The five common pitfalls of Cognitive RPA
This knowledge bite is part of a series about Cognitive Robotic Process Automation (CRPA) best practices.
When companies aim to unlock the potential of Cognitive Robotic Process Automation (CRPA) they usually tend to encounter the same pitfalls. In this knowledge bite we elaborate on the five most common pitfalls and discuss how to overcome them.
- Document volumes are too low for reaching a sufficient confidence level threshold and making a positive return on investment (ROI): One of the most important pitfalls is when there is not enough data for the robot to learn and recognize patterns. The robot needs large amounts of data to classify correctly. Furthermore, low volume processes do not necessarily have a positive impact on your business case, resulting in a smaller (or even negative) ROI. The selection of the right set of data as a use case is essential for CRPA. Check out our previous knowledge bite about the four essential steps for selecting a right use case for your organization.
- Data in the documents is too unstructured to identify data points and to classify them correctly: The cognitive RPA-bot’s first step is to identify data points in the documents and to classify them accordingly via (un)supervised machine learning. When the input data is too complex and unstructured, the robot won’t be able to sufficiently differentiate between the different data points. For instance, an invoice with tables and numbers with few text lines, will be much easier to process than custom legal documents with predominantly large paragraphs of text. That is why choosing document types with too many different structures of data will inhibit your Cognitive RPA results.
- Defining insufficient validation rules to train the bot: Before training the bot with (un)supervised machine learning, you have to define a sufficient number of validation rules or synonyms. You can for instance think of several conventions to write dates in documents such as invoices. When not defining these conventions beforehand, the robot faces difficulties in recognizing the right fields. Therefore, the robot will not deliver the results you expected when outsourcing routine jobs.
- Not including a sufficient number of ‘exceptions’ in the training set: Even seemingly structured documents will face exceptions. When training the robot with supervised learning it is of great importance to include those exceptions in the training set. This will lower your drop-out ratio of documents that need to be processed manually.
- Lack of ownership : Ultimately the robots will learn and change over time, same as the documents you process will change over time. While robots can take over a lot of tasks in the organization, they should be maintained by a team. Therefore, companies should ensure dedicated IT ownership for every robot. In addition, robots cannot reach 100% straight-through-processing of the documents, and therefore exceptions need to be processed manually. Robots can learn from this exception handling via supervised machine learning. The ownership of exception handling needs to be clearly allocated within your organization. Defining the ownership model of the robot and related responsibility is an important part of the implementation.
Contact us for more information about how Cognitive RPA can add value for your organization. More Synechron thought leadership articles:Four essential steps for selecting use cases for cognitive RPA
Five things you need to know about cognitive RPA
Automation the future is now
The evolution of chatbots
Chatbots for financial services
AI solutions for the FSI industry