While not without a few challenges, “The promise of machine learning is rooted in its ability to identify inefficiencies through a process of continuous learning and introduce efficacy, especially in areas where decision-making is required based on a set of criteria.” Geetika Tandon explores this in detail.
As I look at my son filling out his Common Application for college admissions, it makes me go back in time to over three decades ago when I filled out each application on paper and created a thick paper-based portfolio by sticking pictures of my work. I was able to print using word processing software and that was revolutionary. We have come a long way from those days and the change has not only been just a technological one but also a transformational one- one that has changed the way we live, work, and think.
Workflows and paper-based processes have progressed immensely in the last two decades. While several organisations are still playing catch up with automating their paper-based processes, the buzzword around automation is intelligent process automation, and business process automation has given way to business process optimisation. Are we on the brink of another transformation?
Business process automation has been around for more than two decades and has largely been associated with digitising paper processes, while business process optimisation refers to the ability to not just automate but to use insights and technologies to find efficiencies in the system. Machine Learning (ML) algorithms and Artificial Intelligence (AI) technologies have enabled a lot more insight that did not exist in the past. So, what is business process optimisation and how can ML help us enhance the way we do business today?
What is Business Process Optimisation?
Business process automation refers to automating business processes whereas business process optimisation is a more wholistic approach to business process management that looks at the process itself rather than just the automation. When looking at business process management (BPM), there are usually two aspects to it- the automation or digitisation of the process itself and business process improvement or optimisation which requires modelling of existing processes in a modelling software and then painstakingly identifying areas of improvement to increase efficiency. The latter is not necessarily required to perform the former but usually advised and often, organisations engage in a long process of modelling their current business processes and envisioning the to-be scenarios. Business process optimisation is not just automating an existing workflow but in the process of automation, re-engineering the process itself to introduce efficiencies.
Changing the game with Machine Learning (ML)
The advent of machine learning, artificial intelligence, and cognitive technologies has changed the face of business process management. What started with simple workflow automation where repetitive manual processes were digitised to improve productivity and reduce paperwork has moved onto mature digital transformation where robotic process automation (RPA), artificial intelligence and machine learning tools provide key insights that help in increasing the efficiency of processes.
The promise of machine learning is rooted in its ability to identify inefficiencies through a process of continuous learning and introduce efficacy, especially in areas where decision-making is required based on a set of criteria. Some of the major benefits of using machine learning include increased efficiency, faster decision-making, cost reduction due to reduced human intervention, and overall process proficiency for the organisation. Some examples of machine learning and AI fuelling business process optimisation would be:
Accurately assess business processes with Intelligent Data Processing (IDP):
Optical Character Recognition or OCR is widely used to convert pixelated content into real data. However, IDP encompasses a more comprehensive treatment of converting unstructured content into actionable data. Data processing has been around for over a century now and static data processing through relational database systems has been a staple for gathering, storing, and sorting data. However, technologies like machine learning, natural language processing, and computer vision can now help with document classification and data capture functions. This process includes data capture, detection, and extraction; and generation of AI-driven insights, as well as ongoing process refinement. This is especially applicable to paper and data-intensive processes such as claims processing for document analysis, fraud detection where an intelligent fraud solution automates the process of reviewing borrower documents and data while simultaneously identifying suspicious activities, and in the financial sector where IDP systems can verify gathered input from users and provide instant verification.
Improve decision-making through machine-based learning
Human intervention is labour- and cost-intensive and remains dependent on the institutional knowledge stockpiled in the brains of subject matter experts that can delay the process of decision-making. The most common example of these would be in mortgage applications where ML-based systems can analyse credit suitability and limits using data on income, assets, employment, and title. In the medical industry, ML-based systems are being used to provide diagnosis and care suggestions and in supply chain systems, ML-based systems can drive procurement decisions and optimised routes for efficient deliveries.
Predict future behaviour with machine learning
Machine learning can transform departments into profit centres by identifying business improvement using prediction algorithms. This is where the power of ML and AI-based systems comes together in full force. ML models can be used in multiple applications such as churn analysis and prevention, real-time recommendation, fraud analysis and prevention, and the targeting and personalisation of customer experience. Machine learning can extract and analyse the metadata needed to build viewing experiences based on historical trends. Forecasting what customers want, how much of it they want, and when they will want it is vital to any organisation´s success. Supply chain, sales, finance, and other business units are dependent upon accurate demand metrics to identify areas of efficiency. Optimisation through accurate prediction and forecasting can be revolutionary and change the way we work today.
Challenges of Implementing an ML-based system
Every new technology brings new frontiers and challenges with it. Implementation of ML requires organisations and perhaps nations, to rethink the way we look at data, privacy, and security. Below are some of the common challenges of using machine learning:
The biggest challenge to implementing a machine learning-based system is access to quality and adequate amount of data. AI models function best when they are trained on correct data and the accuracy, as well as the size of the data set, is critical. This often requires the reorganisation of current processes to provide data to the ML systems while still ensuring the privacy and security of end users in an organisation. It also poses a disadvantage to smaller organisations with smaller data sets while the bigger conglomerates are at an advantage in accessing large data sets. This is an area where policymakers need to come together to understand and implement the democratisation of data.
Another challenge is to sustain the continuous training needed to let the system produce accurate results. This requires providing the model with accurate training data and constantly re-evaluating the algorithm for accuracy. This is not a one-and-done situation. All machine learning processes (be that supervised, unsupervised, or a hybrid of the two) require the need to ensure that the raw data fed into the training dataset is reliable and exhaustive for the desired outcome. A consistent review of correct, complete, and consistent training data is required to set the foundation for successful machine learning projects.
According to the 2020 RELX Emerging Tech Executive Report, 39% of their survey respondents said the reason behind many of the significant challenges for adopting machine learning in business scenarios is a lack of available technical talent. The human factor remains a large issue in building and maintaining machine learning. Implementing ML requires a team of experts that can determine whether the architecture is purposeful, deployable, or efficient. AI expertise is still limited and there is a very small subset of the workforce that has any experience in implementing ML-based systems.
Data governance and model integration require not just deep expertise but also ownership of a solution, both of which are absent in the legacy organisational structure. We need a structured organisation that can facilitate change management. The governance structure is key to ensuring that transformation continues to expand use cases for machine learning-informed business process optimisation after system implementation. Adoption to transformation needs to be led with active championship from leaders of the organisation and opportunity to iterate by technical experts and domain experts.
Every business process can be optimised, and machine learning and AI technologies can change the game by providing insights and enhancing decision-making. However, as we see above, each technology has its challenges and limits. Most businesses hesitate to step in now because of high AI development costs and the lack of available expertise. Nonetheless, there is immense power in this technology to not just optimise existing processes but to ultimately transform the way we do business. The key to running a successful business in an agile world is starting small by identifying areas of high return on investment and building an AI backbone for the organisation and providing continuous improvement.
About the Authors
Geetika Tandon is Managing Director at Deloitte Consulting LLP with over 20 years of industry experience in technology consulting. She started her career at IBM as a developer working on voice and RFID solutions, moved to middleware implementation, and then acquired deep expertise in IT modernisation, helping multiple government agencies move to a cloud and DevOps environment. She currently holds four patent files and four IP publishes. She has also presented at various IEEE Software Technology Conferences, IEEE Women in Engineering summits and IBM Technology Conferences.
Andrea Jooyeon is a Senior Consultant with Deloitte Consulting LLP where she advises agencies in security, state, and civil government on strategic challenges related to enterprise technology and organizational transformation. She developed strategies to use machine learning to identify recruiting challenges and collaborate with government leaders in human capital.