Artificial Intelligence (AI) and Robotic Process Automation (RPA) are often misconstrued as the same, but don’t be fooled. They are in fact different technologies. Depending on the business case being addressed, one might be a better fit than the other, and it might even make sense for them to co-exist. Each has its own merits and can be applied to solve a variety of business cases throughout the insurance value chain.
Broadly speaking, AI is the ability of machines to interpret data and act intelligently in order to predict, make decisions and complete tasks based on the data at hand – like a human, only faster. Both AI and RPA offer intelligent automation for commercial insurance companies, allowing organizations to automate repetitive tasks and streamline once time-consuming manual processes. RPA is often lumped into the same broad category of AI capabilities, but there are some important differences.
Get our viewpoints delivered to you inbox
So, what is the difference between AI and RPA?
AI is data-driven, so although RPA and AI can both be used to intelligently automate repetitive manual tasks, only AI continues to learn and discover through data. Data is the key ingredient – the more data you feed the model, the more accurate it gets over time. Each time a human knowledge worker corrects a mistake, this information is recorded and used in the next training cycle. This is often referred to as “a human in the loop” and is part of the AI continuous learning cycle that ensures the system gets smarter and more accurate as it goes along. AI systems can capture feedback, learn, iterate, and improve over time, whereas RPA is business rules driven. This means that it relies on a set of instructions and is programmed to perform basic repetitive rules-based tasks to enable process automation. RPA tends to be used for very tactical needs. With RPA, software bots complete tasks according to the rules specified and, should there be a change in a process or the way the system needs to perform, reprogramming is required to update the decision tree that provides instructions to the bot. Unlike AI, RPA does not continuous learn and get smarter on its own.
With AI it has the inherent ability to “continuously learn” and improve the accuracy of its output over time and with exposure to more data. This is not the case with RPA, as it is reliant on the ongoing maintenance of business rules or a decision tree approach like “if this, then that.” Once set up, RPA is typically good at taking a specific action at a specific point within a process. If the business logic changes, then the rules need to be updated to address the new logic. Unlike AI, RPA does not automatically improve processes over time, and it doesn’t adapt to change without human intervention, setup, and recoding.
AI and RPA can both enable human knowledge workers to delegate high-volume, routine tasks to digital co-workers or software robots, however, the difference is that AI does not require reprogramming and can stand on its own. Because RPA is automating a specific portion of an existing process, it typically requires integration with existing IT infrastructures and legacy systems which can lead to additional maintenance costs. However, SaaS-based AI solutions don’t require integration with legacy systems to provide value. AI can, of course, be integrated with RPA solutions, broker management, agency management, rating engines, and core systems to share data and auto-populate data fields within these systems, but it is not a prerequisite for the AI system to deliver value.
Types of AI
There are many types of AI, and not all AI is created equal. Machine Learning (ML) is a core subset of artificial intelligence. It is an academic branch where a lot of research is being done today. There are many different types of machine learning such as supervised, unsupervised reinforcement learning, semi-supervised and deep learning. Deep learning is highly capable of recognizing a pattern and classifying them.
Another subset of AI is Natural Language Processing (NLP), which is a machine’s ability to read language as humans do, pulling out relevant pieces of information, assigning value to those words, and intelligently analyzing structured and unstructured text.
Semantic analysis reads all the words to capture the true meaning of any text. It relates syntactic structures, from phrases, clauses, sentences, and paragraphs to the level of the writing to their language-independent meanings. Basically, it analyzes context in the surrounding text, and it analyzes the text structure to accurately understand the relationships to the words.
Applications like Siri, Amazon Echo or Google Earth, are great examples of question and answer systems, speech-to-text, and text-to-speech. For instance, if you ask one of these apps, “What’s the weather today?”, it will give you an answer. Another example of NLP is content extraction and classification which is the ability to read all the entities in a digital document. An entity is basically a noun – like a person, place, or thing. With NLP, you can extract data from text, tables, and charts, and associate the relationships between entities. For example, the NLP solution contextually understands the relationship between the insured, the insurer, type of insurance, etc. in a policy or binder.
AI solutions based on NLP and ML, designed for insurance, and trained on insurance data models can extract, interpret, and contextually understand data locked away in unstructured insurance documents. For example, the system is smart enough to contextually understand the difference between “orange the fruit” and “Orange, a county in California”. A well-trained NLP solution can recognize and classify natural numbers in a policy as policyholder limits. Named Entity Recognition (NER) is a specific type of natural language processing that is very powerful and applicable to the insurance industry. Compared to horizontal solutions, that can only recognize between 10 to 25 entities or nouns, AI solutions purpose-built for insurance can recognize more than 500 entities. Being able to extract and contextually understand hundreds of entities like limit, deductible, premium, monetary figures, dates, names, etc., convert the data into XML or JSON format and automatically feed it into downstream streams drives efficiency and costs savings.
How AI Works
To some, AI is an abstract concept, so let’s take a closer look at the AI Learning Effect and how it works in three basic steps.
1. Label the data
First, you have to train the data model which requires collecting or gathering together data from various documents – policies, binders, endorsement, applications, statements of value, loss run reports, emails, vehicle schedules, etc. for a specific line of business – and then cataloguing and labelling the data. The labels identify key data points such as insured’s name, address, the broker, the insurer, coverage, limits of liability, etc., in paragraphs, sentences, lists, tables, etc., and will be used to teach the machine.
2. Train the model
Once the data is labeled, then the data model is developed to train the machine to identify and extract the key data points. For example, you may want to extract a list of all the named insured in document or a list of all the limits, the system will automatically recognize the entities within the document, extract, interpret and classify them based on the trained data model regardless of the location of the data in the document.
