Accountants have embraced waves of automation over many years to improve the efficiency and effectiveness of their work. But technology has not been able to replace the need for expert knowledge and decision-making. Indeed, previous generations of ‘intelligent’ systems have generally demonstrated the continuing power of human expertise and the limits of machines.

Artificial Intelligence (AI) is already an integral part of many of our lives, and with increasing use of Edge’s Copilot or ChatGPT, these algorithms will continue to become embedded in everything we touch and do.

These systems do not replicate human intelligence. However, on a task-by-task basis, systems increasingly produce outputs that can far exceed the accuracy and consistency of those produced by humans.

AI may bring many opportunities for accountants to improve their efficiency, provide more insight and deliver more value to businesses. In the longer term, AI raises opportunities for much more radical change, as algorithms increasingly take over tasks currently done by humans.

The framework for embracing AI in accountancy includes understanding the technology, its limitations, and exploring real-world examples of AI usage.

• Find out more in ICAEW’s explainers on AI and Machine Learning.

What is the long-term vision for AI and the profession?

AI systems will likely take over more and more decision-making tasks from humans. While accountants have been using technology for many years to improve what they do and deliver more value to businesses, this is an opportunity to reimagine and radically improve the quality of business and investment decisions – which is the ultimate purpose of the profession.

The profession needs to focus on the fundamental business problems it aims to solve and imagine how new technologies can transform its approach to them.

Focus on purpose

Accountants want to help organisations and economies work better by giving good advice and making good decisions. All the activities associated with accounting ultimately aim to help people make good decisions about the allocation of resources and hold others to account for their decisions.  This underpins investment, growth and confidence in all organisations and economies.

More intelligent systems enable radically different approaches to this ultimate objective and the kinds of fundamental business problems the profession aims to solve. Investors need to gain confidence and trust in the financial results of companies; companies and governments need to ensure the correct levels of tax are paid; management needs to decide how to spend organisational resources and be held accountable for those decisions. Solving these kinds of fundamental problems is essential for companies and economies to succeed and are at the heart of the accountancy profession.

Exploit powerful technologies

It is important to recognise and exploit the power of new technologies effectively. The strength of machine learning and deep learning approaches to AI are regularly leading to major advancements in many areas. Other areas of emerging technology will interact with AI and have a significant impact on business in the future, including blockchain or quantum computing.

To fully exploit powerful new technologies, we need to be clear about their unique characteristics and how they can help to solve real problems. Often, technology can be a solution looking for a problem to solve, or simply something which enables us to replicate how we already do things using different tools.

Instead, we need to encourage debate, interaction and learning between technology experts, business and the profession to reimagine the way that we solve fundamental business problems with the help of new technologies.

Think radically

The profession also needs to be open to more profound change and avoid just defending or incrementally improving the status quo. Where AI enables greater insight from data, it helps human experts make better decisions and provide better advice, where it automates high volume tasks it can free up time for more strategic thinking. 

However, as these systems continue to get more powerful, they will be capable of moving into more complex decision areas. This potential might result in replacing humans altogether in many cases and enabling entirely different solutions, services and models. When looking at the longer term, therefore, the profession must think beyond incremental improvements to existing processes. Furthermore, organisations need to reflect on the specific skills and qualities that accountants bring to businesses. This goes beyond technical knowledge to incorporate qualities such as professional scepticism, ethical decision making, gaining and applying insight from numbers, and ensuring that numbers can be trusted.
This also means engaging positively in debates on, for example, the role of human judgement in more complex business areas. There may well be uniquely human characteristics, such as leadership, empathy and creativity, which might never be replaced by computers. 

We should not underestimate the adaptability and ingenuity of humans. However, ‘human judgement’ is often just a substitute for lack of data – powerful computers with access to new sources of data may well supersede the need for human judgement in most cases. 

Be adaptable

The skills and learning agenda for the future accounting professional has been subject to a lot of debate between professional bodies, including ICAEW, employers and educators. Most would agree that accountants will need more hard skills in areas such as technology and data, as well as a greater emphasis on things like soft skills, critical thinking and adaptability. 

There also needs to be greater emphasis on life-long learning. This is also reflected in ICAEW's Continuing Professional Development (CPD) Regulations, which brought in new CPD requirements from 1 November 2023. The changes give greater clarity on the amount of CPD that accountants should be doing to ensure up-to-date technical (such as data and AI), professional and business skills. 

