It’s difficult to ignore the hype around Big Data. But does it actually offer any ROI? Malcolm Wheatley investigates three ways it can boost the bottom line.
There’s a buzz about Big Data – that much is undoubted. And it is hard to ignore.
Where has it sprung from? A happy combination of advances in low-cost computer processing power, low-cost data storage, powerful analytics techniques, and a ready availability of the raw source material – large volumes of data.
Yet for all the impetus gathering behind Big Data, many manufacturers will be tempted to regard it simply as hype. Why? Because for manufacturers – as opposed to mass retailers, financial services companies, utilities and government institutions – it’s difficult to see the ROI.
But that doesn’t mean to say that there isn’t any. For while mining the world of Big Data isn’t appropriate for every manufacturing business, it would be wrong to assume that Big Data doesn’t have offerings for the manufacturing industry’s mainstream.
What might those offerings be? The picture is still evolving. But it’s already clear that three distinct avenues offer opportunities that many manufacturers will find it worthwhile investigating.
Better still, they’re genuinely new advances – not ‘me too’ conventional marketing and finance approaches given a Big Data spin to qualify for a place on the bandwagon.
Despite the rise of Just in Time and buildto- order business models, many manufacturers are still heavily reliant on demand forecasting.
And in the world of computerbased demand forecasting, not much has changed since the 1970s and 1980s. Traditional ‘time series’ techniques simply extrapolate the past into the future: trends in past sales, the impact of past seasonality, and the effectiveness of past promotions. They’re all carried forward to produce a calculated estimate of future demand. Different methodologies and algorithms attempt to refine the process, but at the heart of any MRP or ERP forecasting module lies familiar well-honed techniques such as exponential smoothing, moving averages, and Box Jenkins models.
No longer. A technique known as demand sensing is turning this logic upside down and is credited with improving the statistical forecast error of product demand forecasts by 30%-40%. Importantly, those improvements are attested to by corporate users of the approach, not just by vendors selling it.
Unilever, for instance, trialled a demand sensing solution from specialist provider Terra Technology in North America, and then took the decision to roll it out in Europe as well, says Fabrizio Bortolotti, the company’s European planning director. “More accurate forecasting and better inventory management complements our lean manufacturing strategies, allowing us to capture growth opportunities and optimise service to our customers without the risk of carrying excess inventory,” he notes. And demand sensing delivers just that by incorporating a much broader range of demand signals, in as near real-time as possible. Simply put, demand sensing takes the traditional forecast as an input, but adds to the mix real world events such as market shifts, weather fluctuations, changes in consumer buying behaviour, social network sentiment, and real-time point of sales data.
All of which adds up to a lot of data – Big Data – which, coupled to advanced analytics software, allows manufacturers to do much more than merely get a better handle on overall sales volumes.
“More and more companies want to use downstream data,” says Andrew Spence, supply chain business development director at Oracle. “Demand sensing allows them to aggregate volumes of downstream data, and apply analytics to make meaningful decisions – demographic by demographic, store by store, and distribution centre by distribution centre.”
“Demand sensing allows manufacturers to aggregate volumes of downstream data and apply analytics to make meaningful decisions – demographic by demographic, store by store, and distribution centre by distribution centre” – Andrew Spence, Supply Chain Business Development Director, Oracle
Forecasting techniques aren’t the only thing left unchanged from the 1970s and 1980s. Scratch a typical manufacturer, and you’ll often find that the frequency of MRP runs is also little changed.
Granted, monthly MRP runs are generally a thing of the past. But fortnightly and weekly runs remain very much a feature of the present.
Why so? One of the original reasons for such frequencies was that computers used to take all weekend to churn through the calculations. Although this is no longer the case, MRP runs can still take many hours.
Other reasons for infrequent runs include the need to suspend order entry and workin- progress updates. Then there’s the sheer practicalities of revising the business’s procurement and production planning schedules on a continual basis.
But SAP’s much-heralded HANA in-memory technology finally threatens to upset the apple cart, by bringing together two aspects of supply chain and manufacturing management previously forced to co-exist separately: supply chain and manufacturing planning, and supply chain and manufacturing execution.
“Our vision at SAP is to realize the real time supply chain. But the closer you move to the real time supply chain, the more the distinction between supply chain planning and execution blurs. It’s about planning better, and executing better – at the same time,” says Hans Thalbauer, senior vice president of business solutions for supply chain at SAP.
