The challenge of gaining visibility and predictability across its multi-billion dollar agency payments business is a thing of the past for Australia Post thanks to analytics technology.
The postal organisation tackled the issue of agency payments cash flow as part of a technology transformation aimed at automating accurate monthly, daily and annual forecasting.
Traditionally, monthly cash flows forecasts were a manual, time consuming process, manager of business systems and development, Armand Mizan, told the audience at the recent IBM Smarter Analytics Conference in Sydney.
There was also inadequate intelligence across short and long-term forecasting, an inability to obtain accurate daily forecasts in a timely fashion, and little visibility of the movement of agency cash flows. Australian Post is an agent for organisations such as the Australian Tax Office, telecommunications and utility providers, and receives payments across its 4,500 retail outlets and online portal. These payments go through an Australia Post bank account and are re-distributed back to the agency.
Agency payments might be a one-line item on the budget, but the business is worth $50bn per year and daily fluctuations have a significant impact on cash flow. Gaining reliable insights which could be used to manage this more intelligently was critical in Australia Post’s overall quest to gain accurate financial forecasts.
To meet the challenge, Australia Post brought together IBM’s SPSS predictive analytics solution with the group’s Cognos TM1 enterprising planning software.
In order to check and determine the accuracy of the information in SPSS, historical data from August to October 2011 was collated and put it through the software to produce a modelling forecast.
“That forecast was compared to the actual data from 2011. We achieved an accuracy level of between 95 and 98 per cent using SPSS,” Mizan said. “This proved the model was worth using. A one per cent improvement on $50bn in cash flow has significant benefits.”
Australia Post now has fully integrated actuals and forecasted monthly cash flows for the next four years. The monthly and daily cash flow models are also aligned, and visibility of agency financial data has been achieved. Four people worked full-time for four months putting the SPSS/TM1 project together, but there was a long parallel run before the solution was pushed live.
However, Mizan admitted the combined SPSS/TM1 project was not without teething problems. For example, agency settlement dates information from the general ledger reflected the transaction date. However, this was different from when the cash settlement went through.
“When we are doing cash flow forecast, we needed that settlement date,” Mizan explained. “In order to overcome that, we applied lags to transactions to arrive at a settlement date.”
The second challenge was the impact state and territory public holidays and special events on have on customers paying their bills. The SPSS software now takes this into account using a calendar of holidays.
Mizan also cited gaps in historical data, along with recognising different trends across different types of agency payments as key learnings during rollout.
While it’s early days in terms of measuring the ROI of the new solution against costs, predictive analytics is firmly on the agenda. The next steps in its quest to improve forecasting and financial performance is predicting customer sales by day, product profitability, price volume models, market basket analysis, customer churn in the parcels area, customer segmentation.
Other unchartered waters ahead include text analytics, Mizan said.
“What you need to think about is solutions – it’s not about going in with a pre-determined technology set,” he advised. “You need to understand the capabilities and use the best tools to get the job done right. In our case, a combination of technologies gave us the outcome we needed.”
It’s also important to remember that successful predictive analytics modelling is about data manipulation. Getting this data into a usable format often represents 80 per cent of work. Once data is in a form that’s usable, SPSS provides the smarts to aggregate and segment data, devising a unique model which can then be deployed to address the specific business pain point.
“The more analysis and more data, the better the outcome and the increase in accuracy,” Mizan added.
3 tips from Mizan on successful analytics:
- Buy-in and sponsorship of c-level management is essential. “This ensures all analytics projects align to the strategic objectives of the business,” Mizan said.
- Align to specific business requirements. “Whatever initiative you embark upon, understand the business requirements and pain points being experienced. If you misunderstand those requirements, you won’t get great take-up.”.
- Take control. “You must take ownership of hierarchies, the quality of the data, testing, communication, accountability and so on because without ownership it’s hard to get a model that meets all the expectations.”