Expand +



Make Machine Learning Work for Your Business

Top Image Systems on Extracting Value

July 20, 2017

Andrew Pery, Chief Marketing Officer, Top Image Systems, discusses Machine Learning with Ken Murphy of SAPinsider. Below is a transcript of the conversation.

Ken Murphy, SAPinsider: Hi, this is Ken Murphy with SAPinsider. Thank you for listening to this podcast on machine learning and what it means for the SAP customer. Here with me to discuss this topic is Andrew Pery, who is the Chief Marketing Officer for Top Image Systems. Andrew, thank you for joining me today.

Andrew Pery, Top Image Systems: Thank you so much for this opportunity, it’s my pleasure.

Ken: I was hoping to start with a definition of machine learning. Tell our listeners a little bit about what it is and what it means.

Andrew: Simply put, machine learning is a form of artificial intelligence (AI). And by virtue of this the computer application itself is able to learn without explicit programming. So for example machine learning technology has the ability to discover certain patterns in data, and based on it analyze new information. Now when we talk about machine learning, typically it falls into two categories. The first is what’s defined as supervised machine learning, and by virtue of this the program is trained by feeding example documents from which the computer software learns. So this is generally referred to as learn by example machine learning. And so for example in the context of invoice processing, the software learns the layout of each invoice that’s in the system, and then compares it to a knowledge-base of previous invoice layouts. So then the system learns the layout and structure of supplier invoices, and then can automatically can extract relevant invoice line item information and for example match it against the SAP ERP master data without any human intervention. And in doing so providing very high recognition rates. When we look at invoice automation in large measure it’s still very much a highly labor intensive and error-prone process, so machine learning can really reduce the transaction cost and accelerate these processes within an SAP ERP environment.

The second form of machine learning is what’s defined as unsupervised machine learning, in which case the program analyzes large datasets from which it finds certain patterns and relationships within that data set. An example of this is predictive analytics that uses data to infer sort of future activity and behavior trends. And this form of machine learning is very useful within for example the services industry that detect consumer behavior, look at their interests which may be a predictor of buying preferences. So for example in an SAP environment for supply chain management applications that can really help in determining supplier preferences and being able to gain much better visibility to supplier relations. So machine learning really helps in reducing transaction costs but at the same time helps management to gain better insight.

Ken: And why is machine learning gaining traction in the enterprise today?

Andrew: Simply put, the growing volume and velocity and variety of information. Just to give you an example, on a daily basis organizations generate 2.5 exabytes of data globally. If you look at what is an exabyte, it really equals to 1 billion GB of data every day. That’s an enormous amount of volume of information that comes from multiple sources. In order for organizations to harness the information they’re looking to leverage an internet machine learning technology for three reasons. The first is to convert this large volume of data into a competitive asset. So machine learning can help organizations gain better insights into leading market trends and customer behavior. It helps the CFO for example to gain better visibility into a working capital and cash flow that can analyze information more readily and therefore plan for the future. The second is equally important and it’s to mitigate potential risks. So today organizations are faced with a large array of regulatory requirements including security and privacy legislation that requires organizations to really preserve and protect information so as not to expose themselves to potential litigation or fines as well as corporate reputation. So risk mitigation is certainly a very important driver for the adoption of machine learning technologies.

And then lastly is reducing transaction costs. So particularly when you look at the finance function, organizations are looking to become more agile, more streamlined and also to accelerate these business processes so when they capture purchase orders or invoices or supply chain management, typically these are driven by paper-based, paper-intensive processes so machine learning technology can significantly reduce these transaction costs.

Ken: What then are some of the typical use cases for machine learning?

Andrew: There are several, and I’d like to highlight a few in particular. The first is digital mailrooms. A digital mailroom is really a process by which all incoming documents that come to an organization from multiple channels, be it paper or fax or email or PDF attachment are digitized at source as soon as they enter the organization. And so machine learning technologies is used to essentially intelligently capture and extract and classify all this incoming information from multiple channels and automatically determine what type of document they’re dealing with and be able to differentiate for example if the document is a customer complaint, or is it a purchase order on the invoice or contract and by virtue of this high level of intelligent automation they’re able to accelerate the routing of information for example to the right organization for resolution. If it’s an invoice, then automatically it gets captured into the SAP ERP environment. If it’s a contract, then obviously it’s also integrated with SAP for supply chain management and being able to then accelerate those business processes. In the case of for example customer response management, to be able to more readily and faster respond to customer inquiries. So that’s one example of intelligent capture.

The other example is obviously invoice processing, or invoice capture. And in this case we’re talking about managing large volumes of invoices and typically today more than 50% of organizations still utilize paper-based methods and means to capture invoices which tends to be highly labor-intensive and error-prone. So machine learning technology allows organizations to automate these processes, capture the invoice line item information, match it against the ERP master data, post that information as well as of course automate the entire capture to payment lifecycle so straight through processing of those invoices which obviously significantly improves service levels, improves access and visibility into cash flow and of course reduces transaction costs.

