Leveraging AI for intelligent process automation

A prominent insurance executive remarked recently that, “the insurance industry is essentially in the business of selling promises.”

These promises all live in the form of documents; thousands of documents and related processes that capture the details of those promises.

As insurance companies contemplate their next digital transformation move, automating and augmenting the workflows related to these documents is likely high on their list.

The opportunity

Most insurance companies today are trying to capture value in one or more of the following ways:

1. Capacity expansion: Like a lot of corporate America, insurers have focused on taking cost out of their businesses for the last several decades. As the returns on those efforts shrink, we are seeing a new focus on capacity expansion. Specifically, growing revenue without growing expense by helping existing resources and team members accomplish more. It’s a huge opportunity to expand margins and grow their business.
2. Cycle-time improvements: Customer service and specifically customer response times are increasingly driving the competitive landscape. The ability to make decisions quickly regarding policy costs, approvals, appraisals, and claims response separates the high performers from the laggards. Implementing process automation is key to unlocking this advantage.
3. Efficiency: Driving bottom-line improvements requires reducing the number of hands that have to touch any given process, enabling those resources to focus on other higher value activities and initiatives. Freeing people from a lot of the mundane repetitive tasks that come with manual processes can improve job satisfaction and performance measurably.
4. Knowledge capture: One of the risks of today’s manual processes is that much of the required knowledge isn’t captured or codified anywhere. Experienced team members often carry around important knowledge in their head about a given process that leaves the building with them every night. This leads to incomplete and inefficient training and on-boarding of new team members and limited scalability of processes across teams. Digital transformation of these processes is a great forcing function to capture the specific steps and context required.

The challenge

Why have manual document workflows existed for so long? The reason is that until recently, the only practical approach to navigating unstructured content such as documents, text and images has been some variant of keyword-based approaches and enterprise search. These systems require elaborate rule-based systems to codify all possible variants of the information being processed in the form of expert systems, taxonomies, ontologies and other rule-based approaches.

Unfortunately, these systems have two glaring weaknesses:

• Brittleness: They work as long as the rules in question have completely captured the problems to solve. The moment the keywords or concepts stray beyond the target’s knowledge, these approaches break down.
• Expense: Creating these elaborate rules to interpret the enterprise content means that either the business must hire full-time information architects to build and maintain these rules, or spend millions of dollars on outside resources to do it (the “Watson” problem).

The breakthrough

The last five years have seen amazing leaps in our ability to understand unstructured content. For the first time, machines are able to understand context as well as keywords. As humans, most of our understanding when communicating with one another is from the unspoken context that surrounds our conversations. If we had to rely strictly on keywords, our conversations would be rife with misunderstandings.

New developments in the field of Artificial Intelligence (AI) are providing that context. A particular branch of AI known as Deep Learning enables machines to understand text and images in ways not before possible. Prior to Deep Learning, managing unstructured content required learning to speak to the computer in its own language, e.g., Boolean search queries. As long as we knew how to precisely form our question for the computer, it would give us back what we asked for. Deep Learning flips this on its head. Now, users simply provide the computer with examples of what they are trying to accomplish, and the Deep Learning algorithms backward solve to form the correct “model” to achieve the user’s goal.

This opens up a huge set of new opportunities to insert AI into existing processes and workflows, where the technology can “learn” a set of tasks related to the business process and in effect give the subject matter experts and business line owners “bionic arms” to dramatically improve the throughput and efficiency of the existing approaches.

Applications and use cases

There are a number of near-term applications and use cases in Insurance that can benefit from this type of AI-driven Intelligent Process Automation (“IPA”).  The common thread is anywhere a given set of documents need to be evaluated, reviewed, and/or classified, etc. by a number of different people. For example:

• Claims Processing: IPA can be used to automatically classify and annotate a new claim such that it can be effectively routed to the right SME for evaluation and processing. This results in faster turnaround time and improved accuracy for a processed claim, driving improved customer satisfaction and organizational efficiency.
• Appraisals: IPA can process both written and image-based information for property and casualty-related appraisals to verify the assets being covered. Home insurance is the best example here where photos of each room in a house as well as exterior photos can be matched to the written property description.
• Commercial Underwriting: Often involving thousands of pages of documentation, major commercial underwriting processes can be dramatically improved by creating underwriting criteria attributes that can automatically be recognized and “scored” using IPA resulting in major reduction in response times when submitting proposals.
• Policy Analysis: A common challenge in insurance is the need to be able to traverse very large collections of policies that often span several decades to understand how the language within the policies is affected by changes in regulatory policies or the competitive landscape. IPA can understand specific clauses in policies and score and classify them for a given use case.
• Regulatory Compliance: In a highly regulated industry with dozens of state and federal regulatory bodies, responding to regulatory inquiries in a timely manner represents a large expense for most insurance companies. IPA is able to create augmented responses to inquiries dramatically reducing the response times and resources required.

As insurers expand their digital transformation efforts to business processes like these, I believe Intelligent Process Automation will be a key enabler of success.

Getting to ROI

To get started down the path of Intelligent Process Automation and its associated ROI, insurers need three core building blocks:

1. An existing, defined process or workflow as a candidate for improvement, augmentation or automation.
2. Representative data, content or documents for the given process.
3. Input from the key SMEs to define and evaluate success.

With these in hand, insurers can achieve initial ROI in just 3-4 months, including development and deployment of the process.

Tom Wilde (tom@indico.io) is CEO of Boston-based Indico Data Solutions. These opinions are his own.

See also:

• 5 trends to watch in insurance AI and robotic process automation
• Using AI and automation to transform claims handling
• Tackling automation challenges in the insurance industry