Next event:

Technologies for DevOps, Test Automation and AI: The QA Financial Forum Paris 2019

8 October, 2019
News and research on financial software quality assurance and risk management

Success for AXA with RPA Creates New Requirements for Test Environments

22 February 2019
The insurer's UK business is achieving savings with bots and robotic process automation. But RPA requires a different kind of testing, says AXA's Simon Clayden (pictured).

AXA has made considerable savings by introducing RPA for bots it is using primarily to check and respond to customers' insurance claims. It has automated nearly 100,000 development and testing tasks, according to Simon Clayden, COO for technology for AXA's UK business.

With robotic process automation, “the testing per se is not the complexity,” explained Clayden. “The challenge lies in ensuring that the development and test environments mirror and synchronise with each other.”

Most of the robots - sets of data-processing and decision-making algorithms –  being used by AXA, have been deployed to scan customer emails and match them to relevant claims records. They have saved AXA around £140,000 (or around 18,000 hours of work) since they were first deployed in mid-2018, said Clayden.

The first bot, dubbed "Harry", was rolled out to staff in the customer property claims team in June 2018. Harry was later joined by "Bert" in the commercial property team and "Lenny" in the liability team. AXA has now deployed 13 robots.

The majority of testing and integration work conducted on the robots took place in-house, creating a unique challenge for the AXA technology team.

Clayden explained the process of trialling these robots and highlighted a key difference between RPA and traditional software.

”For these algorithms, we don’t conduct software testing as such,” said Clayden. “The test scripts were already built-in because as soon as the robot is activated, it is following a pre-assigned logic. So we’re past the days of user acceptance testing.”

While AXA’s bots do not require user acceptance testing as such (because the software process is fully automating a back office process) the fact that they operate in multiple environments, coupled with their low tolerance for infrastructural differences, means that they demand a much higher degree of similarity.

“We do put the robots on an initial probationary period because we’ve found that our main consideration was how to tackle the test environment.” Clayden said. “We have separate test and production environments. We build the robots in a test environment and we found with our first iteration that we had to construct our test environments as an exact mirror of the development and production environments.”

And this test environment requirement is a particular quirk of automated algorithms, Clayden said. “This was never a consideration in the old days when we had people doing user acceptance testing. If the tester saw a text field in a slightly different place, for example, or a piece of data that didn’t quite match the test script, they would be able to work that out and find their way around it. The robot, on the other hand, stopped as soon as it saw a field where it wasn’t expecting it.”

This is a factor that will increasingly come up in testing, as financial institutions explore the potential of AI, machine learning and robotic process automation, Clayden believes.

“Historically, you would have a test environment for a particular application and a separate production environment. What we’re starting to see with robotics is that they will be interacting across multiple applications throughout the process that you’re automating. The initial hurdle is just ensuring that, as the robot moves between multiple test environments and then multiple production environments, it doesn’t stop running.”

All of AXA’s robots, based on RPA software supplied by UIPath, the New York-based vendor of RPA technology, are cases of what is known as 'attended' RPA, whereby the bot helps employees get a task done, rather than taking it completely off their hands. In the case of reading and categorising correspondence, which would take a human on average of four minutes per task in AXA's case, it takes a bot 42 seconds.

AXA has successfully tackled the RPA integration challenge, but its ambitions stretch further. The business hopes to have 100 robots by the end of the year and will eventually evolve their customer communication function to incorporate chatbots.

“This is very much the beginning of our journey,” Clayden said. “We intentionally made the choice to look at customer-facing processes first. We do have ideas around the use of robotics for IT, in areas such as testing, however, we prioritised the customer-facing applications for our launch. ”

Clayden acknowledged that it is still too early to make a meaningful comment on the applications of artificial intelligence for software development and testing at AXA.

“But test automation is certainly an area within IT that we’re very keen to explore further. The core will be the robotics from UI Path, but on top of that, we’re looking at chatbots as the interaction channel. Additionally, we’re looking at voice biometric identification – both for internal IT applications and for our customer service channels,” Clayden said.

Get the latest
by email

Newsletter Sign-Up
First Name*
Last Name*

I understand my contact details will be entered into your database and used to contact me by email with QA Financial's newsletter. See our privacy policy here.

Opted-in to receive newsletter
Source - newsletter sign-up form