High-Performance Business Decision Automation at DecisionCAMP 2018

Dr. Jan Purchase and I will jointly present a session at DecisionCAMP on 17 September 2018.

High-Performance Decision Model Execution by Compilation of DMN into Machine Code

• Drivers for high-performance decision making.
• How to achieve high-performance decision execution.
• Examples and demonstration of high-performance.

Drivers for High-Performance Decision Making

As part of their digital transformation, businesses need to make ever more complex business decisions in increasing volumes and act on them in real-time:

• What mortgage offer to make to a specific client, based on their financial situation?
• What products and services will keep this client loyal?
• Which of these clients has the highest lifetime value and should be prioritized?

These are all high-value, high-volume decisions that need to be made methodically, transparently, automatically and increasingly frequently.

Such automated decision-making requires a fast connection between business insights—often acquired through analytics or machine learning—and action empowered by business rules. Speed is essential, after all: “the customer is waiting”.

As a recent Forrester report noted:“decision management is the highest-value next step for firms that wish to complete the insight-to-action cycle necessary for successful digital transformation.”

But the simultaneous growth in the internet of things, edge computing, and regulatory and compliance transparency requirements is set to simultaneously increase the data volumes required by decisions and constrain the hardware available to make them.

Dr. Jan Purchase and I will describe how the three main drivers for the geometric increase in volume and complexity of decision making are:

• The data stream from wearable and other proliferate devices (IoT).
• The evolution of markets and applications (especially in finance and medicine) towards fine-grain, AI-informed decision making.
• The requirement for increased transparency when processing transactions, rather than traditional aggregate processing, for financial and data protection compliance.

What’s perhaps less well known is that edge computing will impose hardware and connectivity constraints on decision-making environments. If you’re diagnosing disease by performing retinal scans on children in the Sudan, you won’t have a reliable connection to upload your findings to the cloud for processing—you may only have a three-year old mobile phone. So you’ll need an edge architecture and a highly performant decision-making system with a small hardware footprint. The same is true if you’re trying to detect an intruder in a network of in-body, organ monitoring devices.

“The front line of intelligent decision-making will be thousands of small, low-spec devices collaborating in a local, self-healing network—not a multi-node AWS server.”

How to Achieve High-Performance Decision Execution

We will specifically examine procedures used to execute DMN decision-models (DMN is an OMG standard for decision modeling that has been adopted worldwide) at very-high speed.

Delegates will learn about some of the techniques we used to execute DMN in machine code to align well with the architecture of common CPUs, overcoming the performance bottlenecks of earlier implementations.

And we will discuss the features of DMN that present the most significant challenges for high-performance environments. We will show how some of these challenges can be overcome by best practices, and discuss some proposed revisions to DMN’s type system to improve performance with no practical impact to its flexibility.

Live Demonstration

IFRS-17 is one of the most demanding regulatory compliance regimes of recent years. It is a new insurance regulation that governs how and when insurance companies are allowed to recognise profit, obliging them to account for profit and loss incrementally as an insurance product provides its value, rather than when premium payments are made.

We will show a live demonstration of DMN execution, based on one way of implementing IFRS-17. Starting from a DMN model produced by one of the fifteen or so modelling tools on the market—and execute that model locally on a regular laptop at speeds between 100,000 and 1,000,000 decisions per second.

We will discuss factors that can a significant impact on speed, such as using floating-point numbers, logging, number of cores—and potential trade-offs between precisely following standards and decision speed.

Are You Attending DecisionCAMP 2018?

We are looking forward to meeting, and re-meeting, decision management professionals and clients at the conference.

Please let us know if you’re attending and we can arrange a private meeting for you.

info@rapidgen.com