Business Analytics and Machine Learning Applications – Reflections from Durham Analytics Day

On the evening of 16th March 2021, the Durham Analytics Day was held online, at which academics and industry experts discussed how analytics can enhance decision-making processes. The two-hour event attracted over ninety participants to a series of interesting talks delivered by Eduardo Contreras Cortes (4most) on machine learning applications in the UK financial sectors (e.g., credit risk), Prof. Frank Krauss (Durham University) on a wide range of grand challenge projects undertaken by the Durham Institute for Data Science and Angela Castillo-Gill (Deliveroo) on building marketplace products with small decisions. This event was chaired by Dr. Riccardo Mogre, who is a co-director for the Centre of Innovation and Technology Management and supported by the Operational Research Society , the Institute for Data Science, Durham University Operational Research and Data Analytics Student Group and Women in Data Science.

Machine Learning (ML) Applications in Credit Risk

The first talk by Eduardo, who is a principal consultant in the 4most, the largest UK consumer consultancy, began by asking the audience about their experience with the ML. Surprisingly, only a few people indicated that they had some experience with the concept. Given the unfamiliarity with ML, he began by providing an overview of ML and its implications for people and their daily lives. ML is part of artificial intelligence and computer science, which allows people to analyse big datasets, then identify complex patterns in data. The application of ML has significantly impacted our daily lives, for example with regards to spam email detection, self-driving cars and Netflix or Spotify recommendations, among others. In the financial service sector, ML has been widely used in attempting to assess credit risk (e.g., concerning loans, mortgages, overdraft, and credit cards). For instance, an individual's credit score is an assessment measure that assists lenders to make decisions in terms of how they respond to a request to borrow money. The credit score system of an individual is built by data from the individual's financial transactions recorded in the banking system over time, including their track-record as regards repayment of debt. Three decades ago, such decisions were based on individual/leader's accumulated knowledge and experience. With the development of computing, the traditional approach used in building the credit scores has evolved into a statistical approach (e.g., logistic regression) by incorporating all kinds of data, which, in turn, creates different probability models, in an effort to make better decisions. Simple statistical regression is, however, incapable of identifying complex patterns, when there are more data available. ML offers more powerful algorithms in computing to identify more complex and accurate patterns. In a survey conducted by the Bank of England in 2019, two-thirds of the businesses in the UK financial service industry are using ML, particularly in the banking sector. Some possible explanations of ML being widely used in the banking sector could be the legacy of credit risk and the usefulness in detecting fraud, as well as saving staff costs. As is evident, risk management, customer engagement, and insurance are areas catching up with the development of different models of ML that are suitable for their business models. 

Eduardo has also highlighted the importance of understanding the different techniques or algorithms used in ML. For example, the self-driving car area uses image recognition algorithms, while, in financial services, three-based models and natural language processes are commonly used. How is the pandemic leading to new uses for ML? Many banks, including the Bank of England, have found the ML and data science can play a vital role in maintaining the basic functioning through automating processes and applications (e.g., customers' engagement and services), when employees are absent due to the effects of the COVID-19 pandemic. However, the exploration of how ML works is in a nascent stage. Notably, there are some principles, including 'easy to understand', 'transparency' and 'auditability', that are applied in providing guidance when deciding on the appropriate algorithm for use in the different contexts. This is because the outcome has a direct impact on people, such as that people may ask a reason why their loan applications had been refused. The utilised ML model should be able to explain the logical reasons for reaching the decision. In addition, any applications must comply with the rules and regulations to build multilateral trust with customers and the financial authorities. 

From the perspective of management research, Business Analytics and ML are strong driving forces to the 'fourth industrial revolution' or 'Industry 4.0', along with the advancement in computational power and the exponential increase in data. By incorporating a large amount data in the decision making process, the adoption of business analytics or ML can assist in making more accurate decisions in theory, compared to those decisions made by human actors who may have either conscious or unconscious bias. The training process of ML, however, can introduce unintended bias, resulting from the nature of the data, the algorithm, or the data-algorithm interaction (Danks & London, 2017). In addition to issues mentioned by Eduardo, fairness may emerge as an issue when ML is used on a large scale for business. It is possible that members in certain groups may be systematically discriminated against, as regards being granted mortgages or other financial provision, due to the existence of data bias. Thus, regularly reviewing the performance implemented by ML and Business Analytics becomes essential in order to avoid unintended bias. Second, the use of ML-based solutions in automating organisational routine tasks and operations is not new, for example, call centres has existed for some time. Notwithstanding this, such automation based on pre-programmed rules in performing repetitive and monotonic tasks (Parasuraman & Riley, 1997) differs from ML algorithms, seemingly involving more iterative interactions between human and machines, such as Netflix's decisions on directors and actors, and Pfizer's drug development activities. The emerging hybrid organisational forms are challenging assumptions of traditional organisational governance and solutions on a wide range of issues (e.g., trust, reliability, accountability, and governance), which have societal implications relating to equality, security, transparency and employment. Management scholars, thus, are encouraged to broaden their perspectives to explore the distinct behaviours, namely a machine's influence on human behaviours and human influence on machine behaviours. 

