A multinational technology company that set the standard for customer service and support in its industry wanted to develop a preventive maintenance program to detect, diagnose, and proactively fix faults in hardware and systems before customers experience any problems.
The aim of this use case was to leverage telemetry and systems health data from key infrastructure components and combine that with customer, order, technical support, CRM, and product data. The resulting data would power BI dashboards, an alerting system to predict failures, and customer service systems that assist with troubleshooting.
In addition to the core capabilities above, the customer wanted to be able to ask ad-hoc questions of the data such as, “How many devices in this device family are exposed to a bug in any geography AND what is my warranty cost exposure over the next quarter on those devices?” or “How many customers by geography will potentially be impacted by a particular support issue that was introduced by updating a particular firmware version on a given device?”
The client needed direct access to an actionable, integrated view of their data across multiple datasets. However, the volume and complexity of the source datasets posed a greater than expected challenge. The telemetry data alone had climbed to over 780,000,000 records and was generating upwards of two to three terabytes of updates per day. The data was stored, warehoused, and then moved into a 100-node Hadoop-based data lake using traditional techniques. The data lake was intended to serve as a pre-aggregation point to combine all of the data together into a single analytical view. However, due to the sheer amount of data flowing into the systems, it took 10,000 CPU cores and over 600 hours to prepare the data for analysis. Even then, queries to the data lake would take hours and sometimes days to execute depending on the complexity of the query.
KEY OBSTACLE— The effort and time involved in every update combined with the latency of the resulting queries prevented this customer from solving the business problem.
How Molecula Helped—
Molecula was asked to conduct a proof of concept to solve the challenge while the customer evaluated several potential solutions.
Molecula’s competition focused on improving the end user experience, but none of the other approaches were able to solve the underlying latency problems. Molecula loaded the entire dataset into FeatureBase, and demonstrated real-time, ad-hoc queries on the the entire dataset in fractions of a second on a single laptop. With FeatureBase, these sub-second queries were not just possible on each dataset, but on real-time JOINS across customer, order, technical support, CRM, and product data.
Molecula won the business, and the customer eliminated the time, infrastructure, and latency associated with pre-processing and querying their massive amounts of data. This enabled customer service teams to make data-informed decisions at the speed of thought.
“Data that once took weeks to collect, store, and process from multiple systems in order to be analyzed is now available for analysis the second it is updated in the source systems.”
FeatureBase also reduced the engineering time required to prepare the data for analysis by 50%, reduced infrastructure costs by 10X, and gave the client the ability to ask any ad-hoc, granular question of the data in real time without having to re-process any data.
Once the latency problem was solved for the original preventive maintenance use case, several new use cases became possible, including dynamic warranty pricing and real-time asset management. In addition to the cost savings, Molecula’s solution led to new revenue opportunities for the client.