Sound inventory management is an essential ingredient for success in merchandising and manufacturing companies. The challenge for managers is to strike the right balance between maintaining enough stock to meet demand and limiting excess inventory, which can deplete cash

Applications of Al in Accounting 23 reserves and increase storage and carrying costs. Unfortunately, the requirements for inventory management are highly dynamic, featuring demand variations, seasonal fluctuations, and stock-outs (Patil & Divekar, 2014). For some companies, an outdated or manual inventory management system can make managers unaware of inventory levels and thereby result in the mismanagement of inventory space and people (APS Fulfillment Inc., 2018).

The two critical implementations of Al for inventory management systems are (1) demand prediction for inventory management, and (2) reinforcement learning for full inventory management (Hamilton, 2018). Demand prediction involves the development of a time series model that can forecast demand across all items in inventory by incorporating external data sources such as weather. Reinforcement learning is a more advanced approach where the system not only makes predictions about inventory levels but also takes actions to order inventory automatically.

Amazon is a pioneer in using Al to streamline various parts of its operations. The company has implemented Al for inventory management at an unprecedented scale (Hamilton, 2018). Specifically, Amazon uses Al to forecast consumer demand, supplier backorders, warehouse optimization, and stock level optimization. Al-enabled robots allowed Amazon to increase efficiency and safety at its fulfillment centers and to store 40% more inventory. As a result, Amazon is better equipped to satisfy its Amazon Prime offerings and other deliveries, is less likely to run out of stock, and is able to deliver a faster and more consistent customer experience (About Amazon Staff, 2019).

Al can be used to manage inventory levels in real-time by analyzing large amounts of data from various sources, including ERP systems and the Internet. In a recent blog post, Alasdair Hamilton (2018), CEO of Al research firm Remi Al, emphasized the profound impact that Al can have on inventory management:

“We are on the verge of a major upheaval in the way inventory is managed. This revolution is a result of the availability of the huge amounts of real-time data that are now routinely generated on the Internet and through the interconnected world of enterprise software systems and smart products” (para. 5)

Effective application of Al for inventory management can help organizations optimize inventory levels, minimize storage costs, and increase profitability.

Accounts Payable

The accounts payable (A/P) process involves matching invoices to supporting documents, such as purchase orders and contracts. Matching is necessary to ensure that payments are authorized for the correct amount and in compliance with purchasing contracts and that they are encoded correctly in the general ledger. Many companies rely on an outdated analog system for matching paper invoices to other paper source documents. According to research firm Gartner, only about 10% of businesses worldwide receive invoices in electronic form. The lack of electronic invoices suggests an incredible opportunity exists in A/P to increase efficiency and effectiveness through the targeted application of Al services. Specifically, invoice automation combined with an Al-powered matching engine can dramatically reduce manual work, improve process cycle times, and increase the return of investment (Keck et al., 2019).

Machine learning algorithms combined with optical character recognition (OCR) can be used to automate the A/P process by extracting data from receipts and invoices and then classifying the type of expense in the general ledger (Vordenbaeumen, 2019). Unfortunately, current accounts payable invoice automation (APIA) initiatives tend only to address document digitization, largely ignoring the matching component (Keck et al., 2019). Gartner provides several examples of common invoice mismatches that automation can address, some of which include pricing discrepancies, unknown or not identifiable suppliers, and invoice quantities that do not match the purchase order (Keck et al., 2019).

Invoice automation can be supplemented with a machine learning engine to significantly improve match rates. These enhanced Al systems perform better than an off-the-shelf, rules-based expert systems engine. Al-enabled APIA can perform sophisticated matching when an invoice does not initially have a match.

It is straightforward to realize the potency of Al in A/P. The first step is to receive invoices in digital form. The next step is to automatically match invoices to the purchase order using Al. When an invoice does not have a matching purchase order (PO), such as with utility bills and taxes, Al can learn to approve invoices without POs by analyzing contracts and expense categories that were previously approved for payment without a matching PO. Gartner explains that an APIA solution should not only understand the issue with the invoice but also be able to identify the appropriate person to resolve the problem (Keck et al., 2019). Finally, Gartner recommends that APIA systems should prioritize workflows so that high-value invoices and

Applications of Al in Accounting 25 those close to payment or discount dates are flagged over others to ensure a timely resolution.

Gartner provides a sample of APIA vendors on the market, including Medius, The Shelby Group, Basware, Esker, and AvidXchange. For example, Medius developed an APIA solution called MediusFlow AP. It uses OCR and Intelligent Data Capture (IDC) to digitally capture information such as the purchase order number, amount, and date. It then uses a template to automatically enter the data into the A/P system for processing. According to the vendor, companies that use their solution process up to 97% of their invoices via a fully automated and touch-free workflow (Medius Software Inc., n.d.).

< Prev   CONTENTS   Source   Next >