By 2026, 80% of large enterprise finance teams will rely on internally managed and owned generative AI platforms trained with proprietary business data, showed research by analysts at Gartner.

Speaking at the Gartner CFO & Finance Executive Conference, Mark D McDonald, senior director analyst at Gartner Finance Practice, said: ‘The recent entry of large, well-established companies into the generative AI market has kicked off a highly competitive race to see who can deliver revolutionary value first.

‘Leadership teams do not want to fall behind peers; however, as the chief steward for an organisation’s financial health, CFO’s must balance the risks and rewards of tools like generative AI.

‘There are three distinct conversations that CFOs should have across leadership circles to ensure that reasonable expectations are established, and the use of generative AI creates value without introducing unacceptable risks.’

It is important for CFOs to avoid the hype around AI to avoid inflated expectations, especially when presenting adoption ideas to senior management. When preparing a business case, it is critical to define generative AI use cases that are aligned, responsible, and actionable. With a limited regulatory environment writing a policy on acceptable usage should also be high up the priority list.

What to consider when implementing AI processes

Debunk the hype to avoid inflated expectations

Generative AI presents the potential for businesses to comprehensively navigate their data’s growing complexity and volume with ease. However, the technology’s limitations introduce several real challenges to this objective, leading Gartner to consider it at a peak of inflated expectations.

CFOs should partner with senior technology leadership (eg, chief information officer, chief data officer, chief information security officer) to distinguish hype from reality and share results with other executive leadership team members.

Current generative AI solutions represent a collection of modern innovations, including deep learning, natural language processing, reinforcement learning and graph networks, all of which deliver remarkable outcomes. However, the extensive number of parameters and connections used to create these outputs prevent any transparent reconciliation of the algorithm’s response.

Gartner warned that this meant there was an inability to determine if the algorithm has developed unstated objectives or if it is basing conclusions on inaccurate, irrelevant, unethical, or even illegal information.

‘Such limitations form the backbone of conversations that CFOs must have with leadership circles when considering the use of generative AI,’ said McDonald.

Define generative AI use cases that are aligned, responsible, and actionable

With an understanding of generative AI’s limitations, CFOs can responsibly direct a conversation with management teams aimed at defining use cases. They must collaborate with operational management, executive leaders, and representatives from the user community to define actionable generative AI use cases that align with the organisation’s overall strategy and risk tolerance.

‘As with any AI solution, the best use cases exploit a specific business’s strengths and defend its weaknesses,’ said McDonald. ‘Copying use cases from other companies will likely not have the same impact in an organisation with different circumstances. Instead, aligning generative AI’s fundamental capabilities to a business’s unique strategies and objectives delivers a value that differentiates a company from its competitors.’

Develop generative AI governance and guidelines for acceptable use

Generative AI requires human oversight to ensure that outcomes adhere to the nuance of human judgment and fairness. While generative AI’s output may appear human-like and compelling, the results may not always be accurate, unbiased, or reliable.

‘CFOs should engage legal, HR, audit, security, and other relevant corporate support functions to establish usage guidelines to minimise security, compliance, regulatory and other intellectual property risk,’ said McDonald. ‘This discussion must also include the potential impact to the workforce, company culture and necessary training.’

LeackStat 2023