How can the public sector free up resources while continuing to fulfil its missions? By enabling certain time-consuming and repetitive tasks to be automated, the digital assistants or RPA (Robotic Process Automation) are one possible solution. They enable public servants to spend more of their time on tasks with greater added value. Internationally many public sector organisations have launched projects with the RPA technology, including the French tax administration or “DGFiP”.

An RPA success story at the French tax administration DGFiP...

In 2018, the DGFiP decided to partner up with BearingPoint to launch a study on the use of digital assistants or RPA. Given the use of this technology was not widespread within the French public sector at the time, the goal was first to assess the feasibility of integrating an RPA solution within the DGFiP IT ecosystem and test it on certain simple business processes. The Proof of concept (PoC) proved successful, which led the DGFiP to launch a project with the goal of automating several business processes and to reflect on a future scaling up of the solution: how can we identify and characterize processes with high potential for automation? What is the right long-term governance for the RPA project? What IT architecture is most appropriate? etc.

Having automated several processes, the DGFiP is now developing a long-term RPA framework with the introduction of an RPA center of excellence in September 2020 and of a new technical platform in March 2021. Analysis of other business processes is underway to integrate into this this approach.

With the experience as co-leads for the RPA project, Karen Plissionnier – deputy head of the user experience team within the digital transformation service at DGFiP – and Véronique Jung – IT project manager at DGFiP – have answered our questions and share below some precious feedback for any public actor interested in starting a similar approach.

1. Where did the idea of an RPA Proof of Concept come from?

The DGFiP identified the potential for gains enabled by RPA automation, in terms of performance and public servant experience (accelerating processes, reducing the workload associated with of repetitive tasks, improving security and traceability, etc.). By 2018, the DGFIP was interested in experimenting with the technology. At the time, we wondered if it could fit within the DGFiP IT landscape – which is broad and diverse – and if it could offer significant added value to our business activities.

We quickly identified certain constraints:

  • IT security policy
  • Diversity of existing software and tools that the RPA would need to interact with, ranging from old mainframe apps to new web portals
  • Highly sensitive areas of business, covering taxation and public accounting

Before going any further, we decided to launch a first experimentation in 2018 with a PoC on four business processes.

2. How did you choose the processes to experiment as part of the PoC?

We chose several processes to cover a large spectrum of business domains and types of operations. The four prioritized processes therefore touch on tax collection, Human Resources, management of relationships with suppliers and on different types of operations (matching information between internal and external sources, consolidating information asynchronously, optical character recognition (OCR), etc.).

3. What were the main PoC take-aways?

The PoC quickly demonstrated the technical feasibility of automating of all the processes in the experimentation, from the least complex to the most through RPA.  Beyond proving feasibility, it also showed quantitative and qualitative gains:

  • A significant, swift and measurable return on investment: an RPA robot can save between 50% and 90% of a staff member’s time spent on a process (from the first PoC) results
  • Public servants can be freed from repetitive and time-consuming tasks, and focus more on higher value tasks, improving their experience and job satisfaction.
  • The method encouraged the standardization processes, aligned between services
  • The PoC delivered an improvements in data quality and risk-management (disappearance of manual inputting errors, coherence checks between apps)
  • A rapid response to business needs: deploying non-intrusive and dynamic IT solutions easily and swiftly

These results drove us to start a first live deployment on two pilot processes as part of a wider experimentation.

4. How did you choose the processes for this first live deployment and for later expansion?

For the first live deployment, we wanted to move forward fast. Therefore, we put to one side the most complex processes and prioritized those less intrusive for the DGFiP (no inputs to critical DGFiP apps). Analysis of the benefits from the PoC also allowed us to favor processes with the highest return on investment.

Given the very positive results of the live deployment of the two pilot processes in late 2018, a mapping of automation needs within local offices and central teams was undertaken at the beginning of 2019. We analyzed every one of the 130 business requests to assess the adequacy of RPA automation according to different criteria (task repetitiveness, process standardization and documentation, number of systems involved, etc.). RPA automation seemed particularly well suited to four processes that were selected for the new deployment. This first mapping campaign, largely experimental, showed the importance of helping functional areas within the organisation expressing needs to better adapt their proposals to the analysis criteria and potential return on investment. Due to the volume of needs that were identified, a a more structured methodological framework based on a standardized eligibility grid was developed in the summer of 2020 to help scale up and “industralise” the approach.

