In the last decade, technology has managed to completely change the way businesses operate, prompting them to go digital and improve their productivity.
However, thanks to innovative concepts such as AI and Machine Learning, it’s normal that more companies are aiming to reach a whole new level of efficiency, turning their organizational weaknesses into strengths.
To facilitate this shift, many organisations are starting to implement Business Process Automation or BPA to streamline their processes to remain competitive and relevant in the marketplace.
But what exactly is business process automation?
A simple search for BPA on Google will quickly give you a few definitions. Generally, Business Process Automation is defined as a methodology that leverages technology to automate processes in order to cut costs and increase productivity.
In considering BPA, a critical step is the identification of processes that would be suitable candidates for transformation. While this decision is multifaceted, from a logical perspective, two points stand out as critical factors.
The first criterion — one that is potentially obvious — is that a process is repetitive. Intuitively, a highly repetitive process will produce greater gains when improved compared to a process that is rarely run.
However, just because a process is repetitive, does not mean that it instantly becomes a good candidate for automation.
The next critical criterion for the effective implementation of BPA is if the process is codifiable. In general, the main component for a process to be judged as codifiable is if its inputs are structured data and the outputs are non-cognitive in nature.
If the inputs of a process are unstructured or semi-structured data and the outputs are cognitive in nature, traditional methodologies of automation and integration could falter and result in a high number of exceptions, rendering the BPA ineffective.
An example of a process with semi-structured data as inputs is an accounting department processing expense claims or invoices. While this work is repetitive, the inputs differ in terms of format and content because the receipts and invoices can come from many different sources.
Traditional methodologies struggle to codify the ever-changing input format leading to the need for manual efforts in the handling of such processes.
Also, the selection as to which account to book an expense or invoice to can vary based on certain scenarios and has a cognitive element.
Given the technology available, the automation of invoice and expense processing remains a challenging domain across many industries even to this day.
The Future of Business Process Automation
While some definitions of BPA focus on the automation of repetitive tasks, the definition provided by Gartner can potentially show the way forward for this methodology.
In the 2000s, the analyst firm coined the term “Business Process Management” to refer to a vast array of software applications that deal with processes — whether carried out by machines or individuals. Ever since its introduction, companies have been calling BPA the next big thing in the digital world.
According to Gartner, “Business Process Automation is the automation of complex business processes and functions beyond conventional data manipulation and record-keeping activities, usually through the use of advanced technologies.
The concept focuses on ‘run the business’ as opposed to ‘count the business’ types of automation efforts and often deals with event-driven, mission-critical, core processes.”
The definition’s focus on complex processes would suggest that BPA should be applicable to processes that have unstructured or semi-structured inputs as well as outputs that are cognitive in nature.
What’s interesting is that in order to handle unstructured or semi-structured inputs and cognitive outputs, BPA has to deal with processes that are not conventionally codifiable.
The solution to this, as suggested by Gartner, is the application of “advanced technologies”.
Many articles state that the use of Artificial Intelligence (AI) as an advanced technology that will help BPA in such scenarios. While this statement tends to point us in the right direction, it is also a very broad-based assertion.
So the next question is, how can BPA apply advanced technologies in a practical way to handle such complex processes?
A Practical Approach to the Future of BPA
Innovo42’s framework incorporates a multi-faceted practical approach in handling complex processes. A mix of traditional and innovative technologies are applied to automate such processes.
Our approach consists of:
• Stochastic analysis of the process
• Logical validation of the stochastic results
• Exception management
• Iterative learning
Deconstructing Stochastic Analysis
“Machine Learning at its most basic is the practice of using algorithms to parse data, learn from it, and then make a determination or prediction about something in the world.” — Nvidia
Machine Learning, which is a branch of AI, provides us an avenue to deal with inputs that are not structured and outputs that are cognitive. This is generally the application of stochastic models to
This method can be applied to unstructured or semi-structured input data to make predictions about its classification. The classified data can then be used to derive the required outputs.
For example, if an invoice which can widely vary in form, format and content is provided to a stochastic model, Machine Learning will be able to classify the document as an invoice and categorize elements found in the invoice such as total amount, tax amount, line items and others. The classified data can then be extracted and used as an output.
Another stochastic analysis could be run on the classified data to suggest the account to which the invoice should be booked into.
The ability for stochastic models to analyse data and make predictions is not something magical. For stochastic models to function, they must be trained and calibrated. The choice of the model used also has an influence on the result, where certain types of models work best for certain scenarios.
For training and calibration in general, the larger the quantum of data, the better the models will be trained and the higher the accuracy of the results will be. Of course, the assumption is that the data used for training is qualitative and realistic.
Similarly, the outputs of the stochastic models come with confidence level probabilities indicating the accuracy of the output.
Thresholds can be set on these confidence level probabilities so that data below a certain threshold can be flagged as an exception and be verified further.
Logical Validation of Stochastic Results
Due to the nature of stochastic models, logical validations are carried out on the outputs of these models to ensure data consistency. The logical validations include data type and logic checks.
An example of logical validation is checks done on data which are identified by stochastic models as a date. In this case, logical validation will ensure that the data element identified will qualify as a legitimate date.
Another example is stochastic models identifying a total amount in a receipt, in which case logical validation would ensure that the total amount is of numeric type and not an alphanumeric element.
Zooming in on Business Validation
The next step of validating the output of stochastic models is the application of business validation. Unlike the logical validation, the business validation checks the consistency of the data by applying rules that will relate to a business scenario.
Examples of business validation include checking if a transaction date is not in the future or if the tax amount is the correct percentage of the total amount.
Here, both logical and business validations apply deterministic rules on the outcomes of stochastic models. By doing so, the accuracy of the output is greatly increased.
Behind Exception Management
The reality is that no matter how much we cater for processes and train models, there will always be exceptions with any BPA effort.
Such exceptions can arise when stochastic models classify data with a confidence threshold lower than required, or when the logical or business validation highlights an issue with the data.
What is important is that there is an exception management process in place that allows the efficient management of exceptions.
A comprehensive exception management tool which isolates and identifies the exceptions allows the required intervention to be focused and prompt.
Finally, simple and effective handling of exceptions remains key to the successful implementation of any BPA initiative.
Iterative Learning in Business Process Automation
While exception management aids the handling of exceptions, iterative learning ensures the reduction of exceptions occurring in the future. The idea here is that today’s exception is tomorrow’s standard flow.
This is achieved by taking the corrected data resulting from exception management and feeding it as training data into the machine learning algorithms to calibrate them.
This calibration will help increase the accuracy of the stochastic models and reduce the overall exception rates progressively. Over time, the accuracy of the models will continue to improve as they are used.
The Impact of a Practical Approach
The practical framework which combines both advanced technologies and traditional methodologies will help multiple processes that consume unstructured or semi-structured data or have cognitive outputs into viable candidates of BPA.
The application of the framework will help companies automate multiple processes that have so far eluded BPA initiatives across various industries.
Also, as Machine Learning models and methodologies evolve, the number of processes that can benefit from BPA will continue to grow.
Our stance is that processes are companies’ assets and require investment just like any other business asset. Ultimately, with a practical approach to BPA, companies can boost their productivity and go digital.
Liked this article? Check out our article to find out tips, insights and real examples from our team that will help you modernize your expense management workflow.