robotic automation in government – Government & civil service news

Illustration by Katy Smith

Taking on routine, high-volume tasks, Robotic Process Automation can create huge time savings for both public servants and citizens. At a GGF webinar, experts from Finland, Romania, the USA and a private provider explored the potential – and the risks – of this emerging technology. Event chair Elaine Knutt reports

Robotic Process Automation (RPA) has huge potential in government – particularly as public servants work to support social distancing and deliver new services during the pandemic.

Widely adopted in the banking, insurance and retail sectors, RPA software tools can speed up routine, admin-heavy processes by following pre-determined rules. Any public sector organisation interfacing with citizens and service users has IT systems geared to predictable, repeatable tasks – and entrusting them to 24/7 “bots” could improve speed and reliability. And the COVID-19 pandemic has bolted an ever-expanding number of new routine processes on to organisations’ core missions. Whether it’s grant or loan applications, or tools that directly support the coronavirus response in healthcare, new systems have been rolled out at astonishing speed.

Panellists on a recent Global Government Forum webinar agreed that COVID-19 has accelerated interest in RPA in the public sector – noting how quickly it can be mobilised, the results it can bring, and the range of use cases. But if the panellists agreed on the technology’s potential, they also agree that public sector organisations must tread carefully. Poor oversight could mean that RPA tools end up automating errors or replicating biases; poor data architecture could restrict their availability to exploit the technology.

97% time saving in Romania

In a compelling case study of what RPA can achieve, Lacrimiora Corches, a general director at Romania’s Ministry of Labour, Family and Social Protection, described how Romania’s National Agency for Social Payments was tasked with distributing direct payments to self-employed workers whose livelihood had been affected by Covid-19, while delivering on a government pledge of a 10-day turnaround for those deemed eligible.

To achieve this, it moved from first discussions with an RPA vendor on 30 March to full national roll-out just a month later. “On 15^th April, we were able to start testing; after 12 days, we scaled for the whole country. In one month, we went from moment zero to the moment when everyone in the national agency was able to put in place the automated process,” she says. The metrics were impressive: 96% of the 285,000 claims processed were automated, with each taking 36 seconds rather than the 20 minutes when handled by staff.  

“I love the fact that Lacrimiora has been able to do a huge pivot and take humans out of the loop for public and societal benefit,” commented fellow speaker Jack Watts, business development manager for Artificial Intelligence in the EMEA region at IT solutions provider NetApp.    

16 Finnish applications

Mikko Laakso, director of ICT development at the Finnish Tax Administration, discussed how his team set out “to introduce ourselves to different technologies and different RPA suppliers, vendors and platforms” in a five year scale-up process and, after “proof of concept” studies, identified 150 possible use cases. The agency then acquired six RPA licenses in order to run 16 different processes.

The use cases, he explained, could be divided into four types: “mass manual work” where the objective was to get rid of repetitive, monotonous  tasks; “hidden tasks” where early processing could help boost efficiency downstream; multisystem work, where manual work to collate data from different systems often led to errors; and “pulling data”, where RPA could help draw analytical insights from data in different systems.  

Speeding processes in the USA

From the US Digital Service, data engineer and digital service expert Gina Maini discussed her work at the Medicare Payment System Modernisation programme: she’s involved in modernising the national health insurance programme’s Cobol-coded mainframes, introducing RPA to automate certain tasks and developing APIs to, for instance, allow authorised third parties access to legacy claims data.

In terms of the impact on people’s lives, she said, that project shared a core principle with an earlier RPA project at the Department of Homeland Security: there, “we built a technology that adjudicates [asylum seekers] quickly and with high precision, generating legal documents for an asylum officer. In a backlogged refugee or asylum system, you can imagine how many peoples’ lives can be helped or impacted by automating a simple document generation step. It took a process that would take an asylum officer three to four hours, and reduced it to five minutes. That was a game-changing piece of RPA.” 

Good data management essential

But the pandemic has also surfaced some underlying questions about rolling out RPA-enabled services. Speaking for NetApp, which orchestrates different IT services and middleware providers for public sector clients, Watts warned that rapid deployment might lead organisations to overlook the need for thorough protocols on data provenance, accuracy and versioning.

