At its best, predictive analytics provide the executive team with tools to better understand customer behaviour and market trends and improve a company’s business intelligence system.
Predictive analytics bring together statistical analysis, data modelling, real-time scoring, and machine learning to observe trends and project them into the future. Used in scenario modelling, these projections are helpful in recognising variables and making judgements on likely outcomes.
Within the finance function, effective data analytics and performance improvement methods are becoming increasingly popular. They enable accountants and finance professionals to be more proactive.
Gary Cokins, CEO of US advisory firm Analytics-Based Performance Management and author of the CGMA book Strategic Business Management: From Planning to Performance, offers three examples of how predictive analytics is changing finance reporting.
Combined with activity-based costing, which assigns overhead costs to products, data analytics allow for a more encompassing view of channel and customer profitability.
Data analytics have allowed the development of strategy maps to report and monitor both financial and nonfinancial key performance indicators (KPIs).
Traditional cost-centre budgeting and cost variances control are giving way to driver-based rolling financial forecasts using predictive analytics integrated across business processes.
With predictive analytics, executives, managers, and employee teams can better see future demands, such as the volume and mix of products needed to meet sales forecasts more accurately, according to Cokins. He explained that, as a result, the executives can adjust their resource capacity levels and types, such as the number of employees needed or spending amounts. They also can quickly address small problems before they become big ones. And they can transform their mountains of raw transactional data into information to test hypotheses, see trends, and make better decisions.
Where predictive analytics are used
There is now widespread use by companies of data analytics to anticipate mechanical problems, with the benefits of reducing maintenance costs and machinery downtime.
Other applications of predictive analytics may be less obvious. Analysis of rainfall and temperature has been used by wine producers to predict with a high level of accuracy both the quality and quantity of a season’s output. This, in turn, can help decisions on pricing and predict income.
Predictive analytics are also being applied in workforce management, enabling human resource professionals to anticipate their staffing needs and the remuneration packages required to employ the right number of people with the right skills at the right time.
Fifty-eight per cent of HR professionals in a recent survey by the Reward & Employee Benefits Association said they consider predictive analytics to be a game changer in their area of work.
“Employers recognise the value inherent in the data they have, particularly if they can use it to predict which benefits will have the greatest positive impact on their workforces, or how reward strategies correlate with indicators such as performance, leadership effectiveness, or profitability,” Debi O’Donovan, director of the Reward & Employee Benefits Association, a UK professional organisation, said in a statement. However, she added, to fully maximise the benefits of predictive analysis, employers need to invest in technology: Nine out of ten employers rely on generic software such as Excel, limiting their ability to manipulate data.
Know the limitations of your models and data
David Jayatillake, ACMA, CGMA, is head of business intelligence and analytics at subprime lender Elevate Credit’s London office. Until mid-2017, he was senior pricing manager at UK payment processor Worldpay. He has used predictive analytics to assist with setting pricing and for profitability forecasting, with accuracy that has proved more reliable than with traditional methods.
However, he warned that data analysis has limited capacity to predict outcomes where markets are going through change, “so you must use it with a pinch of salt,” he said. “You might be predicting sales volumes with a particular customer and initially be correct, but then if you start doing more or less business with that customer, then the predictions go wrong.
“Getting the right data to use in a clean format is always a challenge, and you should always strive for good data, not perfect data — it doesn’t exist,” he added. “You have to be very specific in building the right models: Models work best in scenarios with a narrow scope.”
Jayatillake, who was a data analyst before he began studying for his CIMA exams, suggested the best way for management accountants to start using predictive modelling is to learn how internal data structures work and how to use them.
“You have to know where your data is, how to access it,” he said. This could involve learning some new skills. For example, learning Structured Query Language, a standard language for assessing and manipulating databases, is very helpful in extracting data. Once you have done that, you need to understand the data, Jayatillake said, and then decide where you think the value in the data resides.
When modelling, you should start simple. Most management accountants use Excel, so start with Excel, Jayatillake said. It allows for techniques such as time series analysis and multiple linear regression that are very useful in finance.
The DataCamp website is very helpful; it allows anyone who has little to no experience in coding to learn, he added. You should start with the same models that you tried with Excel, and then move on to more complex nonlinear models if you feel they can add value.
Putting predictive analytics to work
As the name suggests, predictive analytics can only be used with problems that have an element of predictability to them, and there needs to be a relationship between the available dataset and what you’re trying to predict.
For example, income statement transactions tend to be more predictable over time than balance sheet items. You also need lots of good quality data from which you can learn.
To decide whether predictive analytics is the right method for a particular problem, you first need in-depth knowledge of the business and the available data, said Rhodri Davies, a director on PwC’s assurance innovation and technology team in the UK.
Here are Davies’s recommendations for getting started with predictive analytics:
- Understand your data; research and identify anomalies in the data; identify risk factors in the data; mine new sources of data.
- Recognise the differences between relational and nonrelational databases. Relational databases have data structures that allow linking information from different types of data buckets. Nonrelational databases contain data stored without structured mechanisms to link data from different buckets to one another.
- Use exploratory multivariate statistics, inferential statistics, visualisation tools, optimisation methods, machine learning, and predictive analysis tools. Process-mine using new data analysis techniques and algorithms to isolate and investigate specific processes that might have led to changes to the data/accounting ledgers.
- Use simple vendor risk dashboards and filters to minimise inefficiencies and human error. Perform data and process mapping from a regulatory and risk-assurance view.
- Based on the data analysis and insights, communicate what decisions and actions are required in ways that highlight added value.
— Paul Gosling is a freelance writer based in the UK. To comment on this article or to suggest an idea for another article, contact Sabine Vollmer, an FM magazine senior editor, at Sabine.Vollmer@aicpa-cima.com.