Case Study #3: Autonomous Audit Drones for Inventory Management (EY)

Background

In 2017, EY³ began using Al-enabled autonomous drones (also known as unmanned aerial vehicles or UAV) to assist with inventory counts and observations as part of its audit services. The drones use Al to navigate a site, such as a warehouse, like a human. As Sean Seymour⁴, Global Mobile Technologies Leader at Ernst & Young LLP explains, “understanding which bin, which rack, which aisle to go to, all of this is based on human intelligence.” By using a cloud-based asset-tracking platform, the drones perceive their surroundings, map areas, track objects through QR codes and barcodes, and provide feedback in realtime. Clients also can integrate the data captured by the drone with their warehouse management software (WMS).

Results

One of the key findings was that drones performed the same tasks as humans with 300 times the efficiency (time and costs savings) than traditional methods. The primary benefits for EY from using drone technology include faster, more comprehensive, and more accurate inventory counts, the ability for staff to spend more time on analyzing inventory risks (vs. the time spent manually capturing inventory counts in the traditional methods), and the improvement of audit quality.

Lessons Learned

One important lesson learned was the importance of having a strong network connection. As Seymour said, “whether that’s Wi-Fi or whether it’s a cellular connection that’s built into a modem on that drone when you think about the size of some of the sites that we work on, they can be 1.5 to 2 million square feet. And inside those areas, there may be zero cellular coverage, and there may also be minimal Wi-Fi coverage as well.’’ Seymour also mentioned that EY is exploring integrating a 5 G modem into the drone to ensure a reliable network connection. “So that then means that we can send all of the traffic directly from the drone to wherever the client wants that data to go. And not only are we going to be potentially covered from any kind of cold zones where there isn’t a reception, but it also means that data analysis can be conducted in the cloud rather than on the actual physical devices that are there. And that means that they can conduct a lot more work”.

Seymour cautions firms contemplating the use of drones to do their research and due diligence before adoption. As he explained, “many people still think of the drones and UAVs as toys.” He pointed out that the FAA heavily regulates drones. Seymour indicated that clients sometimes do not “understand the licensing behind the scenes that are absolutely a prerequisite to using drones in the US.” Violations of FAA regulations could result in substantial fines. Further, firms should be aware of the potential legal liability if a drone were to injure somebody or cause damage to property or goods.

Case Study #4: Use of AI to Augment Auditor Judgment (Deloitte)

Background

Deloitte & Touche LLP utilizes various Al technologies for risk assessment in the performance of audits. According to Brian J. Crowley⁵, Audit & Assurance Senior Manager with Deloitte & Touche LLP, an objective of Al deployment is “to enhance the quality of our audits through augmenting auditor judgment by providing them with additional information relevant to a conclusion (e.g., informing an auditor of possible receivables collectability risks based on the public release of a bankruptcy filing on the internet related to an audited company’s major customer).” By supplementing audit analytics with Al-enabled technologies, Deloitte & Touche LLP professionals can evaluate structured and unstructured transactional data to gain increased insights.

Results

In performing risk assessments, Crowley notes that Al models “suggest and predict correlations among datasets dynamically and tailored to those specific datasets provided, and thus suggest and predict tailored outcomes, risks, and procedures to perform in response. With these capabilities, we are pursuing more granular data at the onset of an audit as part of the planning and risk assessment processes.” Crowley explains that Deloitte & Touche LLP uses optical character recognition and natural language processing to “analyze and extract insights from unstructured written language” to “identify patterns in large populations of documents such that anomalous fields or text are easily identified and summarized.” Further, Crowley notes that Deloitte & Touche LLP recently “introduced additional optical character recognition functionality to process transactional evidence en masse (e.g., invoices, purchase orders, shipping documents, etc.).” Finally, Crowley describes how Deloitte & Touche LLP’s ‘Flux’ tool “uses natural language generation Al models to analyze fluctuations in account balances over time. It provides written prose explanations to an auditor regarding the underlying component causes for those fluctuations, which allows an auditor to more easily gain an understanding of the nature of the account balance, and help to identify any risks that may be present”.

Lessons Learned

According to Crowley, some of the challenges with Al implementation include “optimizing user experience and user interface, ensuring the resulting output is a high-quality audit result, and training people on how to use the tool effectively.” Another challenge relates to developing Al models. “One of the greatest challenges to Al model development is data acquisition,” said Crowley. “Data is required in order to train models to identify patterns that are relevant to a given business objective. More data yet is required to test those models and reach an appropriate level of confidence in real-world conditions. But not just any data will do - data format is important. Most of the data we acquire during an audit is unstructured in nature - manually-created sub-ledgers or spreadsheet analyses, contracts, transactional evidence, etc. It is extremely difficult to run these data sets through Al models at any kind of scale without some extensive pre-processing due to the variability in the format of that data.”