How AI redefines reasonable assurance for audit

artificial intelligence
istock_f3d_ai
Brought to you by
mindbridge
Share this content

Artificial intelligence is transforming the audit industry. By providing auditors, fraud examiners, and forensics experts with the ability to ingest, analyse, and report on vast amounts of ledger data, firms can provide a higher degree of risk assurance for clients.

In his book, The Fourth Industrial Revolution, founder and executive chairman of the World Economic Forum, Klaus Schwab, states that a shift will occur when 30% of corporate audits will be conducted by AI, which 75% of people surveyed expect to happen by the year 2025.

It’s already underway, with the Big Four firms making significant investments into AI projects and regulatory bodies discussing its impacts. AI will change the audit landscape by creating a new definition of reasonable assurance.

Improving assurance with AI

Since it’s impossible to assert with certainty that an event, such as a financial misstatement, will or will not occur, auditors have operated under the philosophy of “reasonable assurance” for decades. With too much data, too little time, and limitations inherent in the internal controls of clients, auditors cannot provide absolute and guaranteed assurance on their findings, requiring the use of professional judgement to fill in the gaps.

Both auditors and clients operate on the principle that audit evidence is more persuasive than conclusive, and are guided by the Generally Accepted Auditing Standards (GAAS) to provide reasonable assurance that the assessed financial statements are free of material misstatements.

With AI, the growing amount of client data and scarce amounts of an auditor’s time are less of a concern as the work is offloaded to applications that operate at much higher speeds than people. Additionally, AI improves the testing of ledger data by going beyond traditional sampling techniques to identify misstatements based on risk analysis rather than traditional audit rules.

Big data, no problem

Previously, the only feasible method for analysing large quantities of ledger data was to take small random samples and forego the time and effort necessary to pore through it all. For years, this has been deemed “good enough” and both auditors and clients have accepted the fact that anomalies outside the sample set are potentially missed.

This is known as sampling risk, or the risk of reaching a different conclusion by examining a small set of the data rather than the entire data set. This risk increases when the potential misstatements are rare, including those buried within larger patterns of activity.

Some modern financial analytics tools can relieve a few of these limitations but they often require technical skills and programming knowledge that are not part of an auditor’s core skill set. This makes adopting these tools a cumbersome and potentially time-consuming process. AI offers auditors the ability to focus their judgement and expertise on the entire set of data without requiring these specialised skills or data science expertise.

Identifying new types of risk

The very nature of AI means that it is constantly learning about and identifying patterns in data. As these systems learn more about client transactions, they are able to analyse secondary data and cross-correlate hundreds of variables to establish the transactions that seem correct, and those that require further investigation by a person.

One example is a classical AI method called “expert system,” drawing on the knowledge of real-world professionals and practices to identify unusual patterns. By working with audit practitioners to understand suspicious and risky transactions, an expert system can be built that knows hundreds of account interactions and their associated concerns.

These methods democratise human knowledge at scale, and can be quickly run on large amounts of client data to highlight issues that might be missed by traditional methods.

AI can also identify rare items using empirical methods, leveraging the science of determining what is usual versus unusual. Called “unsupervised learning,” or inferring patterns in data without having labelled or known outcomes, these methods can detect outliers in transactions without bias or history, letting the data speak for itself. These methods go beyond traditional auditing rules or what a human is able detect.

Both methods are especially important when it comes to the ICAEW guidance on material misstatements: “Consider and make enquiries of management about factors which might lead to increased risk that the financial statements may contain material misstatements or be non-compliant such as significant, unusual or complex transactions or events.”

It’s important to understand that AI is not displacing the role of the auditor rather AI works alongside people to automate large data analysis tasks and provide new insights. By enabling auditors to dive deep into 100% of the data, clients can be advised on their financial health and compliance with more comprehensive evidence and greater confidence in risk assurance.

About John Colthart

John Colthart

John Colthart is General Manager, Audit and Assurance & V.P. Product Management at MindBridge Ai. He strives to lead clients to effectively use analytics to change the course of their business

.John has made clients successful in every major market worldwide during his 17-year career in technology, leading world-class sales and professional services organisations. 

This started after his departure as a corporate finance and accounting practitioner in 2000 so he could grow a startup to over 425 employees and exit to IBM with the role of VP Sales Operations in 2010. During his stay at IBM, John held global roles running sales enablement, offering management and design leadership within the IBM Analytics division. 

Replies

Please login or register to join the discussion.

avatar
02nd May 2019 11:47

Interesting.
How times have changed.
I remember a senior colleague stating, many years ago, that if people wanted a 100% audit, it would take 12 months. Apparently not the case now.

Thanks (0)
avatar
15th May 2019 14:31

Are you saying the reasonable assurance approach will no longer be valid John?

Thanks (0)
avatar
to AuditFuture
16th May 2019 21:35

My perspective is that the current approach to reasonable assurance will need to change, for some reasons that aren’t even about the current technical capabilities.

In the market today all of the accountancy bodies are talking about the expectation gap - meaning the differing of understanding of what an audit is by the public, the audit committee, the corporate executives and the external audit teams. Some expect any fraud will be found, some see it as a compliance check, while others see it for what it currently is - a level of assurance that the financial statements as presented by management are free of material misstatement.

What I am suggesting is that with the advent of technical capabilities, and in response to this gap, the current process and agreement on what reasonable assurance will change. Many audits procedures and tests completed today are greatly improved by the state of the art tech capabilities augmenting the human capacity. Artificial Intelligence is just one of these ways that will enhance or replace traditional CAAT tools and processes to redefine this assurance. A major step that has taken place is the ability to read 100% of the entries and transactions, classify the outliers or anomalies, and present an easy to navigate way to assess them as an accountant or auditor. This by itself could change the definition vs the human coping mechanism of sampling continuing in its current common implementation.

Thanks (0)