Dynamic Pricing Accelerator
The Dynamic Pricing Accelerator illustrates a reference architecture that shows how Predictive Analytics and Streaming Analytics can help businesses move towards an algorithmic business model. It provides a platform to improve competitiveness with bespoke pricing models and reduces reliance on expensive black-box software. The platform evaluates the model effectiveness in real-time giving the opportunity to take immediate action in response to changing market conditions.
Also see this article on the TIBCO Blog by Richard Price explaining the concept of the Algorithmic Insurer:
(since the last release of Insurance Pricing Accelerator 1.0.0)
August 14, 2018, release of Dynamic Pricing Accelerator 2.0.0
- Upgraded to latest SB10.4.4
- Implemented rebasing of models to complete analytics loop
- Updated DXP to include live dashboard
The Dynamic Pricing Accelerator is illustrated with a use case from the Insurance market. However the principles applied are valid across a range of business domains.
Historically, the insurance software market has been dominated by a handful of specialist vendors whose products can be both expensive to deploy and difficult to customize without the involvement of the IT department.
Operated by IT departments largely as “black box” solutions, leading insurance firms are effectively all using the same algorithms to assess and price risk, with little opportunity to fine tune the logic used or to do so quickly in order to, for example, react to market trends and legislative changes, or differentiate insurance products for competitive advantage.
Beyond that, the data analytics and algorithms applied by most legacy applications are, for the most part, basic and well behind the technology curve. The majority are designed to work purely with static data and are unable to handle multiple real-time data streams, or apply advanced predictive models to better understand and forecast possible risk. Added to which the black boxes are failing to add more advanced analytical capabilities, like those available through statistical languages such as R, which is rapidly becoming the tool of choice for data scientists.
Lastly, legacy insurance applications can be difficult to integrate, particularly into a modern IT fabric spanning a mix of public and private cloud, as well as on-premise platforms. That, in turn, creates barriers when it comes to the automation and streamlining of business processes—a growing requirement as companies move towards wider digital transformation.
Benefits and Business Value
By unlocking the algorithms from the proprietary software, this opens up the possibility of gaining business advantage over competitors. Moving beyond the black box approach allows insurers to leverage human capital in the form of their data scientists to come up with new and innovative models. Providing the right tools for the job helps accomplish this goal. The Insurance Pricing Accelerator is a set of tools that allow insurers to build their own pricing and propensity models, evaluate their effectiveness, and hot deploy them into a running system that will then execute against them in real-time. Feedback from this execution can be rapidly fed back in to the modelling process and the models can be re-evaluated and updated more frequently.
The accelerator demonstrates the insight-to-action loop for insurance pricing prediction. The analyst can use the Spotfire DXP to model the propensity for customers to accept or reject an insurance offer based on a number of customer properties. Once the analysis is completed, the model can be hot deployed to StreamBase so that quote requests can be scored in real-time.
A simulator can then be run to send through a series of quote requests. Each request is scored against the model for propensity. Based on this, a commission and discount are computed based on the probability of acceptance, and a premium is generated. This information is combined together and an overall quote offer is produced and returned to the customer. The simulator then acts as the customer and either accepts or rejects the offer. All this information is stored in Live Datamart and a real-time dashboard provides summary statistics about what’s happening as quotes flow through the system.
Historical data is retained as well as the outcome of the quote. This data is fed back into a rebasing model, which refines the predictive capability and automatically updates and deploys back into the engine.
At the heart of the accelerator are the Event Manager and Analytics Server. They form the principle components of an analysis-action feedback loop. The Event Manager are implemented using TIBCO Streaming and TIBCO Enterprise Runtime for R (TERR). The Analytics Server is implemented using TIBCO Spotfire and TERR.
As quote requests flow through the Event Manager, they are scored and published to a real-time datamart implemented using Live Datamart and displayed on a real-time dashboard implemented using both LiveView Web and Spotfire.
Once scored the Event Manager returns the result back to the calling application which presents the quote to the customer who then accepts or rejects the offer. The Event Manager also retains a history of the quotes and results. On a configurable interval it invokes a rebasing model which then recalculates the predictive model based on the previous set of results. This new model is then deployed into the system.
|TIBCO Streaming Artifact Management Server||1.4.4|
|TIBCO Spotfire Desktop||10.4.0|