RPA and AI in Claims Processing

As many of us know, the insurance sector is in a low growth phase, not helped by ever-increasing regulation and compliance and the inevitable ‘margin pressure’. But the biggest change driver in the sector, in my opinion, must be the rapidly increasing customer expectations. Customers really do want it all: 24 hour access, available through multiple channels, with a virtually instantaneous response, all without impacting their premium prices. None of which, of course, helps those margins.

Many other industries have been able to respond to their customers’ 21st century demands: Retail, banking, media, utilities even, have all transformed the way they engage with their customers digitally. But the insurance sector has lagged behind, and not, generally, through any fault of its own. There are, in my opinion, four things (plus one specific to claims) that are acting as barriers to digital transformation in the insurance industry:

  • a whole host of legacy systems, most of which cannot easily interact with each other;
  • which is mainly down to a lack of data standardisation (although ACORD is going in the right direction, it is not adopted widely enough);
  • which means a lot of data has to be manually keyed leading to unacceptable accuracy levels (one of my clients quotes that 68% of data errors are due to humans);
  • and finally, the curse of the PDF – this document format, especially the scanned variety, seems to be favoured by most people in the industry but plays havoc in trying to get data flowing efficiently between parties (in one insurance sector BPO client of ours, 80% of their processes have a pdf in them somewhere, and 40% of those are handwritten and scanned).

 

If we focus just on claims, the PDF tends to be the document-of-choice for the overly-manual submission process, but we also need to remember the inherent unpredictable demand (from natural disasters etc.) that adds in cost and complexity to the process.

RPA and AI in Claims Processing

I want to show how some of the more cutting-edge technologies can overcome these barriers within the claims environment. I’m going to talk quite a lot about ‘automation’ so it’s worth me spelling out exactly what I mean, and don’t mean, about automation.

So, every claims processing organisation, whether it’s in-house or outsourced, will have some form of claims automation. For example, at a minimum, organisations have automated systems that are used to post transactions, feed information to the General Ledger and disburse funds to claimants. Slightly more advanced would be the inclusion of automated scanning and OCR, especially if this is done at the front-end of the process. And more advanced still, organisations may have implemented automated workflow.

But, none of these really addresses the barriers I talked about earlier. Why? Because they still leave people involved in the process, either typing in data from scanned PDFs, interpreting non-standard data, collecting information from third party systems (and typing that into their own systems), or making decisions on the validity of a claim. And, because many of the ‘traditional’ automation solutions often work in silos, rather than an integrated part of a whole, it can create a situation where automation can actually lead to increased human error as data is entered and re-entered in these diverse systems.

So the process of engaging and responding to a customer quickly and accurately is a challenge for everyone – most have traditionally solved this by using either a lot of people or a lot of technology – both of which are expensive. It is these combined issues of siloed technologies and a dependence on humans which make the claims process slow, costly, inaccurate and focused on business hours only. And if you remember right at the start, we decided that customers want low cost, 24 hour, online, instantaneous and accurate engagement with their insurance agent.

So, what sort of solutions are out there that can provide that ideal customer experience without severely impacting margins?

Robotic Process Automation

The one that underpins most of the others is Robotic Process Automation. This differs from what most people refer to as automation because of that word ‘robot’ – these software agents have the capability to replace the human beings doing the transactional, rule-based work (rather than automating the workflow between the humans). The beauty of RPA is that it works at the ‘presentation layer’ of all of the other systems, so it is non-intrusive and easy to train: configuring a robot is a very similar experience to training a new (human) starter. And then, once it has been trained, the robot will do the same thing again and again and again, 24 hours a day if necessary, as well as leave a detailed audit trail of exactly what it has done.

Humans have traditionally been used as the ‘integration layer’ across and between the diverse systems, because they are flexible and very trainable. But now the robots, who are equally flexible and trainable, do it all, and for a fraction of the cost (usually between a third, for offshore FTEs, and a ninth, for onshore FTEs).

RPA therefore works very easily (and cheaply) with legacy systems, has 100% accuracy, and can be ramped up and down quickly in order to manage peaks in demand. That’s three of the five issues dealt with by one solution.

I am aware of at least 10 major insurance firms that are using RPA technologies in the UK alone. I also know of a number of other firms that are exploring RPA. It is clear that RPA is being used across the whole spectrum of the insurance sector, including carriers, brokers, re-insurance and BPO providers, and across all lines of business, including Property & Casualty, Health and Life. Some large users of RPA have now automated up to 35% of their processes.

Specific examples of processes that are being automated through RPA include:

  • First Notice of Loss
  • No Claims Discount validation (including making system updates and issuing correspondence)
  • Fraud checking
  • Policy renewal (including data gathering and recalculating the policy premiums)

 

In some cases Turn Around Times (TAT) for claims queries have been reduced by 85%, and accuracy of claims data improved from 45% to 95% (with the remaining 5% due to input errors).

