My journey towards analytics and big data

JOHN RANYARD, COMPANION OF THE OR SOCIETY

I joined the British Coal (then NCB) OR Branch (subsequently the OR Executive, ORE) in London in 1963. My first section leader was George Mitchell, later to become Director, ORE and President, Beale medallist and Companion, ORS. An early study involved assessing when underground coal cutting machines should be withdrawn for overhaul. The relevant data, i.e. type of breakdown vs time in use, was usually in engineers’ notebooks at collieries, often oil stained, which required long hours for collection, classification and analysis. The results showed that the interval between major overhauls could be extended, probably because the reliability of the machines had been improved over time by the manufacturers.

In 1965 I started an in-service PhD at Lancaster University OR Department, then headed by Pat Rivett (former President, ORS). My topic was evaluating newly introduced electronic monitoring of underground machinery, which I completed in 1972. These new devices produced masses of real time data that needed analysing so as to produce information for decision making, which these days might be labelled data analytics.

In 1972 I was appointed Manager, Yorkshire Unit, ORE. Many studies were pragmatic, requiring fairly simple approaches and few involved large databases. One exception was a responsibility we had for safety statistics, which were often discussed at board level and with the relevant government Ministry. At one time, a particular set of safety statistics was quite controversial and the consultant* responsible and I checked the analysis in detail to ensure that it was correct. Whilst I was satisfied, I was aware of the impact of any errors so I commissioned a statistician at Sheffield University to check our analysis (for a fee) and he confirmed our results. This was just as well as shortly after our results were quoted in an interview with a Coal Board representative on the Radio 4 Today programme, which was and still is a forum for debate on controversial issues.

John Ranyard
Companion of The OR Society

“shortly after our results were quoted in an interview with a Coal Board representative on the Radio 4 Today programme”

My real involvement with Analytics and Big Data (ABD) began after I retired from British Coal in 1992 and, shortly after, joined the Lancaster University Management Science Department (as the OR Department had become) as External Liaison Manager. One of my roles was to set up 4-month projects in industry for OR Master’s students at the end of their 8-month taught course. I also supervised projects, some of which did involve masses of data.

Both Barclaycard plc and Littlewoods Home Shopping had the characteristics of several million customers in their databases and were mining this information for increased profit. Scorecards were being devised by the respective OR teams, so as to inform decisions such as which applicants to accept as customers and their initial credit level. These scorecards were based on logistic regression, decision trees and occasionally neural networks, which were becoming key skills for many OR consultants. Each new scorecard would be tested on a sample of the full database and once it proved to be successful for the corresponding decision it would be rolled out to the full database for that decision and be labelled the ‘Champion’.

Many customer characteristics and other relevant criteria, sometimes as many as 80 pieces of data, were included in  the scorecards, some of which would change over time, so scorecards needed updating at regular intervals to maintain their value. The in-house teams were big enough to develop and maintain the most important scorecards but the more minor ones were updated less frequently and this is where Master’s students, who by this time had acquired most of the necessary skills, could assist, particularly as they were available when many staff would take their holidays. The cost charged by the University (£4k plus expenses at the time) would be repaid many times over by the updating of a scorecard.

So, the students would gain valuable experience about how their newly learned techniques were being successfully applied in practice. This would include how the integrity of a very large database is established and maintained and how to take representative samples for developing and testing a scorecard. (Over time, as the power of computing hardware increased, together with the increased efficiency of the software used, the full database could be used to develop and test a scorecard, leading to improved predictions). The students would also be involved in ‘selling’ their updated scorecard to the relevant marketing or financial manager, who would need to be convinced that any changes in parameters from the previous version would lead to improved results. For this reason they were resistant to neural networks, which are in effect ‘black boxes’, providing no explanation for an improved performance. In contrast both logistic regression and decisions trees have visible parameters, some of which captured the impact of demographic variables. These particular variables could be discussed with the client, who would be very reluctant to approve a parameter change that contradicted his or her market knowledge!

Whilst analysts in both these organisations were recruited from the OR community, with many members joining The OR Society, this was not always the case in similar data rich organisations that also needed scorecards for improving profits and remaining competitive. In my job at Lancaster and in various roles for The OR Society, I promoted the value of OR Society membership to such staff, not only to enable their technical skills to be continually updated but also to understand and embrace OR’s mature methodology that helps to maximise the prospects of implementing the results of analysis.