Type of automated process  
Business domain

Role of the RPA robot

Tax recovery / control Collecting different data within IT systems and publishing a consolidated result.
User-relationship management

Downloading documents and publishing a consolidated result.

Downloading documents and sending them to the user.
Accounting and recovery Downloading a bank statement, identifying a list of operations to be treated and then enter corresponding inputs in the adequate IT systems.

5. How was the approach scaled up across DGFiP?

Two major subprojects were central to the scaling up of the RPA solution at DGFiP:

A. Governance

  • A detailed reexamination of the governance, reviewing the roles and responsibilities of different actors and adapting the structure of governance bodies. In order to accelerate and safeguard the live deployment of processes, the different steps of RPA projects were updated and set out in more detail.
  • Scaling up the method for mapping business needs, analyzing and selecting processes to automate.
  • Establishing an RPA center of excellence, based within DGFiP IT services with a full-time team dedicated to RPA robots.

B. Infrastructure

Progressing the technical architecture of the platform: an analysis on the scale and sizing of the infrastructure was undertaken based on the future ambitions for process deployment. We then implemented a highly-scalable technical platform in line with this ambition.

Switching RPA solution editors: the state of our project allowed us to change editor without risk or major business impact.

These subprojects enabled us to have:

  • A stable RPA platform capable of hosting new processes in the long run.
  • Clear ways of working (actors, phases, decision-making) to deploy new processes, from mapping needs to live deployment and maintenance.

6. How was the choice made to switch RPA editors while scaling up?

During the deployments, we encountered difficulties with the initial RPA technical solution, which was chosen at the PoC phase. With hindsight and in looking to scale up this technology, another solution seemed better suited to our needs and requirements, in particular from a technical and support perspective.

The decision to switch editor also raises the question of the choice of solution: even if an adjustment is still possible down the line, it is preferable to compare the different available solutions and to select the one which seems most adapted to the context from the very start.

This change of trajectory was possible because we were at a turning point in the project cycle. The relatively limited number of RPA robots deployed allowed us to change solution without a major impact on existing processes. This decision was also taken to meet the ambitious objective of deploying RPA robots on a major process involving numerous special cases and impacting a large population.

The switching of solution was key in the scaling up of processes at DGFiP. The implementation of a new technical platform supported this scaling up and the center of excellence, which needed to be reinforced and to develop skills on a new technology.

7. What lessons were learnt during the scaling up phase?

During the scaling-up phase, it’s necessary to anticipate and write down all the potential blockers that can be encountered.

During the scoping of needs and requirements, we realized it was important to have the following elements in mind:

  • Identifying and writing down all the business rules of processes, in order to analyze at the most detailed level the expected behavior of the robot.
  • Never underestimating the complexity of processes, for example the number of specific special cases which can generate deployment difficulties and the effectiveness of the robot.
  • Anticipating issues with GDPR conformity, which may take time to resolve and impact the timeline for rolling out robots
  • Defining well in advance the expected orchestration of RPA robots, so as to ensure the best possible maintenance.

During development, three main items appeared critical to us to ensure project success:

  • Defining in advance all “exit doors” for the RPA robot, to allow it to react in case of unforeseen events.
  • Mobilizing IT developer profiles for the RPA customizing phase.
  • Not underestimating the duration of preproduction work.

Several key challenges were also overcome, and must not, be forgotten:

  • Moving from demonstration on a restricted scope to live production for many services or offices.
  • Scaling up a method, a governance and an associated set of committees.

With the success of our RPA approach and the benefits we measured, and the business actors have recognized, we are moving forward with further robot deployment:

  • We are automating new processes.
  • We are further developing existing processes and completing a project to repatriate the processes that were deployed initially to the new technical platform.
  • We are expanding and reinforcing our center of excellence to increase our rollout capability and capacity.


Would you like more information?

Si vous souhaitez en savoir plus à ce sujet, nos experts sont à votre disposition.

Toggle location