During this phase, tokenization and stemming occurs, which is the process of segmenting text into words, clauses, or sentences – words are separated out and punctuation is removed. For instance, stop words such as “the”, “of”, and other commonly used words are removed as they are unlikely to be useful for machine learning. Behind the scenes, there are several statistical methods such as TFIDF, scoring of words, sorting algorithms, word economics, etc., that take place to train the machine to understand the contextual relationship. For example, in any given insurance document, there are three key parties: 1) the policyholder or insured, 2) the insurer, and 3) the broker or producer. With machine learning, natural language processing, and named entity recognition, these entities can be automatically extracted from an insurance document, and the system can contextually understand their relationship to one another just like a human knowledge worker.
3. Understand and Predict
The final step in this basic three-step process is the “understanding” phase, where the system makes a prediction based on the modeling. Now, the initial predictions are not always perfect – typically they are about as good as your average human on their best day. However, if it does not perform to that standard right out of the gate, you have the ability to correct it which is part of the continuous learning loop. With AI, there is always a human in the loop that is feeding human intelligence to the system, telling it when it is correct and when it isn’t, so it gets better over time. The more data sets the system consumes, the better it becomes at recognizing, interpreting, and classifying the data. This is called reinforcement learning.
By labelling the data, the machines learn from people and develop ways of providing human-like results. Every bit of data gives the system a better understanding of how humans read, classify, contextualize, and understand information. The quality of data that is provided to the machine through human input is critical to improving the accuracy over time.
For more technical insights into how AI works, listen to the FNO: InsureTech podcast as hosts Rob and Lee interview Ron Glozman, CEO and Founder, Chisel AI.
The short answer is yes. Whether you deploy AI and/or RPA depends on the business case that needs to be solved. Both are different tools that solve different problems, and sometimes you may need to use them in unison to get the job done. You may even consider integrating your AI system with your RPA system, process mining, analytics, and other tools to achieve hyper-automation for greater efficiencies and business agility.
The enterprise RPA market is growing at a CAGR of 65%, from nascent in 2016 to $3 billion in 2021. Likely higher. By 2021, Forrester estimates there will be more than 4 million robots doing office and administrative work as well as sales and related tasks.
Gartner has predicted that by 2021 AI augmentation will generate $2.9 trillion in business value and recover 6.2 billion hours of productivity. For insurance companies looking to increase revenue and customer-centricity, embracing AI-powered solutions will enable them to do more without adding staff.
AI in Commercial Insurance Underwriting
Insurance is one of the most data rich industries – with more than 80% of the data unstructured. The insurance industry experiences losses of more than a billion dollars a year, stemming from errors and omissions. With an AI solution purpose-built for commercial insurance, carriers and brokers can intelligently extract data from unstructured insurance documents and contextually understand the data like a human – only at superhuman speeds and with greater accuracy. This data can be extracted from applications, submissions, binders, policies, statements of value, loss run reports, etc., eliminating the need to rekey data into rating engines and other core insurance systems. By unleashing data and giving it to underwriters, they can make better risk assessment and pricing decisions.
Today, brokers and carriers employ teams of human knowledge workers or outsource the time-consuming task of manually reviewing insurance policies to safeguard against potential errors and omissions prior to policy issuance. Armed with red pens, yellow highlighters, and paper copies of policies, knowledge workers spend hours reading through policies and toggling across screens to verify the policy language, coverages, limits, insured addresses, endorsements, etc., to ensure that all facets of the policy are accurate. These policies can be hundreds of pages in length, which means it can take days to compare a policy against a binder. Often, insurance organizations can only spot check 50% or less of the policies due to the lack of manpower and the amount of time it takes to review policies and related documents that are many hundreds of pages in length. Not checking policy leaves the broker vulnerable to potential costly E&O risk exposure.
By leveraging AI, insurance carriers and brokers can automate and accelerate the policy checking process with real-time data extraction, on-screen checking against a virtual checklist, and color-coded error detection which enables them to quickly identify and correct errors and/or omissions.
“A traditional insurance broker’s day is made up of 70% admin and 20% business sales, with only 10% left for providing expertise to clients.” – Paul Donnelly, Executive vice president for EMEA at Munich Re Automation Solutions – Insurance Business
Industry analysts are predicting that online insurance sales will continue to increase, amplifying the need for insurance carriers and brokers to reimagine how they do business today. Automating high touch elements of the insurance value chain such as application intake, policy checking for brokers and policy checking for carriers will be a necessity to accelerate the quote-to-bind process, invoke straight through processing, and capture new revenue opportunities.
“Novarica predicts that the market size for direct online sales of small business insurance will increase to $3.7 billion by end of 2020 and could grow to $12 billion by 2025.”
Manual underwriting processes lack flexibility and are difficult to scale. With AI, the process of policy checking and application intake can be automated and streamlined, saving time, reducing operational expense, and allowing skilled knowledge workers to spend more time nurturing customer relationships.
Summing It Up
In short, RPA is process-centric and rules-based, whereas AI is data-driven and designed to grow more intelligent over time. Depending on the use case, SaaS-based AI solutions for insurance deliver significant benefits for commercial insurance companies:
- Increase data extraction processing speeds and eliminate the need to spend two to four hours per day rekeying data into rating engines and other core systems
- Digitize the checking of policies to reduce the risk of errors and omissions
- Automate the application intake process, expanding underwriting capacity by 50%
- Reduce operating expenses and increase customer response times
- Provide the flexibility and agility to easily scale across lines of business and geographies
For more insights on the power of AI in commercial insurance, check out these educational resources:
Read the Carrier Management article “AI Accuracy and the Pursuit of Perfection”
Review the AI Learning Effect with Colin Toal, CTO, Chisel AI
- Listen to AI Wisdom Ep. 6: The Crucial Role of Data in Your AI Strategy with Colin Toal