fashion, offers can be made to a customer faster and at the perfect time, for example, when certain threshold levels have been reached.

Customer data for a digital twin can be gathered from CRMs, account-based sales strategies, logs, order processing info, and other sources, and is ideally collected in a data lake. Models are trained on the data using various analytical techniques (machine learning, etc.), and the trained models are held in the digital twin, which runs them in real time.

A technology particularly suited for such a digital-twin approach is the Tarantool Data Grid, which features an in-memory/disk combined database along with an application server.

In Tarantool, the application server holds the rules engine for the digital twin, which uses the aforementioned precalculated models, along with lambda functions held in Java, Python, or R containers. There is also a data “window” that streams dynamically updated flow views (and as mentioned, data ends up in a lake). Tarantool holds in memory the current state of the model and a list of events upon which to react. Less relevant events are transferred to disk and the disk also holds an archive of all of the source flow data. For connectivity, Tarantool features a REST API for integrating with any UI, pub/sub for subscribing external systems to the events, and connectors to export data to external systems like Kafka and relational DBMSes.

And Tarantool is fast and consistent, as you can see from the graphic below, which shows the performance of 120,000 transactions per second in five flows, executed on a cluster of eight servers:

Tarantool also features key-value, JSON and SQL (in alpha) structures; fully ACID transactions; sharding; replication within clusters and among data centers; querying by GraphQL; geospatial indexes; a proprietary queuing mechanism; connectors for all mainstream relational databases; and drivers for most mainstream languages.

Hopefully, at this point you are imagining some of the advantages that customer-digital twins could bring to your business, and you can understand why Tarantool is particularly suited for the use case. To get started with Tarantool, you can Download Tarantool Here, read the Getting Started Guide, or Contact Us to discuss your project in greater detail.