Predictive maintenance for a car owners' association
Predictive maintenance app for car owners increased customer satisfaction resulting in higher association membership retainment and increased garage bookings, lower incidents on road due to serious notifications regarding engine and battery faults.
What the business got
Challenge
An automotive association looked for a new way of serving their existing customers.
A multi-million user base targeted with any solution on day one has to be scalable and reliable. With a very recognizable brand, any public flop would have resulted in negative press coverage, which led to extremely slow development and too much fixation on details that never mattered.
Strategy
Harnessing decades of car maintenance data, they decided to create predictive models for their customers, so they can schedule a garage visit before the breakdown becomes a costly roadside event.
Our delivery
We delivered the end-to-end first viable product with full cybersecurity, data privacy and reliability coverage.
Both their regulated environment and their internal group processes were fully satisfied with our technical documentation, we worked with the solution architects and all related fields from day one to ensure launch goes smoothly.
During our engagement, the scope needed normalization on multiple fronts and while we kept all capabilities that were necessary, we vastly simplified a few areas, e.g. notification logic from a 14 page specification was dropped to 2 sentences and worked well even months after launch.
Outcome
As launch window arrived, all key fields immediately let the green light:
- cybersecurity
- business continuity
- data privacy
- DevOps & platform
- and architecture committee.
The launch quickly scaled and succeeded in delivering the new value by the first engine and battery predictions.
Client testimonial
"We were focusing on so much the tech components, and forgot the end to end user in the process until Sandor's team pulled us into the same room"
— product owner.
The team setup
Scope
Build · create new components
Integrate · reate adapters for existing apis
Customize · configure and set up existing tech
Reuse
Event stream · decouple
3 squads
Product owner and tech lead and principal data engineer and principal data scientist and solution architect, one per squad
3 backend engineers
4 mobile engineers
3 data engineers
5 data scientists
3 integration engineers
Whose people
ours
the client's
The container architecture
Highly scalable, multi-cloud approach was the core to serve the appropriate workloads: Azure data lake for the daily model inference over large data set, while AWS application workloads and DynamoDB served the customer facing experience.
Data lake separately handling a non-production daily inference load and monthly retraining and fine-tuning of the model weights.
While the main application swallowing daily tens of thousands of active users, the core of the system was decoupled with event streams and a modular approach utilized materialized views per customer to serve rapidly.
Complex integrations were necessary between a number of mature software systems under already large production loads.
The delivery calendar
In-app payment for scheduled garage and work moved after public launch saving over 6+ months of waiting for a viable product.