I caught up with a local real-time decision making start up recently called Inkiru. Inkiru’s mission is to revolutionize real-time decision making by using the increasing volumes of data available to drive both insight and action. The core team is ex-Visa, PayPal, eBay, Chase, amazon with a focus in payments, fraud detection and security. The company has 22 employees now, mainly in engineering and data science/analytics and has one major client that has been co-developing the platform with them. Inkiru see a clear opportunity in providing both predictions and real-time actions based on predictions. In particular the move to mobility combined with the pace and volume of data being generated is forcing companies towards real-time analytics and decision making. Their initial target is one of the industries most impacted by this, retail.
The primary driver for developing a new platform was their experience in real-time predictive analytics. Real-time predictive analytics is a hard problem, regardless of whether you are talking about real-time modeling (where models adapt to data as it streams in) or even to embedding real-time scoring (of a model built earlier) as part of a transaction time that must be measured in milliseconds. They felt that the lifecycle of predictive analytic model development, deployment and updating required a new platform for a real-time environment. In particular they wanted a platform that handled runtime execution of models natively and that handled large numbers of data sources so that they could be available both to build the model and to score the model when it was deployed.
The platform is envisioned to have four products:
- The initial phase is what they call Adaptive Intelligence. This is a meta data driven cloud-based solution for data augmentation, predictive analytic deployment and configuration driven dashboards. This product is dependent on customized models developed by Inkiru.
- The next steps is to produce an Enterprise Intelligence platform for dedicated/on-premise deployment that also allows customers to use their own/contract analytics teams to add models.
- Phases 3 and 4 involve packaging up solutions to create standard models and offer this as both as Decisions as a Service (DaaS) offerings for specific decisions and as a PaaS offering for people to add their own rules and models.
The core of the platform is a decision engine runtime for deploying Predictive Models and an Eclipse-based Decision Workbench for configuring the decision engine. This is combined with a noSQL data store to support flexibility in data changes. Data augmentation is a key focus area to the platform discovers schemas automatically as data sources are added. For performance reasons the data store creates a graph as data arrives and continuously updates summary metrics. This allows data that arrives in real-time to drive analytics in real-time.
Decision logic is handled by their own business rules engine and predictive analytics through a combination of their own modeling techniques as well as R and SAS algorithms. While most models are built offline and then scored in real-time the engine also supports machine learning to allow models to be built and updated as data flows in.
The tool has a fairly robust syntax, allowing high level concepts to be expressed simply, but they are very focused on performance due to their real-time focus so the language is still fairly technical. The business rules capability can be used to support Champion/Challenger and A/B testing but this not built-in and model / decision performance reporting is coming soon.
Architecturally the configured decision is deployed as a classic Decision Service capability that delivers real-time “advisories” or answers as configured by the workbench. Results can be returned in real-time or be called separately in batch and the deployment can also support an event processing architecture.
The product has a web-based “desktop” for displaying dashboards/metrics/model performance/results and logs for rule execution. It would be good if they could bring some editing capability to this environment (especially for business rules) in the future but today all editing is within the Eclipse environment.
The workbench, security, event steam processing and analytics environments are available today with self-learning, visualization and various data technology plug-ins coming soon. These capabilities have been packaged up into Fraud/Loss and Know Your Customer/Authorization packages today with plans for Inventory, Segmentation and others coming in the future as they move beyond fraud and into marketing and contact optimization as well as inventory management etc.
You can get more information on Inkiru here and Inkiru is one of the vendors in our Decision Management Systems Platform Technology report.