Recent breakthroughs in artificial intelligence have excited many executives seeking personalized marketing, sales boosts, customer demand prediction, and risk identification.

However, the rapid adoption of Google Bard, Microsoft Bing chatbot, and OpenAI ChatGPT has also sparked significant backlash. Critics view AI chatbots as job threats and tools for creating 'deepfakes,' noting that their insights, while seemingly credible, are prone to inaccuracies and bias.

Tesla CEO Elon Musk, Apple co-founder Steve Wozniak, and over 26,000 signatories stated in an open letter released last month that these chatbots pose 'profound risks' to humanity.Open letterThe letter stated that AI developers are 'locked in an out-of-control race to develop and deploy ever more powerful digital minds that no one—not even their creators—can understand, predict, or reliably control.' The signatories called for a six-month pause on developing systems more powerful than ChatGPT-4 and for establishing AI governance structures.

For CFOs weighing whether to adopt AI tools, the challenge lies in seeing through the hype while overcoming anxiety. AI experts and finance executives point out that the challenge is integrating technology into operations and achieving substantial returns without shirking the responsibility to limit risks and ensure investment value.

'The challenge for CFOs right now is that the frenzy around generative AI is creating pressure to scale it broadly,' said Tad Roselund, Managing Director at Boston Consulting Group. 'This commercialization pressure is prompting them to explore all the benefits of generative AI.' He emphasized that even the creators of these tools cannot always explain the reasoning behind insights on core strategic issues like emerging risks, capital allocation, and new market opportunities, so CFOs need to 'demystify' AI tools and identify low-risk, high-return applications. In an interview, Roselund advised that to avoid failure when adopting the latest AI, CFOs should seek 'a balance between urgency and prudence.'

The wave of AI democratization

Generative AI and conversational AI (such as ChatGPT) are forms of machine learning that have greatly expanded access to advanced computing power. These software can write college essays, computer code, market research reports, jokes, translations, blogs, legal documents, and assist in medical diagnosis and drug discovery. Roselund noted that they remove technical and development barriers, achieving 'democratization' of AI, allowing frontline employees to use computing power that was once available only to a few.

Broader access has driven record adoption rates for conversational AI. According to data from DiploFoundation and KPMG, ChatGPT attracted 1 million users inless than a week, while Instagram and Spotify took 10 weeks and 20 weeks, respectively.

CFO Dive, data source: DiploFoundation, KPMG

According to UBS forecasts, by 2025, the conversational AI market will surge to $20 billion, equivalent to20% of total AI spending. Goldman Sachs notes that the potential market size for generative AI could reach $150 billion, compared to $685 billion for the entire software industry. Over a decade, generative AI could contribute a 7% increase to global GDP and boost productivity by 1.5 percentage points.

As AI innovators like OpenAI, Microsoft, and Google compete for market share, the use of AI tools has grown sharply. By offering free or low-cost access, they can collect massive user feedback to improve algorithms and gain an advantage.

New AI tools help CFOs analyze data, make financial forecasts, prepare financial statements, manage risk, and oversee treasury and accounting tasks more easily. Employees freed from routine tasks can shift to more creative and fulfilling work. Adobe, Shopify, Instacart, and Zoom have already adopted generative AI tools. Salesforce announced in March that it plans to integrate ChatGPT into Slack and embed generative AI into its CRM software. Walmart uses chatbots to inform customers about order status and simple issues like returns. Employees use the 'Ask Sam' app for voice queries to locate products, check prices, and view messages or work schedules.

CFOs can also use AI tools to cut labor costs, but this may affect employee morale. Goldman Sachs says AI automation could eventually affect about two-thirds of U.S. occupations to some degree. A list of high-risk occupations highlights the technology's potential to disrupt the workplace. Researchers at the University of Pennsylvania and OpenAIconducted a studyshowing that large language models like ChatGPT pose a threat to workers in dozens of occupations, including accountants, auditors, financial quantitative analysts, blockchain engineers, interpreters, mathematicians, and journalists. The researchers stated that over time, the technology will simplify at least 10% of tasks for 80% of workers and half of tasks for 19% of workers.

Finance executives and AI experts say CFOs should not rush into adopting generative or conversational AI without assessing numerous risks. Some AI tools may end up on the list of criticized innovations—those that cost early adopters dearly while overpromising and underdelivering. An article introducing findings from MIT Sloan Management Review and Boston Consulting Group states: 'AI has the potential to be another technology on that infamous list, especially given so much speculation that AI systems could eventually replace workers.'

Beware of 'shadow AI' risks

AI tools bring multiple risks. First, the proliferation of generative and conversational AI may weaken a company's ability to manage or document AI usage. Employees who keep their jobs will likely gain access to powerful computing capabilities, potentially engaging in 'shadow AI' activities—computing outside company oversight—and intentionally or unintentionally leaking proprietary or customer data.

Second, companies that cannot fully explain how AI tools generate insights may face scrutiny from regulators, legislators, shareholders, and other stakeholders. The 'black box' risk is particularly tricky for asset managers and other AI users with fiduciary duties.