“When you can take a proof of concept and move to pilot really quickly, then scale it up to alpha and beta testing before going through model validation, it’s a great procedure to look at what data set versioning [systems] you’ve got, and make sure it’s all tracked and traced all the way through,” he advised. “Because when you’re in production with RPA, you may recognise there is some kind of bias or potential error, so you want to be able to go back and check what data sets were used, and how the model was built and validated.”

He also made a plea to public sector agencies to prepare the ground for deploying RPA or other automation tools – such as artificial intelligence (AI) – by building the right hardware architecture and software frameworks. “My ask is to consider what a good standard of AI platform looks like, so you can give the best tools to your scientists and they have the freedom of experimentation.”

A patch, not a fix

Finland’s Laakso noted that RPA is likely to be a time-limited technology, suited to improving efficiency within legacy systems. But when those systems are themselves re-booted into the 2020s, interfaces and processes should dramatically improve – reducing the need for RPA. For example, after the FTA invested in a new off-the-shelf system, GenTax, it found during the pandemic that the latter could flexibly accommodate new COVID-related add-ons. “So for us RPA is a band aid; a temporary solution,” noted Laakso.

An audience question drew attention to a problematic RPA roll-out in Australia: in the “RoboDebt” scandal, thousands of citizens were wrongly sent RPA-generated letters demanding repayment or money owed after the system’s designers made assumptions on how to calculate people’s taxable income. What happens when bots’ ability to work unattended and 24/7 means that errors or biases are silently replicated 24/7? How can public sector organisations safeguard against this?

Fears that robots will introduce errors as well as avoid them are common – and well-founded, said Laakso. “When introducing RPA, that’s a comment you often come across: if a person can do 200 cases per day and the robot can do 20,000, what if the robot does it wrong? After the work day is over we only need to fix 200 cases if it’s done by a person, and 20,000 if it’s done by a robot.”

He recommended pre-emptive measures, such as test automations and limiting the robot’s workload – for instance, setting thresholds of the maximum number of cases expected per day. The second line of defence is to find a way to make the correction automatically, rather than manually. But overall, he added, RPA processes need to be quality-assured and monitored as thoroughly as conventional processes.

The blurred boundary with AI

Meanwhile, other data automation and decision-making technologies are clamouring for attention. Watts described “federated learning”: a technology already deployed in at London’s NHS King’s College Hospital with the support of NetApp. It allows hospitals to share patient care image data for the purpose of training cancer detection [machine learning] models. “Having a ‘master model’, trained on millions of data points, to detect cancer with incredibly high accuracy and robustness, is paramount,” he said. “I’m sure we can all we agree that AI will have the most impact societally in healthcare. But as a concept from a collaborative data point of view, it can be applied in any industry.”

When an audience member raised a question on “hyper-automation”, Laakso noted how the FTA is exploring the field of natural language processing and handling unstructured data – such as the text in emails or chat logs. However, he noted that data science professionals cannot accelerate beyond of the legislative framework. “At the moment, Finnish regulation does not recognise AI. There is an effort going on to renew the legislation on the automatic processing of decision, but that is something that restricts us.” However, the tax agency is thinking ahead: “We have already drafted our ethical principles when using AI, so that we still do things ethically. That will be the keystone.” 

How to move forward?

Asked to nominate one shift or mindset change that would help public sector organisations to realise the full benefits of RPA, Maini argued that policy makers and data scientists both need to ensure that they are creating adequate feedback and accountability loops.

“When you deliver policy changes to agencies, you [should] also give a mandate that they must own the outcomes and the KPIs must also be accounted for,” he said. “It should be a key indicator to say whether or not the policy did what we said it would. Oftentimes we forget all about that, and we’re looking to the next fiscal year and we’ve forgotten about [last year’s projects].”

Corches, meanwhile, hoped that projects such as the successful RPA roll-out at the National Agency for Social Payments will raise expectations of how the public should experience their interactions with public servants – in turn fostering better uptake of process automation. In fact, there are signs that this is already happening in Romania, where staff in other departments have been asking if they too can have new robot colleagues.

“So in the beginning we imposed the change using RPA, and now HR are asking us to come up as quickly as possible with new solutions, similar to the new processes we have applied,” she said. Ultimately, she hoped, “the use of RPA will change the mentality for the public servant. We will interact with each other in shorter ways, and with better communication.”