“We have already moved our advanced planning and scheduling and sales and operations planning capabilities onto HANA, and putting demand sensing and inventory optimisation on the same platform, using the same data model, makes perfect sense.”
Better still, adds Adrian Simpson, chief innovation officer at SAP UK, a manufacturer can also schedule multiple MRP runs – utilising different demand scenarios, different price points, and different inventory management assumptions. And it can all happen incredibly quickly, so you can say goodbye to lengthy MRP batch runs.
“What we’ve recognised is that there are some business processes that are very time-consuming, because of their nature,” he says. “An MRP run can take eight hours, which doesn’t offer much scope for optimisation and analytics. That’s the opportunity that in-memory processing provides.”
“Plant-floor data is a constant stream rather than a series of transactions and arrives too fast for a traditional database to handle. Data volumes of 25,000-35,000 ‘tags’ or data-points in 200 milliseconds are quite common” Sue Bailey, Software Consultant, SolutionsPT
At Wrexham-based PET plastic container manufacturer APPE, a plant-floor data historian application is credited with helping the business increase output by 25%, with fewer people. What’s more, product quality has increased as well, says APPE’s continuous improvement champion Tim Manuel.
Sourced from specialist industrial automation provider SolutionsPT, the Wonderware plant historian constantly tracks machine temperatures, pressures, setpoints, statistics on machinery downtime and spoilage, and helps the company monitor KPIs such as manufacturing effectiveness, unplanned downtime and cycle loss.
While plant historian applications aren’t new, manufacturers outside specialist process environments have often been slow to recognise the value of the insights that they can provide. The APPE installation is the first, large volume plant data historian implementation to unite a high speed data acquisition and storage system with a traditional database management system, says Sue Bailey, a software consultant at SolutionsPT.
“Plant-floor data doesn’t look like the data that you’ll find elsewhere in a manufacturing business, and is in volumes that you don’t find elsewhere, too,” she points out.
“Plant-floor data is a constant stream rather than a series of transactions, and arrives too fast for a traditional database to handle. Data volumes of 25,000-35,000 ‘tags’ or data-points in 200 milliseconds are quite common.”
Mark Dunleavy, UK managing director of data integration specialists Informatica, which numbers among its clients 67 of the top 99 Fortune 500 manufacturers, says manufacturers are generally slower technology adopters than other industries, but are increasingly aware of the benefits of exploiting plant-floor data.
And while improvements in operational efficiency are an obvious target, a growing body of opinion advocates using plant-floor data to predict imminent equipment breakdown, prompting targeted – and timely – preventative maintenance. For in the run up to a breakdown, equipment often gives off warning signs, albeit in the form of signals that are buried in the mass of data that typifies most plant floors.
Gauges on machinery such as gearboxes, for instance, can detect changes in vibration levels, oil temperatures, and pressures. Special microphones can detect noises inaudible to the human ear, while sensors installed in machinery can methodically count usage- or cycle-time based wear patterns.
The problem is not collecting the data – plant historians and similar applications handle that requirement – but analysing it and spotting the tell-tale signs of imminent breakdown before a breakdown actually occurs.
The good news? Such capabilities do indeed exist, with providers such as IBM, SAS and Accenture all offering combined analytics and consulting capabilities.
“Manufacturing companies pay out millions of pounds per year in downtime costs because of unplanned maintenance on machinery,” sums up Bob Finney, head of the analytics business unit at high-performance data specialists OCF.
“Previously, there was no way of avoiding this, but with the use of predictive analytics based on Big Data platforms, the early identification of maintenance requirements is genuinely possible.”
Taken together, the bottom line-enhancing power of such advances is certain. From demand sensing to targeted preventative maintenance, and from improved operational efficiencies to lower inventories and better scheduling, it’s difficult to argue that the impact will be anything other than positive.
Most manufacturers, though, will have a more prosaic concern: Can I afford it? But here too it seems there’s welcome news – though you may have to wait. Such capabilities aren’t yet built into ERP systems but according to Gordon Fleming, chief marketing officer at QAD they will be one day.
“I see ERP evolving to deal with most common Big Data situations,” he says. “That’s not every Big Data problem that is out there, but those that are most commonly experienced, and which offer the highest ROI. The result: enhanced affordability – and arguably improved easeof- use.”