Ken: How then do companies derive value from these use cases?

Andrew: There are three areas where organizations derive value. The first is reduced transaction costs and reduced operating costs to take out from the business highly labor-intensive and error-prone manual types of activities as they relate to for example invoice capture and invoice processing. Anything that’s document-intensive to automate those processes.

The second area where organizations derive benefits is improved visibility to the business. Better insight into supplier relations then enables the treasury to extend for example preferential discount rates to highly valued suppliers, the ability to gain better insight into working capital, plan better for the future and also to look at how the organization is able to take additional costs of the business. So for example if you look at invoice automation and invoice processing, best-in-class organizations that have machine learning technology implemented and fully automated accounts payable processes can reduce their invoice processing costs by as much as 80% so when you look at a typical manual process the average cost of processing an invoice is just over $15. Using machine learning technology and fully automated straight through accounts payable processing that cost can be reduced to as much as $1.50 to $2 an invoice. In terms of better visibility into the business, ability to have timely access to information and reports and analytics enables organizations to reallocate highly labor-intensive resources to more effective, higher value transactions for example dealing with exceptions, dealing with cash flow management, better allocation of working capital. So the result of that is organizations become a lot more agile.

Ken: How then does Top Image Systems deploy machine learning technology?

Andrew: That’s our core competency. We’ve been in this business of intelligent capture, document-process automation, machine learning, since 1991. Our core competency is to empower our customers to capture and transform incoming business content from multiple channels into digital data that enables them to better leverage that information and accelerate their business processes. So by providing this advanced machine learning technology and process automation, we’re able to deliver to our customers superior recognition rate, accuracy in the capture and collection and management and processing of this information, and we help customers along three dimensions. The first is to onboard information much more effectively and efficiently so that when all business content that comes into the organization from these various channels, we automatically capture and organize this data and make it understandable and digitally actionable. So we’re accelerating these business processes and helping companies become much more efficient.

The second is to accelerate customer engagement by automating these highly labor-intensive processes we do reduce transaction costs and enable organizations to improve customer service level. And thirdly, as a result of this we have companies become a lot more agile. One of the areas we’re particularly focused on is delivery of a complete, integrated accounts payable solution within the SAP environment whereby our process automation solutions for invoices lives within the SAP familiar user interface that enables the accounts payable organization to have single point of access to all of the invoice and purchase order related data, the ability to initiate within SAP workflows to post this incoming invoice data and information to be able to match it against the SAP ERP data and the ability to post that information and ultimately approve payment. Or in the case of escalation to be able to route that information to the appropriate resource for resolution. And of course for the treasury and the CFO to then gain better visibility into the status of invoices to look at the volume of invoices that they produce, the throughput, and also to understand where they stand in terms of payments to suppliers.

Ken: Can you share an example for how your machine learning capabilities deliver value?

Andrew: One of our largest customers is called the Bosch Group. It’s a company in Germany and this company has 450 subsidiaries in 60 countries with a total annual turnover of over €70 billion so it’s one of our larger customers. It’s an SAP-based accounts payable implementation and they process more than 450,000 invoices per month. So it’s a very large volume of invoices and they have over 100,000 suppliers across 15 different jurisdictions and countries. So it’s a highly complex invoice processing solution within the SAP ERP platform and within this environment we’ve fully automated the capture of the supplier invoices with 450,000 invoices that are coming into the organization on a monthly basis. And to a large measure, especially for PO-based invoices, we’ve fully automated this process. By virtue of this, over 40% of all incoming PO-based invoices are processed automatically using machine learning intelligent capture and workflow technology without any human intervention. So it’s a straight through process from capture to posting to payment. Which of course significantly reduces transaction costs, improves the recognition rate, and enables the organization to really reduce and take out significant costs from the invoice capture process. And by virtue of that significantly reduce the transaction costs and at the same time accelerate payment of invoices to suppliers, it enables them for example determine suppliers that might be eligible for early payment discounts in which case if they’re able to extend those early payment discounts it enables the organization to realize higher returns on invested capital, because there’s no sense to hang onto cash. So this is I would say probably one of the more useful examples where our solution is fully integrated with the SAP ERP environment and provides significant benefits.

Ken: And lastly, Andrew, where can a listener go to get some more information on the topic?

Andrew:They can go to our website, And they can go directly to the accounts payable landing page, as well as resource pages where we have information about our solution, in particular case studies that would reinforce the value proposition as well as I would encourage listeners to go to our blog page where we provide thought leadership around our solutions with a particular focus around accounts payables and financial process automation that the audience might be interested in. Ken: Andrew thank you for joining us today. Andrew: Thank you so much for the opportunity. I really enjoyed it, thank you so much.

An email has been sent to:

More from SAPinsider


Please log in to post a comment.

No comments have been submitted on this article. Be the first to comment!