Insurer Unum aims to deliver benefits of $20m using AI to re-write old Cobol code

Analysing 1.5 million lines of code

Last year, Unum partnered with Amazon Web Services (AWS) and Pegasystems, which provides a low-code platform to Fortune 500 companies to automate their business processes, in a project to re-engineer Unum’s Cobol computer code. The insurance company is using AWS’ AI-powered Transform platform to translate its old code into modern business workflows which run on Pega’s cloud platform.

Unum used Transform to ingest and analyse 1.5 million lines of Cobol code in a matter of hours. The platform mapped where the Cobol software integrated with other IT systems and was able to document the IT systems and processes it depended on.

The process is analogous to renovating an old house where the original building plans have long since disappeared and the architects are long gone. There are hidden wires and pipework in the walls often with hidden connections and unexpected links that need to be discovered before the building can be updated.

As Guatam Roy, senior vice-president and chief technology officer (CTO) at Unum told a conference last month, software’s hidden pipe problem is similar – it’s not clear what processes depend on what other processes, what connections are live and what happens if something breaks – a key question in a regulated industry.

Unum, howeverm says it has been able to use AWS Transform to uncover these hidden relationships and interdependencies in its mainframe code.

Reimagining the business

The aim of the work is not simply to replicate the Cobol code in a modern computer language, but to “reimagine” how the company conducts its business, says Anderson.

The insurance group is modernising its business processes project by project, she says. Business leaders choose projects that meet the immediate priorities of the company and help with the long-term goal of moving away from old technology.

“We are actively looking at our application suite. In some cases we may say let them run for now because they are good enough. But for others that are going to be more differentiating for the company, we will start looking at having more active modernisation plans,” she says.

Each Cobol workstream has a dedicated work team. They include dedicated project leaders, technology and data analysts, a project stream owner and an architect. Each project aims to deliver results in 30 to 90 days, says Anderson.

Claims processing

The company started by re-working its claims processes. Anderson’s goal is to have fewer manual processes and to make faster decisions on customers’ insurance claims. That meant simplifying the work processes used by Unum staff when they deal with incoming insurance claims.

“We actually have multiple windows, multiple touch points and many manual steps, all of which were preventing us from delivering the customer experience that our customers deserved,” she says.

 “It’s not just the one-time investment ... you really need to look at the total cost of ownership and how much that’s going to solve for your business”

Shelia Anderson, Anum

Unum has documented and analysed its claims logic and re-designed multiple workflows in Pega’s Blueprint software, and has begun moving a number of workflows into production. Blueprint is intuitive enough for business specialists without a background in IT to understand.

So far, the company has reduced the number of screens that its claims staff have to view, including old fashioned green screens, from seven to just one. By simplifying the process, staff will be free to focus on higher value tasks, allowing them to offer a more personalised service to customers.

Jeremiah Moody, assistant vice-president at Unum for Pega Platform Engineering, told Computer Weekly that the work is likely to save 7,000 man hours a year by simplifying how claims managers hand over insurance claims to claims specialists. Some parts of the business are also buying-in AI capabilities or off-the-shelf software, for example, from startups, when they need to get capabilities up and running quickly.

“We have a number of those that are yielding results already as well as in our claims processing,” says Anderson.

Data is a priority

Anderson says that data is now a priority area for the company and that teams are looking at different categories of data, such as customer data, to find ways to simplify business processes.

“Usually, you will find many different entry points and change points through your manual processes and systems that will create quality challenges in your data,” she says.

Unum is simplifying the processes by reducing the number of manual data entry points, where people can make changes that could affect the quality of data, while at the same time cleaning up the data.

Testing is vital, says Anderson: “A lot will depend on the readiness of your data layer, so that the more time you have put into prepping the data layer in the beginning, the faster your testing will be.”

Understanding the costs

One of Anderson’s focuses is to strike the right balance between adding new capabilities and retiring old technologies that will “offset” the costs of new technology investments.

“We are at the point now where we’ve got the information and that data to make better decisions going forward. Into this year, we will be taking a much more informed approach around where you stop, start or continue investments,” she says.

With any project, its important for IT leaders to understand the real costs of technology over time, not just the upfront costs.

“It’s not just the one-time investment, but as you are making a decision, you really need to look at the total cost of ownership and how much that’s going to solve for your business, the business value and how much it’s going to cost you to run and how you compare that against the ongoing business benefit,” says Anderson.

That is going to be particularly important following rises in the costs of AI tokens which are making the use of AI development more expensive. Although AI is making it possible for business leaders without any programming experience to build applications, there needs to be processes in place to reign that in, she says

Unum will be drastically different in the future

Anderson says that the company’s customers and employees’ experience will be “drastically different” in the future. Rather than having customer agents manually populate claim forms, that information will be automatically added from data the company holds.

“Think of a future where you know your customer and, based on the customer information you already have, [that will give you] most of the information that you need to file that claim,” says Anderson.

That could include data already held by Unum and data from third-party data services, such as Lexis Nexis and other providers that could be combined into a data lake.

When a customer calls, the claims staff will be able to understand what products each customer has and know why they are calling.

“We can even look at industry data to understand significant moments in the life of a policyholder. So, if you could tell that they visited an emergency room, for example, you could then tell which policy they were likely going to be claiming against,” adds Anderson.

Automating and simplifying business processes in this way will remove manual steps that can lead to errors in data. Ultimately, it should be possible to adjudicate claims decisions using rules-based decisions, with little or no manual intervention.

“That’s ultimately going to result in high customer satisfaction, and with those streamlined processes, your employee’s are going to be focused on more of the high-end services that really matter,” she says.

Unum, however, will always have a human involved in the decision-making process, particularly if an insurance claim is going to be denied.

“That is a big operating principle that we have as a company to make sure that we are serving our customers with the empathy and support that is needed, because, many times, if we’re interacting with a customer and they are on the phone, they are calling because of a really tough situation,” she says.

Future opportunities could include using AI to summarise claims, giving intelligent coaching to insurance staff and using AI for fraud detection.

Anderson says that the work to transform mainframe applications is still at an early stage, but ultimately the plan is to start deprecating old IT systems, adding: “To be able to achieve the value, you have got to be able to shut some of those things off, fully decommissioned.

“Where most companies end up is implementing a log of great new things but, for whatever reason, you can end up with a small lingering tail of capability and cost that is staying on the mainframe. We have got to get to the point where we’re making hard calls to push that all the way through to completion.”

Learning from mistakes

Anderson is a great believer in learning from mistakes – for example, a team is working on an AI capability for the company in Unum’s labs, which she says is “a brilliant, game-changing idea for us, yet they started out with the hardest set of use cases, the very most complex”.

She adds: “But we can actually get benefit for the 80% of use cases that are much simpler and it might have been better to start there. That is not a failure, it’s a learning.”

She advises other organisations that are thinking about modernising their legacy code to not wait for the perfect moment. It’s more important to get started and to learn and test. “It’s a bit like building that plane while you are flying it,” she concludes.