Third, AI tools may expose companies to lawsuits or reputational damage due to 'hallucinations.' They may access flawed or biased data, draw incorrect conclusions, and perpetuate inaccuracies, discrimination, or stereotypes. OpenAI admitted last month that ChatGPT-4 has an accuracy rate ofless than 80% in commercial applications. Google stated in a disclaimer: 'Bard is experimental, and some of its responses may be inaccurate, so please double-check information in Bard's responses.'

Fourth, malicious actors inside and outside companies may use AI tools to improve deepfakes and other disinformation, or for phishing, impersonation, and intellectual property theft.

Regulators in the U.S., Europe, China, and other countries have proposed guardrails for AI tools. UNESCO's 193 member states unanimously adopted a framework to prevent AI misuse. UNESCO Director-General Audrey Azoulay said in a statement last month: 'The world needs stronger ethical rules for artificial intelligence. This is the challenge of our time.' The U.S. Department of Commerce this month sought public comment onmethods to limit AI risks. Alan Davidson, Assistant Secretary of Commerce for Communications and Information, said in a statement: 'Responsible AI systems can bring enormous benefits, but only if we address their potential consequences and harms.'

Finance executives and AI experts say CFOs can limit risks and reap the benefits of AI tools through the following four steps:

1. Ensure 'responsible AI'

CFOs and AI experts say the latest AI tools highlight the need to build safeguards against misuse. Roselund believes: 'You need an enhanced version of responsible AI' and to embed it across all levels of the company. 'Tone from the top is crucial,' he said, emphasizing the need for a senior executive to be fully accountable for outcomes.

As described by consulting firms like Boston Consulting Group and KPMG, responsible AI also ensures technology serves a broad range of stakeholders; mines high-integrity data; defends against attacks and malicious use; protects user data; avoids harming people, property, or the environment; and is explainable, transparent, and reliable. Companies adopting AI should form a cross-functional oversight team including data scientists, lawyers, and heads of various departments. This team should uphold testing and quality control standards, regularly assess risks, and recognize that broader AI use within the company will require greater agility and shorter response times. Roselund said all employees should be empowered to innovate freely within clear boundaries and explore new uses of AI tools. They need to 'think outside the box within the box.'

2. Broaden traditional ROI concepts

CFOs and AI experts say AI tools raise the importance of ROI measurement due to their exceptionally high returns and risks. Wayne Kimber, CFO of SymphonyAI, said in an interview: 'AI—despite all the hype—still needs to prove itself. It must move from being a science project for data science engineers to something of real business value.'

So far, AI returns are far from universal. According to a survey by MIT Sloan Management Review and Boston Consulting Group of1,741 respondentsacross 100 countries and 20 industries, 37% of executives said their companies derived value from AI, while 30% said they did not. The researchers noted: 'Coordinating individual and organizational value from AI remains a work in progress,' while cautioning against burdening employees with tasks that serve machines.

Arijit Sengupta, CEO of Aible, said in an interview that when evaluating ROI, CFOs need to recognize that valuable uses of AI tools will emerge unexpectedly as software analyzes massive data. 'People need to shift their mindset, shifting their traditional approach of setting ROI targets and expectations for projects based on a deterministic world,' he said. 'You can't paint a rosy picture upfront before you actually train AI on data.' Sengupta said CFOs should first focus on increasing revenue—which is relatively easy to measure—to maximize returns from AI tools. Second, they should aim to cut costs and limit risks.

3. Adapt to AI limitations

OpenAI is candid about the weaknesses of its latest software, ChatGPT-4. In areportlast month, OpenAI stated: 'Despite GPT-4's capabilities, it shares similar limitations to earlier GPT models. Most importantly, it is still not fully reliable (it can 'hallucinate' facts and make reasoning errors).' OpenAI recommends 'avoiding high-risk uses without additional context or human oversight.'

Sengupta said AI tool creators will gradually eliminate hallucinations through collaboration with customers on focused applications. He added that Aible's software is designed to highlight errors by double-checking AI tool answers. 'The hallucination problem—use case by use case—will eventually be solved,' he said. 'But before enterprises can use it, this issue must be addressed.'

4. Start small

CFOs and AI experts say that given opportunities to gain competitive advantage—and risks of losing market share—CFOs should avoid delaying evaluation of AI tools. Mukund Kalmanker, head of Wipro's global AI solutions, said: 'Organizations need to move quickly to establish a clear vision and a project-led transformation approach.'

Meanwhile, Kimber believes CFOs can maximize returns and avoid waste by initially focusing on proven AI tools at a limited scale. They should 'partner with companies that have industry knowledge,' he said, rather than some covert 'skunkworks project.' 'We're not telling customers, buy our platform and then find a needle in a haystack,' Kimber said. 'We say, hey retail customer or fintech customer, we've identified use cases that work.'

Finance executives and AI experts say CFOs who take an uninformed approach may push technology into the same 'hype cycle' that has disrupted innovation adoption for decades. 'The hype cycle in technology is well known,' Sengupta said, noting swings in executive attitudes toward expert systems and automated machine learning. 'People get excited, then overinvest, then lose a lot of money, then become disillusioned, and then underinvest,' he said. 'Now we're seeing this with generative AI.'