• Visit ICAEW’s Finance in a Digital World CPD hub or our webinar catalogue for CPD resources relating to the use of technology in the profession

How do artificial and human intelligence work together?

AI systems can be very powerful and are improving quickly. They provide outputs that can be extremely accurate, replacing and, in some cases, far superseding human efforts.

However, they do not replicate human intelligence. We need to recognise the strengths and limits of this different form of intelligence, and build understanding of the best ways for humans and computers to work together.

Human decision-making

AI systems provide accurate outputs and can surpass human efforts, but they do not replicate human intelligence. Humans make decisions through intuition and reasoning, with intuition being quick and unconscious, and reasoning being conscious and logical. Accountants use both intuition and reasoning in their decision-making process.

Intuitive thinking is powerful but subject to biases and inconsistencies.

Approaches to AI

Early AI systems focused on replicating human reasoning such as representing knowledge and codifying logic-based rules. In essence a top-down approach where the model would act in the way it was built to act. These systems attempted to capture the rules knowledge of experts but were limited by the complexity of the real world.

Recent advancements in AI, particularly in machine learning and generative networks, take a bottom-up approach and create their own rules based on observation. By combining approaches in machine learning with developments in other areas of AI, such as knowledge representation and reasoning, computers can be used to complement and increasingly improve on both ways of human thinking.

Strength of machine-learning

Machine learning techniques tap into human cognitive strengths and enable breakthroughs in natural language processing, machine vision, and generation of content.

This approach could also enable computers to move far further into decision-making processes than was previously possible, when they were defeated by the complexity and ambiguity of pre-defined rules. Indeed, research in areas such as medical imaging and diagnosis increasingly shows machines producing far more accurate results than humans do.

While there is nothing new about some algorithms performing better and more consistently than many experts, AI systems ‘turbo charge’ this capability and potentially lead to much more powerful decision tools than have previously been possible. This reflects three features about models and the algorithms they contain, making them valuable for analysing big data.

• AI systems can process large data volumes
• pick up complex patterns
• Generate outputs faster than humans

Limits of machine learning

Data quantity and quality is fundamental, and not all problems have the right data to enable the algorithm to ‘learn’. Many models require substantial amounts of data to be trained. The big breakthroughs in areas such as computer vision and speech recognition rely on very large training data sets – millions of data points.

Although that is not the case with all areas of machine learning, success depends on having sufficient data of the right quality. Data often reflects existing bias and prejudice in society. Consequently, while models can potentially eliminate human biases, they can also entrench societal biases that already exist if not governed properly.

Furthermore, not every problem will be suitable for a machine learning approach, which includes the use of generative models like ChatGPT. For example, there needs to be a degree of repeatability about the problem so that the model can generalise its learning and apply it to other cases. Additionally, many machine learning models are trained to give you the most likely answer based on the training data which means outputs may not be accurate or misleading (these are known as hallucinations). For unique or novel questions, the output may be far less useful or even be ‘hallucinations’.

How are accountants using AI capabilities?

Although AI techniques such as machine learning are not new, and the pace of change is fast, widespread adoption in business and accounting is still somewhat in its early stages. The release of generative language models like ChatGPT and Copilot have meant that use of AI models has become widespread for low-risk use cases. The accessibility of these tools has prompted much more interest in the technology and its use.

To build a positive vision of the future, we need to develop deep understanding of how AI can solve accounting and business problems, the practical challenges and the skills accountants need to work alongside intelligent systems.

Accounting problems

Accountants apply their technical knowledge about accounting and finance to help businesses and stakeholders make better decisions. To support their decision-making and advice, accountants need high quality financial and non-financial information and analysis.

Accountants have been deploying technology for many years to help them provide better advice and make better decisions. Technology can help them do this by solving three broad problems:

• providing better and cheaper data to support decision-making;
• generating new insights from the analysis of data; and
• freeing up time to focus on more valuable tasks such as decision-making, problem solving, advising, strategy development, relationship building and leadership.

The very nature of machine learning techniques lends itself to substantial improvements across all areas of accounting, and can equip accountants with powerful new capabilities, as well as automating many tasks and decisions.

Therefore, it is important to identify accounting and business problems where machine learning is likely to be particularly fruitful and where problems may be less suitable for these techniques. This will ensure that adoption efforts are driven by business need, rather than simply technology capabilities. To date, there has been limited use in real-world accounting but early research and implementation projects include:

• using machine learning to code accounting entries and improve on the accuracy of rules- based approaches, enabling greater automation of processes;
• improving fraud detection through more sophisticated, machine learning models of ‘normal’ activities and better prediction of fraudulent activities;
• using generative models to facilitate interactivity with businesses internal policies, or faster analysis of contracts;
• using machine learning-based predictive models to forecast revenues; and
• improving access to, and analysis of, unstructured data, such as contracts and emails, through deep learning models.