Data Science in Tackling Grand Challenges

The second talk was given by Prof. Frank Krauss who differentiated between two mainstream views on data science, one focusing on ML and AI, and another one referred to as 'Core Data Science ML' that is at the interface between statistics, machine learning, pattern recognition, and human interpretation. The applications of the cutting-edge techniques are implemented in different areas through PhD student placements. For example, the first placement is to help a local small enterprise, which is an X-ray imaging company, to develop software and solutions for improving X-ray images. Another placement with the UN is to work on the identification structure of large refugee camps in Jordan and Syria to distinguish residential tents from other tents (e.g., kitchen). Other applications include investigating how to produce fake news based on natural language processing, which has been used during the COVID crisis, and applying deep learning methods to improve trading strategies in financial companies, as well as predictive maintenance work for the nuclear power plant at Hartlepool in reducing unscheduled downtimes, and the simulation of epidemiology (e.g., the construction of a digital twin of the UK population). In collaboration with Geography, the ongoing research is related to some remote sensing activities on the pathogen identification in a plant (e.g., coffee plant in Thailand) and monitoring mangroves which is important because of their functions in stabilising the coastline and absorbing CO2. There is a big impact in terms of helping organisations such as the World Wildlife Fund for Nature preservation and mitigating the impact of climate change by suggestion appropriate plants in different regions. 

Prof. Krauss' talk demonstrated the power of algorithms that can be manifested across different disciplines and areas. Not only doe it help humans to work more efficiently and effectively but it also augments our capacities in tackling a wide range of issues faced by humanity. What has been offered here reminds me of the ongoing debate in management on automation (e.g., machines taking over human tasks) and augmentation (e.g., humans working closely with machines to perform a task). From relational ontology, there is a decomposable relationship between automation and augmentation and the challenge raised here is about how to find a balanced approach embracing both automation and augmentation in business and management practices, which, in turn, should result in positive societal implications. 

Building a three-sided marketplace product: the case of Deliveroo 

The third speaker, Angela Castillo-Gill from Deliveroo, shared insights into how to use the framework 'if this...then that' to quickly make 'good enough decisions' in a hyper-competitive business environment. Operating in the three-sided marketplace constituting consumers, riders, and restaurants is very challenging particularly when it comes to decision-making under time pressure. Every decision has the potential to impact all aspects of the market, namely three actors: consumers, riders, and restaurants. It is essential to have a process which can indicate the likely impact of any decision. There are two inherent concepts introduced: the interconnectedness of the marketplace and the cost of delay. The interconnectedness refers to non-decomposable relationships among three actors in the marketplace, which jointly drive the net profit for businesses. The cost of delay is a key driver to launching a marketplace product sooner than later because winning a favourable time is critical to survive in the severely competitive environment. Thus, how to quickly make 'a good enough decision' under time pressure is a central focus and this can be achieved by 'thinking upfront'. The notion of 'thinking upfront' from the perspective of data science is contextualised by the 'if this and 'then that' rules governing the interconnectedness and balancing the cost of delay. This approach is deemed suitable because it considers human inputs in the necessary informed quick decision-making. Angela's talk offers a fresh view on how a 'born in digital' business operates in the three-sided marketplace for value creation. By incorporating both humans and machines in the business model, this might be a good starting point to observe and explore the human-machine interaction. 

The rapid growth in technological development and digitalisation driven by the fourth industrial revolutions brings with both positive and negative impact on organisations and society. The complex interactions between human and machines have begun to be manifested concerning several issues relating to fairness, transparency, trust, reliability, accountability, among others. Undoubtedly, this has not only presented methodological challenges for researchers in terms of how to access these problems but also has offered an opportunity to develop a more innovative and collaborative approach in advancing our research. It is important to continue today's dialogue across different disciplines and sectors to better understand the state of the art on the use of business analytics and machine learning and explore how to attune the power of algorithms towards more sustainable development goals. The fascinating and informative event was closed by Matthew Robinson (IBM) from the OR society (https://www.theorsociety.com/) . Other analytics groups in Durham include

Analytics at Durham University Business School (Dr. Riccardo Mogre):  research stream on business analytics in the Centre of Innovation and Technology Management (https://www.dur.ac.uk/business/research/management/operations-management/) ; Durham Rutgers Accounting Analytics Network; MSc in Business Analytics (joint with Computer Science). 

OR and Data Analytics Student Group (Angus Bolton): offering careers advice and networking opportunities; teaching student's relevant skills and increasing understanding of the sector. 

Women in Data Science (WiDS) (Yara Jubran and Rowan Morris)(https://www.widsconference.org/) the WiDS initiative aims to inspire and educate data scientists worldwide, regardless of gender, and to support women in the field. It runs 200+ regional events worldwide in more than 60 countries. For the first time, we are bringing the WiDS to the Northeast. In June 2021, the WiDS X Northeast event will feature remarkable women doing outstanding work in the data science field; provide an opportunity to hear about the latest data science-related research in a number of domains; learn how leading-edge companies are using data science for success; connect with potential mentors and collaborators in the field. 

*The views expressed in this post are solely those of the author.  Jacqueline You, PhD in Management, the founding Chair of Doctoral Student Group at the Centre of Innovation and Technology Management, Durham University, UK, and Lecturer in Strategy, Operations and Entrepreneurship at Essex Business School, University of Essex, UK.

The author would like to thank Dr. Riccardo Mogre and George Karmatzanis for their insightful comments on this post. 

Reference:

Danks, D., & London, A. J. (2017). Regulating autonomous systems: Beyond standards: IEEE Intelligent Systems, 32(1), 88-91.

Parasuranman, R., & Riley, V. (1997). Humans and automation: use, misuse, disuse, abuse, Human Factors, 39(2), 230-253