RPA takes over repetitive business processes, such as invoice processing, so staff can spend their time on tasks that add real value for the company and themselves. Initially, it was based on how the user performs the task, and would mimic or repeat the manual processes in the interface. But we wanted to see how the technology has evolved, and how it is being used in real situations.
Jon Holvoet from Credendo set the scene: the company has begun evaluating the possibilities of an RPA solution. Then, Joris Van Ostaeyen of robonext explained how to strategically implement RPA. Finally, Bram Vanschoenwinkel of AE introduced us to ‘Cognitive RPA’, which adds Machine Learning into the mix.
Presentations and the recording from the event are available to our members (after log-in).
Case: Evaluating the potential use case and return of RPA
Jon Holvoet, CCOE Manager at Credendo, started with a list of the promises and benefits that come with RPA (slide 3). He shared the one that is perhaps most important to him: it is a non-invasive technology that doesn’t disrupt underlying legacy systems.
Once you are convinced of the benefits, your challenge lies in selecting the right use case for your organisation. Possibilities can be found in your ordering processes, accounts receivable and payable processes, vendor creation, sales, or within your technical support organisation. So there are plenty of options at your fingertips!
At Credendo, the first use case will probably come from finance, which still has a lot of manual tasks (e.g. expense management and invoice processing).
Jon wrapped up with the areas where he could use some help for RPA:
Getting strategic with RPA
Next up was Joris Van Ostaeyen, Chief Strategy Officer at robonext. He framed RPA as a software technology with the possibility to create a digital workforce, responsible for automated, repetitive tasks. He defined potential processes for automation as those that:
For these processes, RPA robots can
Including RPA in your mix certainly enables productivity gains for your organisation. But while it is often promised as a way for quick wins, these quick wins come with risks, such as:
Going fast is good, but going too fast is highly unadvisable with RPA.
Next, Joris turned to the foundations of a solid and strategic RPA program (slide 12). Typically, when choosing a process for automation, select one that is well-structured and standardised. Automate a bad process and you will only accelerate the flaws.
The number of vendors in the RPA market is growing (slide 16). While it is still fragmented, it is also concentrating around 3 larger providers: UiPath, Automation Anywhere and Blue Prism.
When looking at the people-aspect, there are two angles: who will develop the RPA and who will be impacted by it? RPA is mostly used to increase the company’s productivity or to redeploy employees within the organisation (slide 19). Few use it to lay off employees.
And last but not least: how is governance of an RPA programme organised? Often, RPA is led by business and governed by IT. Many organisations set up an RPA Centre of Excellence for the governance.
Cognitive RPA as a key enabler to enhance human abilities and empower people
Our final speaker was Bram Vanschoenwinkel, Hive Lead Analytics at AE. While traditional RPA supports automation based on structured data, he explained, cognitive RPA uses Machine Learning to automate more complex, less rules-based tasks where complex interpretation is required, like those based on unstructured data sources.
The automation of customs declarations, for instance, is a use case suitable for cognitive RPA. This process is highly time consuming, involves plenty of documents and sub-documents (often poor-quality scans). A second example is AI-enabled auditing of food processes and product information (info about nutrients and allergens).
When looking at AI, there's a lot of buzz in the market (slide 5), causing organisations to struggle with choosing the right solutions for their needs.
Bram shared the cognitive solution AE came up with for the customs use case (slide 10).
The second step is the proof of value, with an assessment of the technical feasibility. It is important that various document types can be recognised (invoices, confirmation of discharge, package list, etc.). Once that is possible, blocks of text that logically belong together must be identified, as well as the ability to detect what kind of information is in these text blocks. Typically, this is also related to the document type. Finally, the information itself is extracted from the documents (the 'data retrieval').
Once feasibility has been assessed, the actual changes can be planned and implemented, and the solution can be developed.
Bram strongly believes in an incremental and iterative approach: think big, but act small (slide 14). Rather than the AI replacing people, Bram pleads for a technology for, by and at the service of people.
Smart technology comes with AI, neural networks, machine learning, natural language processing, deep learning, etc. Solutions can be put on a spectrum that varies from simple (solutions you can buy, with e.g. existing APIs) to very complex (with the development and research of natural language processing and computer vision models, which typically require many years to build).
Bram takes a stab at the middle. By retraining existing models ('transfer learning') by giving them additional data for the specific context at hand, the models improve and can be used for this context.
Bram wrapped up with 3 takeaways:
Access to more information about this topic and/or to download the paper is easy and fast, but exclusively for Beltug members (just login to get access).
Beltug gathers a lot of information. Here you find the advantages of Beltug membership
The Beltug Team
Click here to login