Machine Learning And ERP’s Autonomous Future
March 13, 2017 Alex Woodie
What makes Netflix so good at predicting movies you’ll like? A recommendation algorithm based on your viewing history, and the viewing histories of others like you, of course. While consumer technology is rife with such technology, HarrisData president Lane Nelson sees a future when ERP applications augmented with machine learning algorithms can automate nearly all of the back-office decisions currently made by people.
Machines have been taking over the back-office since the days of the typewriter. But the next wave of innovation occurring around big data analytics and machine learning will likely make the transition to a people-less office nearly complete, according to Nelson, who’s also holds the title of chief evangelist at HarrisData, the Brookfield, Wisconsin-based provider of enterprise software that runs on IBM i.
“Automation is coming at the back office. It has been for 10 years at least,” Nelson tells IT Jungle. “It’s not that we want to put our customers out of work. But information robots are going to do to the back office what robots did to the plant.”
HarrisData is in the early stages of building those “information robots” that will eventually do much of the work that office workers do today. Those bots will pull data from the ERP system and external systems to build a model of how companies run their businesses. After watching a CEO or a VP of sales or a finance manager make enough decisions, the machine learning bot will be able to replicate the decision, given the same variables, and offer up a recommendation to the decision maker.
That’s the general way that Nelson sees machine learning impacting ERP systems. The bots will start by emulating the most pedestrian processes in ERP systems.
“You can start with things like payment processes out of AP, or MRP kinds of processes,” he says. “You’ll have lots of sensors along the way, capturing lots of data that’s happening. You capture it. Then throw it at the machine learning, and then it knows what a normal process looks like and you know the other factors going on.”
The bots will start out as alerting systems that will detect anomalous conditions in the business. “You can start to apply machine learning to say, tell me as early as possible when it goes off the rails so I can take action to put it back on or reschedule,” Nelson says.
Eventually, the bots move up the chain of command, and start advising managers on day-to-day decisions that are typically made by people in the workflow. “I’ve got a stack of cash and a bunch of bills to pay. Who should I pay?” Nelson says. “What do you want to order for procurement? Which things should you commit to? Who should you buy from? Who should you forward on to a collection agency? Which customers should you offer free shipping to?
“These are a lot of little tactical things that they start to hit on,” he continues. “You could do this with a decision tree, but it would be better if you routed it through a machine learning engine where it would say, based on all the things I know – including soft things like how their stock price is doing or whatever you want to throw in there – this is what you would typically recommend.”
The actual data science work – gathering the data, normalizing it, building models, tweaking models, retraining models, and eventually scoring real-world production data to generate recommendations – would not occur in the ERP system, but in an adjacent system running on-premise or on the cloud. The IBM i platform is an exceptional transactional processing platform, Nelson says, but Hadoop runs well on Power Linux machines.
“I think the ERP system ought to be focused on the transaction processing,” he says. “If you’re going to do an analytical database under the brand of your ERP, terrific. But technologically, that belongs in a different spot.” (SAP, with its HANA platform that unified analytics and transaction processing, obviously would disagree.)
Nelson says there’s a huge amount of free or cheap data available that companies could use in their models to augment the bots’ decision-making capabilities. “The public Internet has a bunch of stuff and if you want to get some crawlers to go out and grab more information, you grab it and tag it to vendor ID and throw it in the pile,” he says. “On the whole that’s not hard to do. But I don’t think ERP has to be a host for that stuff, necessarily.”
Eventually, the ERP system is making decisions in a somewhat autonomous fashion. “We’re more inclined to present a list with check boxes next to them: ‘Here’s what we preselected for you,'” Nelson says.
With permission from the managers, the algorithms will automatically make decisions based the training data from human decisions-makers inside the company. At some point, customers may subscribe to models based on data obtained from other companies, and the recommendations made by the models would reflect the training data from human decision-makers at those other companies.
“The model can also be fed with how are other midsize companies interacting with these vendors, or these products,” Nelson says. “How are those factors affecting their decisions? People that are privately held, also from the Midwest, also the same size as I am in my industry, and you get closer and closer matches.
Once that happens – and it won’t happen without explicit permission granted from the companies – then it will create interesting feedback loops in real time decision making.
“Are you willing to kick back and let your business run on autopilot the way everybody else is? God I hope not,” Nelson says. “Or are you going to take those normal decisions on 80 percent of what you do, and you want to focus on 20 percent that’s interesting so that you can do what’s different? I think that makes sense in a lot of places.”
While the big data analytics industry is moving fast, the machine learning technology won’t find its way into ERP systems overnight. But as the core place where business processes are housed and executed, ERP systems are ripe for disruption by machine learning tech.
Lane Nelson is president and chief evangelist at HarrisData
“We’re excited about seeing that [machine learning] technology grow and about getting those hooks into the ERP because we think the ERP, the info processes and business processes, should be automated, except when things go off the rails and you need to fix them, much like a car plant is,” Nelson says.
Nelson says HarrisData is working with a handful of machine learning tools to solve “detection and advisor kind of problems.” It has started instrumenting its human resources app with the hooks and workflows and interfaces needed to add machine learning capabilities at some point down the line.
“Now we’re at point where we’re capturing the data, we’re starting to play with the models and the feedback loops so we know what we want to deploy as a general purpose solution,” Nelson says. “We really are getting ready to close the loop on this, then we’ll roll out ERP next.”
So when can HarrisData’s IBM i installed base expect to see an ERP system that uses machine learning to automate 80 percent of the decision-making in the enterprise? “We want to play in the lab a little longer. I don’t want to rush stuff,” Nelson says. “It’s coming. I’m not going to say it’s going it’s here next year or the year after that. But it’s coming.”
RELATED STORIES
What IBM i Shops Should Know About Digital Transformation
Unwinding Python’s Data Science Potential On IBM i
