Why CFOs should have artificial intelligence on their minds

Getting smart about AI and ML

The term “artificial intelligence” refers to the concept of making machines smarter, enabling them to mimic human cognitive functions. Machine learning (ML) is an algorithmic branch of AI that involves allowing the machine to teach itself based on the data it receives. As the algorithms derive predictions or insights from the data, and get feedback on the outcomes, they keep refining their capabilities to achieve higher degrees of certainty.

Such an approach is clearly well-suited to the finance function, which routinely relies on large and complex volumes of data, both financial and operational, to fuel its many processes. In the State of AI survey, 67% of respondents reported that they are currently using machine learning, and almost all (97%) plan to use it in the near future. Among executives whose companies have adopted AI, many envision it transforming not only businesses, but also entire industries in the next five years. Using ML to optimize and transform processes, CFOs can automate tasks that typically require manual intervention, improving the accuracy of the accrual process or speeding up account reconciliation and, ideally, eliminating any traces of human bias.

But ML’s place isn’t just in back-office applications, where it can boost efficiencies. By partnering with the commercial side of the business, ML can produce insights and boost predictability, providing increasingly accurate predictions as to, for instance, which customers are likely to make repeat purchases or which suppliers are likely to default.

The algorithms, of course, only know what they absorb from the data—which is based on countless human decisions and a vast array of systems. As such, its knowledge base reflects and projects flaws, ranging from inconsistent data quality to potential human bias. Identifying and eliminating such deficiencies requires ongoing maintenance and testing, subjecting the algorithms to quality control so that, for instance, a bank doesn’t unfairly reject the lending application of a creditworthy individual. ML would know better.

Around the blocks

The technology’s capacity for learning depends on not only the volume and quality of data it receives, but also how well it is aligned with the problem. To lay down a solid foundation for the technology, companies need to assess and mitigate any quality issues involving data, undertaking data-cleansing initiatives to boost integrity and accuracy. Companies that set their expectations high, and find the availability of relevant data low, are setting themselves up for disappointment.

To support AI, any data governance issues also need to be addressed beforehand. Internal wrangling over data (some data owners may be risk-averse about sharing theirs, while others want to keep control over data because of its value) can result in needless delays.

But CFOs who remain focused on realizing the ultimate benefits of ML sooner rather than later—aware that it can free up their teams to spend more time on strategic issues—can see past the initial questions they may have, including:

• How can we fund our AI projects? Taking a cross-functional, integrated approach to ML will likely produce the most value for the enterprise, resulting in a shared decision-making tool. But companies can start with point solutions, aiming the technology at a specific problem rather than investing in a more costly enterprisewide solution. Barriers to entry for AI have dropped significantly, as platforms offering ready-made infrastructure and algorithms have become available.³ If necessary, finance leaders can explore creative funding sources, such as vendor subsidy and ecosystems programs, coinvestment strategies, and other models, to provide funding for technology innovation within finance. Teams can also explore venture capital models to fund AI use cases and to use the outcomes as proof points for further investment.
• Which early use cases are likely to yield a financial return? ML’s self-learning capabilities means it gains value over time. But identifying a specific problem, and defining the desired outcome, can enable CFOs to measure the technology’s impact early on. The greatest opportunity for short-term ROI may lie with streamlining back-office activities, including transaction processing (particularly in shared services). Decreasing labor-intensive, repetitive tasks will quickly and clearly justify long-term investment in AI technology. In the State of AI survey, respondents cited improved process efficiency as the top benefit that AI has enabled them to achieve. The best use cases tend to be function-specific, but should also offer broad visibility, if possible.
• Should we build or buy AI? Finance leaders may want to collaborate with their technology counterparts (CIO, CTO) to determine whether to partner with third-party AI providers, develop solutions internally, or pursue a hybrid approach. In making this decision, finance and IT should investigate other use cases being implemented in the organization and leverage home-grown experience and talent to understand what suits the current environment. Organizations frequently mix bought capabilities and home-grown models. When evaluating whether to expand partnerships with cloud vendors and other providers or to foster new ones, consider if the problem is shared across other areas of the enterprise, and ensure alignment of the organization’s AI ambitions. Is the process you are solving for specific to finance (e.g., revenue forecasting)? Or is it a solution that could benefit other areas as well (e.g., invoice matching)?
• How can we quickly develop in-house expertise? Assessing off-the-shelf solutions and designing realistic use cases requires deep competency in AI. One option is to outsource the technical end to a provider of managed AI services, enabling finance to focus on excavating data out of functional silos. Developing in-house expertise can begin with prioritizing AI-related skills in recruitment and training. It may be helpful to stage a hackathon to solve a specific business problem, using it to identify a group of developers who are interested in becoming ML engineers. By making it part of their job to do so, the company can build a knowledgeable team.
• Why should we trust AI? The notion of intelligent machines replacing—and overpowering—humans is strictly the province of blockbuster movies. CFOs should evaluate their current operations and identify opportunities to update the operating model, ensuring that the right people and capabilities are in place to maintain an AI-fueled finance function. Employees embedded in transactional tasks may need to retool themselves, developing stronger analytical skills. The finance function’s overall skill set may shift, but not necessarily its population of FTEs. Humans with the right training need to manage AI systems, leveraging the appropriate capabilities.
  