Practical challenges

Transactional accounting data is well-structured and high quality, and therefore should be a promising starting point for developing models. However, given long-standing issues around data in many organisations, especially those with complex and unintegrated legacy systems, this is still likely to be a major challenge in practice. Firms will need to think systematically about their IT systems, software, governance and internal skills to effectively utilise AI.

A more principled limit will be privacy and ethics, especially where AI systems are drawing on personal data. Fraud detection, for example, may draw on the text of emails sent by employees, which will encounter legal and ethical limits.

Adoption will also be ultimately driven by the economics and business case around AI. This will reflect two different ways that organisations will adopt machine learning capabilities.

First, machine learning is increasingly becoming integrated into business and accounting software. As a result, many accountants will encounter machine learning without realising it, similar to how we use these capabilities in our online searching or shopping activities. This is how smaller organisations are most likely to adopt AI tools.

Second, conscious adoption of AI capabilities to solve specific business or accounting problems will often require substantial investment. While there is a lot of free and open source software in this area, the use of established software suppliers may be required for legal or regulatory reasons.

Given the data volumes involved, substantial hardware and processing power may be needed, even if it is accessed on a cloud basis, it has been found that most of the emissions for ChatGPT are not related to the training of the model but to prompts that occur after.

As a result, AI investments will likely focus on areas that will have the biggest financial impact, especially cost reduction opportunities, or those that are crucial for competitive positioning or customer service. Other areas, while potentially beneficial, may lack a strong investment case. Likewise, using machine learning to develop more intelligent products in specialist accounting areas may lack the market potential to justify investment from software developers.

Roles and skills

Organisations will also need access to the right skills. Clearly, this starts with technical expertise in machine learning. But, as with data analytics, these technical skills need to be complemented by deep understanding of the business context that surrounds the data and the insight required.

There will also be new jobs. For example, accountants may need to get involved in projects to help frame the problems and integrate results into business processes. Other accountants may be more directly involved in managing the inputs or outputs, such as exception-handling or preparing data, or using models to analyse corporate reports, or generate these reports.

This evolution will be reflected in the skills required of accountants. Some roles, such as training models, may require deep knowledge of machine learning techniques. In other areas, accountants may just need a more superficial knowledge of machine learning to be able to have informed conversations with experts and other parts of the business. Critical thinking and communication skills are likely to become increasingly important.

In addition to skills, accountants may need to adopt new ways of thinking and acting to make the most of machine learning tools. For example, different ways of thinking will be needed for spending more time on predictive and proactive activities, such as  putting predictions in context, or building capabilities to change course quickly.

Institutional issues

Accounting has a wider institutional context, and regulators and standard setters also need to build their understanding of the application of AI and be comfortable with any associated risks. Without this institutional support, it is not possible to achieve change in areas such as audit or financial reporting. Therefore, the active involvement of standard setters and regulators in these areas is essential.

For example, standard setters in audit will want to examine where auditors are using these techniques to gain evidence and understand how reliable the techniques are. Such bodies are already debating the impact of data analytics and machine learning capabilities on audit standards.

Another issue in this context is the interpretability and explainability of models. We do not fully understand how the most complex models – such as deep learning models – derive their outputs. If organisations and audit firms increasingly rely on black box models in their operations, more thinking will be required about how we gain comfort in their correct operation.

Regulators can also actively encourage and even push adoption where it is aligned to their work. Much of the investment in this area, for example, is coming from financial services organisations to support regulatory compliance and pressure from regulators.

Next steps in ICAEW’s work on AI

ICAEW’s future work on AI will focus on building understanding of the practical use of AI across business and accounting activities. In addition, it will lead and encourage wider debate about the long-term opportunities and challenges.

ICAEW’s aim is to provide a strong platform to build and share understanding of the specific application of machine learning technologies. We also support other stakeholders in the profession who need to understand the capabilities and issues here, including:

• educators and training providers, who are considering the future skills of accountants;
• regulators, who are considering the risks attached to new technologies; and
• governments and policymakers.

ICAEW will actively explore ways in which it can help the profession to think more radically about a future working with AI and translate innovative ideas into practice.