When Davenport and Harris (1) publicised the value of data mining for many data rich organisations in their seminal book in 2007, they started the Analytics/Big Data (ABD) revolution. Before long many organisations discovered they would no longer be competitive without embracing it. New jobs began to appear and be in demand: analytics consultant, data scientist, data engineer  etc. The OR Society quickly realised that many of the skills required, particularly at the decision end of data mining, were already in the toolkit of most OR people and was determined to ensure that OR had a major role in this new wave. The OR Society’s Council decided to set up a working group in 2011, of which I was a member, to inform the Society’s response. The outcome was a series of recommendations to Council, including establishing an annual one day ABD event to publicise the key contribution that OR could make; setting up an Analytics network that anyone could join; circulating, widely, a free online quarterly publication; and setting up an academic analytics journal – the Journal of Business Analytics, which was established in 2018. John Hopes led The OR Society’s response to the Analytics challenge by chairing the working group and then, as Vice President, then President, by ensuring that the working group’s recommendations were implemented. I helped John to organise the first three one-day events, each of which attracted well over 100 delegates. Subsequently these have become known as the Analytics Summit and continue to be successful.

“The in-house teams were big enough to develop and maintain the most important scorecards but the more minor ones were updated less frequently and this is where Master’s students, who by this time had acquired most of the necessary skills, could assist, particularly as they were available when many staff would take their holidays"

In the early 2010s, Robert Fildes (Distinguished Professor, now Emeritus and Founding Director, Centre for Marketing Analytics and Forecasting, Lancaster University Management School and former Vice President, ORS) and I looked at how the boundaries of OR practice were changing, as part of a study for IFORS into the state of OR practice globally. We published our results (2) and summarised the key outcomes in a series of articles in Inside OR in 2016 (3), one of which focussed on the impact of analytics. We concluded that OR had the potential to secure a significant role in the development and application of analytics but that some OR consultants needed to improve their skills in advanced statistics, data management and data mining; that successful analytics case studies illustrating the added value of using OR’s methodology needed to be publicised; and that research in analytics needed to be stimulated so as to enable continuous improvements in practice.

In the early days of ABD, there were discussions within the Society as to whether the name of the Society should be changed to emphasise and promote the Society’s commitment to supporting those in the analytics community. Should we become the OR and Analytics Society? I was in favour for two reasons: first because in the UK (and more so in the USA) some OR groups, consultancies and academic OR/Management Science departments were changing their names to Business Analytics or similar for marketing reasons; and second because Operational Research does not really indicate what practitioners do (neither just operational problems nor research!). The proposal proved too controversial to Council and was rejected. One advantage of our current name is that it does not confine OR to an over-specific role, which I do accept is an advantage.

“I promoted the value of OR Society membership to such staff, not only to enable their technical skills to be continually updated but also to understand and embrace OR’s mature methodology that helps to maximise the prospects of implementing the results of analysis.”

However, other than in government and defence, the OR label was not being used by many practitioner groups and some started to use ‘business analytics’ or similar to help clarify their role. As Secretary to the Heads of OR Forum (until May 2022, when John Hopes, a long standing member, took over), I was supportive, as were a majority of members, to extend the name to the Heads of OR and Analytics Forum and this was adopted in 2018. There is currently a discussion about whether this name should be further extended to include data science!

It is gratifying to note that today The OR Society and the OR community have established a key role within the analytics landscape, particularly focusing on decision making, sponsoring workshops, training and the academic journal of Business Analytics.

References

Davenport T.H. and Harris J.G. (2007) Competing on Analytics – the new science of winning. Harvard Business School Press.

Ranyard, J. C., Fildes, R., & Hu, T.-I. (2015). Re-assessing the scope of OR practice: The influences of problem structuring methods and the analytics movement. European Journal of Operational Research, 245, 1-13.

Ranyard J. C. and Fildes R. (2016) Extending the Scope of OR: Business Analytics. Inside OR, issue 550, 18-20.

*Elsie Cropper, who died in service in around 1990, aged 44 years. The Society now manages the Elsie Cropper award, set up by donations from those who worked with her, which provides grants for OR sandwich students.