X

Smart Advice. Agile. Personalized. Transparent.

Has AI made OPA obsolete? Nope!

Artificial Intelligence is everywhere these days. Many Oracle customers are exploring AI, and some of them ask me “If I have AI do I still need Oracle Policy Automation?” The short answer is that OPA and AI both provide enormous value to any organization, but are very different and, in some cases, complementary.

Simply put:

  • AI learns from data to create models that can decide the best action to take. AI is not about hard and fast rules. It is about recognizing and classifying patterns of input into particular types of response.  
  • OPA makes predictable decisions based on clearly defined rules it is provided with. It will never change its decision and can always explain how each decision was reached.

Let’s look at a couple of examples.

If you ask a Toyota salesperson whether leather seat protectors or insurance are most likely to be purchased by someone buying a Prius, you’ll get an answer based on their own experience. An AI system can be trained using data from Toyota dealers all around the country, and can result in much more consistent upsell success. That AI system can also be updated in real time as more data comes in. While OPA can be used to define upsell rules, if you want those rules to change, you will need to change them in OPA. OPA won’t automatically learn which product combinations are most likely to lead to more sales. OPA can, however, tell you exactly what sales bonus the salesperson should get, and what insurance products the customer is eligible for, and why.

Or consider HR policies your company has around study leave. You can use OPA to describe those rules and provide a self-service advice tool to all your staff to understand exactly how and why the policies apply to them. An AI system would need to be trained based on the advice given to employees in different situations, and would then need to be retrained whenever the policies change. It would only be as good as the data provided to it: if the training data contained incorrect or incomplete advice, the system would too.

Choosing the Right Tool

AI requires data science expertise to identify the data and machine learning techniques needed to meet a particular business challenge. To simplify things for our customers, Oracle provides Adaptive Intelligent Applications. These are pre-built for specific business needs such as responding to supply chain demand, finding best fit job candidates, and determining next best sales offers. They don’t require customers to have data science experts to create and maintain them. As the name suggests, they do a great job of using lots of data to provide high quality, adaptive decision making for specific scenarios within Oracle’s SaaS applications.

By contrast, OPA is designed to be easy for non-technical business users to apply to any domain. Perhaps you need to make sure your field inspectors always follow particular steps when conducting inspections or repairs of a particular type of medical equipment.  Describe those rules in OPA’s natural-language rule format, then provide an app to your technicians that walk them through exactly the required steps every time.

Examples of other questions that are a better fit for OPA than for AI include:

  • If I am the sole wage earner in a family with 3 children, and one of them is deaf, how much money can I earn before I will no longer be eligible for the family tax benefit payment?
  • A customer in Ireland is opening a new multi-currency bank account. What due diligence actions must I perform to comply with US and International anti-money-laundering laws?
  • The Bluetooth connection in a customer’s car has stopped working, but it’s been out of warranty for 6 months. Will we still pay something towards the repair cost?
  • How much tax must I pay on my salary, with deductions?

These are black and white questions. For reasons of fairness and compliance, they require consistent and accurate decisions that don’t change over time unless there is a specific need to do so. OPA is really good at helping you answer them quickly and easily in your customer service and inward-facing business processes.

Transparency

OPA makes it easy for business analysts, lawyers, policy experts and any other subject matter experts to review, understand and update decision making rules. The end result is a transparent system that can comprehensively explain every decision that it makes, based on the inputs it was provided with. For these fixed decisions, this is crucial for auditing and compliance purposes.

By contrast, the internal workings of an AI algorithm are generally opaque. Like the human brain on which they are modelled, machine learning systems can’t easily provide a human-readable explanation for why it reached the decisions it did. A particular brand of lipstick may be recommended for a shopper at checkout based on what is in their cart, but is it also because of who is in their social network, their age, or the time of year? It is likely a complex combination of all these things and more.

Summary

The following table summarizes the key differences between AI and OPA.

 

Artificial

Intelligence

Oracle Policy

Automation

Learns from data

Yes

No

CAN Adapt in Realtime

Yes

No

Expertise Needed to Apply to New Domains

Data scientists and Domain experts

Business policy experts only

Pre-defined Solutions for Specific domains

Yes

No

Transparent Logic

No

Yes

Always Consistent

No

Yes

Explains EVERY Decision

No

Yes


What About Using OPA and AI Together?

Hopefully, this has helped explain the key differences between AI and OPA, and why they are both useful to your business.

In the next blog post in this series, we will talk more about some of the ways in which OPA and AI can complement one another. To whet your appetite, here are a few ideas:

  • Using machine learning models to detect unusual data entry patterns in OPA interviews
  • Embedding OPA decision making within an NLP-based virtual assistant
  • Using AI to optimize the policies that OPA is enforcing

Until next time.

Davin Fifield

VP Oracle Policy Automation Product Development

 

Next Post: Deep and Meaningful. How far do you chat(-bot)s go? 

 

For More Information:

Oracle Adaptive Intelligent Apps

Oracle Policy Automation

Join the discussion

Comments ( 1 )
  • Clement Joseph Thursday, October 4, 2018
    Good and relevant post.
    Machine Learning and AI can complement existing systems to improve decision making in some cases.
    Example: Diagnosis of diseases.

    In policy modelling case, I agree if these two systems work independently on different problems (as rightly stated) to achieve best results.
    Example: Recommendation Systems -- AI / ML might have an edge over Policy Based Systems.
Please enter your name.Please provide a valid email address.Please enter a comment.CAPTCHA challenge response provided was incorrect. Please try again.