Artificial Intelligence

Whereas personal lines generally have more mature systems, there is still a need for RPA – firms find it easier and quicker to implement RPA than code and configure the legacy systems. In the more complex lines, such as Commercial and Speciality, where Straight Through Processing is more difficult, RPA can play an even bigger role.

In the external claims processing market, there is fierce competition: one third-party claims processing firm has deployed 27 robots that work on 14 core processes, completing 120,000 transactions per month. They claim this saves them 30% of the cost of the process (by redeploying staff), as well as delivering improved service quality, higher accuracy, faster turnaround times, scalability, increased compliance and a better strategic positioning in the market. For the smaller niche players RPA provides a fast-track approach to growth and expansion.

Although Artificial Intelligence technologies are less mature than RPA ones, there are a number of tools that have been successfully deployed in insurance firms. The key task that these systems achieve is to turn unstructured data into structured data – in most cases this means taking those dreaded PDF documents and being able to read the information off of them and transpose that data into the system-of-record. The technology works through applying training algorithms to learn where the relevant information is likely to be on the form – they can cope with relatively wide variability and are able to continually learn as they gain more experience.

The main thing to note here is that the RPA and AI technologies complement each other very well indeed: the AI systems provide the structured data that the robots need as their inputs. Thus, one can imagine a complete process, from an email or a completed PDF form arriving from a customer right through to its satisfactory resolution, all being handled without ever touching a human being.

One UK-based insurance firm that, amongst other things, provides a claims processing service to their own insurance customers has automated the input of unstructured and semi-structured data (incoming claims, correspondence, complaints, under writers reports, cheques and all other documents relating to insurance claims) so that it goes into the right systems and queues. Using an AI solution, a team of 4 people process around 3,000 claims documents per day, 25% of which are paper.

The AI automation tool processes the scanned and electronic documents automatically, identifies claim information and other metadata, and deposits the results in SQL databases and document stores ready for processing by the claims handlers and systems (which could either be humans or software robots). It also adds service metadata so performance of the process can be measured end-to-end.

Some documents can be processed without any human intervention, and others need a glance from the human team to validate the AI’s decisions or fill in missing details. A few are illegible, and need people power to get on track.

The ability to scale on demand was key to the firm, whether that growth came from competitive success or natural disaster, and this solution provided exactly that.

Data migration is another interesting example, and one that exploits both RPA and AI. For policy migration, for example, the task is normally carried out at the time of the renewal, policy by policy, which can obviously take a long time to complete. RPA is able to manage the collation of large amounts of disparate data very quickly and accurately – it uses rules to do this but when it gets stuck there is a much more controlled process for handing off the queries to humans. The AI is able to add additional capability by being able to extract specific data fields from free-form fields, such as a Comments section.

Other Solutions Are Available

Of course, RPA and AI aren’t the solution for everything. Having better inputs into the processes in the first place, whether that be data-enriched PDFs or XML files, would make a huge difference to the efficiency of the process. Also, when the input documents are poorly hand-written so that OCR is not an option, firms can call on crowd and impact sourcing solutions to get remote humans to read the fields as micro-tasks. And there would be no need, of course, to automate manual processes if they were digital in the first place. As new ‘green field’ firms are created, they will have the luxury of being ‘born digital’, thus massively reducing the need for much of the systems integration we have been discussing.

However, the trend for insurance firms is with disaggregation of processes, and we know that the biggest opportunities and challenges will still centre around automating and implementing processes and procedures that are shared by multiple business partners, so RPA and AI will still have a place within this type of organisation. One area that RPA and AI add particular value, and especially in the more forward-thinking firms, is in analytics: the RPA solutions are very good at extracting big data, and the AI tools are great at searching out those elusive patterns. The people that these firms do bring on will have all of the heavy-lifting work already done for them, allowing them to focus on providing real value to their customers.

The Real Winner

So, we do know that up to 50% of claims are still processed manually, mainly due to those legacy systems and manual inputs (note that only a few of ACORD’s standards are actually associated with claims processing), leaving us with a sub-optimised customer experience, poor accuracy and higher than necessary costs (claims processing accounts for between 3% and 12% of premiums).

Of course different lines will have different results; life claims are generally low in complexity, whilst health claims are high volumes and high cost. P&C claims usually have longer cycle times. But, there is no doubt that many, many opportunities exist today to use RPA and AI tools to speed up the work, improve its accuracy, make it cheaper to run and, most importantly of all, improve the experience of the customer. It won’t be long before the vast majority of claims can be processed without any human interaction at all, and the real winner will be the customer.

Share on:

Leave a comment

Follow on Feedly