How DWP brought the business on-board with RPA

Using AWS cloud, the Department has scaled its use from four robots to 1,300 in 18 months

It's a rule that, the larger the organisation, the more time-consuming its processes become. The Department for Work and Pensions, which supports around 20 million members of the public each year, is one of the largest government departments in the country, and in late 2017 was ripe for digital transformation. In recent years demand for DWP's services has risen, but staff numbers have stayed static. A solution was needed.

The Department implemented a small robotic process transformation (RPA) pilot, which since then has grown to more than 1,300 robots. Hayley Addison, Lead Product Owner in DWP's Intelligent Automation Garage (IAG), led the pilot and eventual deployment. During a roundtable session at Computing's recent IT Leaders Summit, she said:

"We have around 80,000 staff members in DWP, and a lot of those work in operational lines-of-business functions dealing with customer enquiries, processing claims, making payments, etc. A lot of the tasks that they do are high volume and repetitive: cutting and pasting information, moving information between different systems, reading information from paper forms; and a lot of it was rules-based, logic-based, and ripe for automation."

Early days

The initial proof of concept, in October 2017, ran just four robots using on-premise physical machines. The IAG made sure to talk to people throughout the Department to find the pain points where automation could help. The team also brought in people from the business so they could understand what was involved in RPA - which did involve disappointing the ones who wanted to see robots working at a keyboard.

Once the real deployment began, it was obvious that an on-premise implementation was not feasible, and DWP now runs its robots in the AWS cloud. That makes spinning them up and down to deal with specific tasks easy, and that "you're not paying for robots when you don't use them."

Scaling upward and outward

DWP deals with around 200 million calls from members of the public each year, and exchanges roughly 10 million data records every day. In addition to that, Addison described the IT estate as "very varied [and] very complex," with "some...really, really old legacy applications," as well as other parts that are "shiny and new." Integration between those systems is challenging, and that is where the Department chose to begin its RPA journey.

Since that time, the Department has scaled its use of software robots from 20 to 1,300, managing 22 different processes across six lines of business - with the expectation of adding as many as 12 more each year.

Delegates at Computing's IT Leaders Summit

Addison warned delegates - who were mostly interested in or trialling RPA, but had not implemented a full rollout yet - that none of DWP's robots had standard out-of-the-box functionality.

"If you try and apply the out-of-the-box functionality you get from different vendors - there's a handful of six or seven which are leading the way in industry at the minute - it will be very vanilla, and vanilla didn't work in DWP because everything's very complicated and complex," she said. The Department had to write a lot of custom code, and use many different libraries and open-source systems, to get the functions it wanted.

To all good things...

There is a danger with any process that what was originally intended as temporary becomes permanent. Addison said, "[DWP's RPA] was very tactical, it was very short-term, and it was to act as a stopgap between the strategic roadmap for the Department and how we try and transform and change our IT estate." She added that, because of DWP's size and scale, "short-term" could last anywhere from two to five years.

The IAG is avoiding an over-reliance on RPA by continuing to work with technical and business representatives, so that everyone involved understands that the robots are "a tactical solution." They might spin up for as little as three months for a simple task like data migration, before being turned off.

"Robotics is really good for one and done job[s] as well, because it's dead cheap and it's quick to build, but also you could have something which is a year, two years, three years, four years... [I]t's about having those regular conversations throughout the live phase of a robotics process to say, [when] reviewing your volumes, 'Has there been a decrease in volumes, and...therefore can we look to decommission the process? Where's your tipping point in terms of ROI? Where does it become no longer viable to run this as a service, and actually it would be better to revert to a human process because the volumes are now so low?'"

DWP asks these questions throughout the lifecycle of all robotic processes. "We are really keen that we don't just create a load of processes that just sit there and go on forever and ever and ever and ever," stressed Addison. She wants her team to be looking at the next steps, rather than simply keeping the lights on on tens if not hundreds of processes.

"[DWP's next step] is about using machine learning. Still using robotics in there to do the do, but it's about using machine learning and looking at how we can gather data, and help improve any decision-making across the Department; so this is where we start to talk about [business] transformation, and the longer term part of the roadmap."