Financial functions are highly standardized at Faurecia, a leading global manufacturer of automotive parts headquartered in Nanterre, France. Since 2007, the company has worked with SAP, rolling out a single, integrated ERP system to about 98% of its 300 sites around the globe. "We have 35,000 users in SAP," said Anna Berger, an SAP expert and the functional design lead in finance at Faurecia.
So, when Berger and her IT team embarked on a robotic process automation (RPA) project with Redwood Software, Faurecia was "already advanced in its harmonization of processes," she said. But the automotive supplier was determined to drive out additional costs and improve efficiency.
A major restructuring effort to convert 25 IT shared services to five or six regional platforms provided an opportunity. Instead of two accounts payable services in one country, and three in another, this massification, as Berger calls the shared services consolidation, "creates a bigger hub of people around the same topic."
"Our ambition [for the RPA project] in this context was to eliminate manual, repetitive tasks, which have limited added-value, and to automate an end-to-end process," she said.
One function ripe for automation was the posting of journal entries, or the transference of recorded business events from the general journal to the ledger. "For a more or less simple process, there were many steps, many people involved and lot of back and forth communications, so it was very inefficient and time-consuming," Berger said.
She chose Redwood Software for the RPA project because of the vendor's longstanding partnership with SAP and its experience in automating standardized back-end processes. (For RPA projects involving less standardized processes, as well as those that involve multiple smaller processes, Berger works with RPA vendor UiPath.)
The RPA project with Redwood Software kicked off in 2017 with the initial aim of automating three processes by April 2018. The first bot went live in May, was fine-tuned through July and is slated to be deployed worldwide at the end of November. A second bot it is running on two pilot sites and a third bot is in redesign.
We asked Berger to discuss her strategy for implementing RPA and some of the challenges of getting her project off the ground.
Editor's note: The following was edited for clarity and brevity.
RPA strategy: Not waiting for perfect
One of the debates in implementing an RPA project is whether to optimize the process first or automate it as is. How did that issue play out for Faurecia, where processes are already optimized?
Anna Berger: We were not starting from zero with RPA, but we had not achieved 100% optimization. What we decided was we would not wait for the process to be fully optimized, but instead to automate it, as you say, 'as is,' and then when the robot was already working to come back and improve the process itself. Our philosophy is if you wait for 100%, you will never go.
Anne Bergerfunctional design lead, Faurecia
How did you go about analyzing which parts of the process should be automated and which parts are perhaps better left to humans?
Berger: We gathered some key players from two regions for our pilots. We went through how they work today. It was interesting because, as I said, we do have one single SAP model and thought we quite harmonized; but in the end, we figured out there was local variance, and so, in Romania, employees were not working exactly the same way as in Mexico. That was one of the important steps.
At this point, we decided what should be the workflow, who starts what, what action triggers the next step. We also defined what should be automated and what should be done by humans. For that, we basically asked, 'What are the steps which fully follow the standards and the defined rules -- which don't require, let's say, intelligence?' What we had left for the human part were [steps] where a decision or a validation is needed.
How much training was required to get a bot to do what you needed it to do?
Berger: A lot! For a very basic robot, in theory, you can apply it out of the box, but the truth is you always have difficulties when you want to adapt it to the company's specific rules or processes. And there is the point where it becomes tricky and where it takes time, and where there are a lot of iterations going back and forth to try to address that work.
What's tricky about bot training?
Berger: The challenges are around comprehension. What the business describes is not necessarily what the robot engineer hears. So, you really have to go several rounds to make sure you have the same understanding and are speaking the same language.
In the worldwide context in which [Faurecia] operates, we really had to make sure we had the right people on board to ensure we ended up with the same process. What we saw as they defined one process was that the users basically agreed, but when they started testing the [automated] process in the system, they saw that a small detail was missing, or a small detail was different between Mexico and Romania, and then they had to go back to integrate the small details.
Definition of the design is very, very important, and this is where you have to spend time to make sure you have [captured] 90% or more of the process, so you don't have to redesign later on.
RPA project results, benefits
What gains are you seeing from the bot that went live in May?
Berger: Well, journal entry posting is a very high-volume activity -- just to give some figures, we have ramped up from 500 journal entries posted by the robot to 7,000 posted in August. This means there are 7,000 journal entries which are not posted by the shared service anymore, but by the robot.
In terms of gains, the processes are running quicker, because we have automated the steps. What is not that easy to measure is quality, but we have a [metric] on quality, because we have implemented additional controls -- coherence checks and things like that, so that [metric] is coming. Also, our auditors have better traceability of what we have done with postings, because our general entries is based on an Excel file, which is attached to the posting.
Lessons learned from implementing RPA
For CIOs who haven't done an RPA project, what would be useful for them to know?
Berger: For me, choosing the process to start with is so important. It should be a process that you understand well. Also, if it's a complex process, don't be too ambitious to try to automate everything right from the start. You can also go step by step, meaning you only optimize the first two steps and then you continue later on to automate more.
If you're too ambitious at the very beginning, it will take too much time and you'll just be frustrated. And, as I said, involve the right people. It is very, very important to have a good mix between IT people and key users with operational knowledge of the process. And don't underestimate the time spent, especially in the design phase, where you have these iterations I talked about and where communication is important -- you must be clear where you are and where you want to go. This investment in the beginning will pay off in the end when the bot is put in production.
Have your bosses been pleased with this implementation of relatively new technology?
Berger (laughs): Yes. Even if parts of it were painful, it was a very good experience. At the start when it was not that easy, we were fortunate to have very good relationship with Redwood.
Automation is something we're all aiming for -- and we're working in the automotive industry, where it is very important to be efficient using robots on the production side. So, to introduce this transformation in support functions, it was a great experience. And it feels good to help the company; it shows that IT is not just a burden but can really contribute to the transformation of the company.
Editor's note: Read a discussion on why more enterprises are taking a multivendor approach to RPA here.