Using modern machine learning (ML) techniques, banks can know with more certainty whether a loan applicant will make their repayments, governments can decide whether a benefit application appears fraudulent, and health insurers can recommend which therapies are most likely to lead to long term positive health outcomes. So where does this leave Oracle Intelligent Advisor, which is used to make policy-based decisions in all these industries?
As I outlined in Has AI made Intelligent Advisor obsolete? Nope!, artificial intelligence and machine learning are complementary to Intelligent Advisor in several ways. In this post I give examples of how machine learning models can be used in combination with rules managed directly by humans, to deliver the best possible advice to customers.
Let me first give an outline of what these experiences look like:
- A customer is interacting with your organization, to get advice or receive a decision. Will you give me a loan? Which benefit should I add to my health cover? It is worth claiming my minor car accident repair costs or will my premiums go up by more?
- You probably already know some specific things about this customer. Their birth date, model of car they own, how many children they have. But you also probably have gaps. What course of study they want to apply for or whether their health or employment situation has changed recently, for example.
- To provide the right advice, you ask questions to fill in gaps, and confirm what you already know is still correct. Ideally, how many questions you ask depends on their personal situation, and what information actually is needed to reach a decision. If their insurance loyalty status will prevent their premium going up at all, then you need not ask questions to estimate how much the repair cost will be, for example.
- Once you have enough information, you provide an answer, ideally providing the customer with a transparent explanation of how it was reached. Usually, you will save this decision, and will often generate printable documents for the customer to download and refer to later.
- In many industries you will have to show regulators or other auditors that the decision you reached was fair and appropriate to the customer’s specific situation.
Intelligent Advisor is designed to deliver these experiences. It lets you capture detailed rules for making decisions, and there are many proven applications where Intelligent Advisor can provide all you need to give your customers the right and best advice. But if you have an ML model you think will help deliver better service, you can use Intelligent Advisor and that ML model together.
Where machine learning comes in
Building a machine learning model requires both data science and domain expertise: training data must be prepared, and appropriate learning algorithms must be applied. Once created, however, machine learning models often outperform human-created models. For example, many micro lenders are building portfolios of well-performing loans by applying machine learning techniques to assess repayment risk. Earlier risk models used simpler rules that humans could design directly, and can’t cope with applicants that have no credit history.
The machine-learned model is then embedded into an experience like the one outlined above: the user indicates they want to apply for a loan, and you collect from them the information needed to make your approval decision. By using Intelligent Advisor, you can improve the customer’s experience by avoiding asking for information that you don’t need, and by the end have a detailed explanation of how you reached your decision for whether to approve the loan or not.
More examples
There are many applications where ML models could be combined with Intelligent Advisor in this way, such as:
- Car warranty. An ML marketing model returns the likelihood of a customer being ready to buy a new car. Intelligent Advisor rules that determine warranty payment eligibility use this model to help choose which customers should be paid for a repair, even if the cost is actually not covered by the customer’s warranty.
- Equipment leasing. An ML credit risk model assesses the risk of customers making their payments. This is combined with an Intelligent Advisor application that auto-approves very low risk and auto-denies high risk leasing applications, and has rules for how to decide and which questions to ask when the risk is in the middle.
- HR benefits. An ML benefit matching model, that determines which of several company-provided health benefits options is likely to provide the most benefit to the employee. For an employee to complete their benefit selections through the self-service HR portal, an Intelligent Advisor interview shows the plans they are eligible for, then uses the ML model to help them pick the program that is best for them.
In each case, Intelligent Advisor provides the interactive experience and contains rule logic that uses the score from the ML model – when it is needed – to help reach the best decision. Intelligent Advisor provides the collected data to both the Intelligent Advisor and ML engines, and generates the final result and explanation of how the overall decisions were reached.
Better together
So, how exactly could machine learning models and Intelligent Advisor work together to deliver these experiences?
Once you have an ML model (or models), you would use Intelligent Advisor to:
- Build logic that uses the recommendation from the ML model to reach a final decision. For example, if ML risk rating is below 0.05, then automatically approve.
- Based on the Intelligent Advisor rules and ML model inputs, design the questions the customer needs to be asked
- Design the preferred order in which questions should be asked, and when certain questions can be skipped
- During each customer interaction, actually ask the questions and verify/update things you already know
- Pass the collected data to the ML model – possibly in multiple iterations
- Combine the ML model’s advice with the Intelligent Advisor rules to reach a final decision, and any related calculations
- Combine the explanation from the ML model into Intelligent Advisor’s explanation of how the decision was reached
- Save the result, and generate the final documents
Not all ML models can provide explanations, but this is an area of ongoing improvement. A key area to consider is how to incorporate those explanations into the documents and audit trails that arise from the decisions and recommendations that are given. The approach you take will depend on what AI explanation technique you adopt, and what style of explanations you need to provide to your customers, regulators and other stakeholders.
Summary
The industry is still in the early stages of incorporating machine learning into enterprise software applications. As shown in the examples above, there are many opportunities to use AI to deliver better service. Organizations at the leading edge will find ways to use AI to do so, leading to higher customer satisfaction and better returns.
By combining Intelligent Advisor with ML, you can have the best of both worlds:
- Transparent, easy to define decision making that complies with regulations and policies
- Cutting edge decision models that leverage learned patterns to deliver the best outcomes
If you are interested in exploring how Intelligent Advisor could work with ML in your organization, please leave a comment, or reach out to me directly. This is an area that we are seeing evolve rapidly.
Davin Fifield
VP Product Development, CX Service Intelligent Automation
See also:
Has AI made Intelligent Advisor obsolete? Nope!
Deep and meaningful. How far do your chatbots go?
For More Information
Oracle Artificial Intelligence