Artificial intelligentsia: Humans working with machines

In the past, many CFOs have led campaigns to incorporate analytics, particularly in FP&A. But the stakes and rewards involved in championing ML are much higher, given its potential use across the enterprise. And while uncertainty surrounding the pandemic continues, the fact that ML is not rules-based (there’s no limit to its ability to evolve) means it can change and adapt as the next normal takes shape. Similarly, finance leaders can lay the groundwork for the technology by taking piecemeal steps, such as:

1. Inventory internal capabilities. AI technology, including ML, is function-agnostic, so before CFOs plant their flag, they ought to check to make sure other functions haven’t preceded them. Marketing may have begun using it to understand how to better retain customers in a virtual environment. Supply chain may already be relying on it to help anticipate production or inventory disruptions. If significant internal capability exists, finance can get a jump-start by piggybacking on those other functions’ efforts.
2. Choose a data strategy. ML benefits from receiving a very complex set of inputs and analyzing how that data fits together. Data often ends up residing in functional silos, where it isn’t rich enough to inform the ML model and help with decisions. Finance needs to come up with a strategy for breaking down those silos—or design a limited pilot project that maximizes the data within a silo, such as classifying sales data for use by finance and commercial teams (see “The CFO guide to data management strategy,” Crunch time series, Deloitte Development LLC, 2020).
3. Establish a cross-functional task force. Since just about all functions can leverage ML, it’s best treated as a collaborative, iterative process. It’s vital that whatever initial issues ML is deployed to solve can produce results that will have a strong impact on the business. Furthermore, any such tests should be on the radar of all members of the senior leadership team. Experimentation may result in success or failure—or something in-between, depending on the different perspectives that are brought to bear.
4. Build an AI center of excellence. Depending on the company’s AI maturity level and the breadth of its capabilities across the organization, it may choose to establish a finance AI center of excellence. Such a resource can help spread the technology throughout the business, evangelizing for AI, deploying resources for pilots, leading training, and advising on hiring. Such a COE could be launched as a subset of an existing continuous improvement function within finance or within an innovation group that sits in the CTO’s office.
5. Confront fear of AI. Finance executives may instinctively resist the appeal of ML, as many have previously reacted to other mammoth IT projects. They may feel that they aren’t sufficiently knowledgeable or conversant to engage their peers in a dialogue about AI. But through cross-functional partnering, CFOs will gain fluency around AI, which will translate into confidence and a regular seat at the table when it comes to tech strategy.

No matter how they initially react, most CFOs can learn to appreciate what ML can cut well. Yes, implementing it successfully may well produce some misses, likely the product of misalignment and data quality issues. That’s why it’s so important for CFOs to approach AI projects with humility regarding the company’s talent and technology. It’s an emotion that separates humans from other species—including, presumably, even the most intelligent machines.