The use of artificial intelligence (AI) using machine learning (ML) will be the single most important trend in business software this decade because it can multiply the investment value of such applications and provide vendors an important source of differentiation to achieve a competitive advantage in what are today very mature software categories. I assert that by 2025, almost all Office of Finance software vendors will have incorporated some AI capabilities to reduce workloads and improve performance. However, software vendors will be challenged to apply innovations in this area quickly while ensuring that the AI capabilities function well enough in the real world to foster rapid adoption while avoiding user frustration. The failures of the Apple Newton and Microsoft’s Clippy office assistant stand out as examples of too-ambitious-too-soon attempts at infusing intelligent automation.
Of course, AI is already at work in a lengthening list of use cases. For example, recommendation engines are common in consumer-facing services such as retail or entertainment. Navigation applications “learn” the best routes for time or distance at specific hours of the day, while spam filters automatically adapt to evolving attempts to evade them. Road warriors routinely use their smart phones to take pictures of receipts and expense-management applications convert the images to business-useful data used for automating travel and entertainment expense processing.
Because AI conjures up images of computer systems that are entirely independent of humans, some technologists prefer to use the term "augmented intelligence," which are systems designed to enhance the capabilities of the humans employing them, especially in improving decision-making and eliminating the need for an individual to perform the rote parts of a process. This is more in line with the intent of what is meant by AI, even if people use the term artificial intelligence.
Here, in summary form, are some of the AI-powered capabilities for finance and accounting departments that either exist in early-stage form or will be the ones introduced earliest in the evolution of AI:
- Analytics, including predictive and prescriptive analytics, constraint-based planning and goal seeking.
- Forecasting with driver-based models that highlight serious divergences from plans in real time and the source of the divergence.
- Anomaly detection to highlight possible outliers and inconsistencies in planning, analysis and reporting.
- Recommendations for planning, task completion, best fit, staffing and so on.
- Task supervision which looks for errors and omissions in data entry to build data quality into processes and prevent the need to fix mistakes while reducing auditing requirements further along in a process.
- State supervision which supports process owners without the need to spend time creating rules-based systems to highlight late events or process anomalies, as well as provide alerts on an incident or condition to facilitate management by exception.
- "Autofill" task management (capabilities just shy of full task automation) for repetitive routines that drive even complex and cross-functional/cross-application workflows.
Software vendors will be adding to the list of AI-supported capabilities continuously in coming years. To foster adoption, they will need to provide business users and executives, especially those in finance organizations, with a clear understanding of how AI is or will be relevant to them in their business role. To keep customers informed, vendors must provide a clear roadmap for incorporating specific AI capabilities and describe the business benefits customers can expect from these additions.
AI capabilities by themselves will not be enough. Historically, one of the biggest impediments to the use of AI using ML in business applications has been the challenge of having clean and consistent data to enable rapid and accurate training of these systems. This remains a major challenge but one that can be addressed.
Approaches to unifying data management in an organization, including the concepts of "data fabric" and more recently "data mesh" architecture (there is considerable debate about these terms and their exact definition), along with more robust APIs, RPA and productive data-cleansing automation, make it easier and more economic to assemble the clean, consistent data sets necessary for the reliable use of AI.
The idea of having a single, virtual management layer on top of distributed data has been extended to enabling distributed groups of teams to handle data to fit their needs. This is in line with a general concept in information technology of systems moving toward more loosely coupled structures to be more malleable and adaptable. This approach lends itself to what I have called (tongue in cheek) the "digital aggregation device" (DAD) or "data pantry" that is associated with a specific business application. It is a pantry because all the data ingredients that anyone needs for a specific function or task are easily found, not stacked high in some ginormous warehouse, with labels that are instantly recognizable and readable to those that need to work with the data. In geek speak, this means the semantic layer is designed for a specific application or use case, not a general-purpose metadata label that makes sense mainly to the IT professionals who created it.
Data pantries are especially well suited to vendors using a platform architecture and will be an essential element in making ML workable for a specific application. Especially so if there is a need to incorporate third-party data in algorithms to achieve greater accuracy in, for instance, sales, marketing and financial analysis and forecasting software. Organizations that rely solely on internal data for informing decisions or providing analysis of conditions or events that involve the outside world will almost certainly fall prey to serious errors caused by endogeneity, especially omitted variables. SaaS software vendors can include third-party data sets or offer them at additional cost as a subscription, greatly facilitating their use.
Looking at software aimed at the Office of Finance, AI is not going to put robots in charge of the finance and accounting department anytime soon. It is doubtful that there is a "killer app" using AI on the horizon that will fire up imaginations. Increasingly, though, it will take the robotic work out of an individual’s day and enable them to be more productive and utilize their skills and experience more effectively. However, there is a need for caution in introducing AI capabilities in business software because business routines are inherently more complex than services aimed at consumers, and the negative consequences of bad advice in commercial decisions are often far greater than a bad movie selection. For that reason, the introduction of AI capabilities and features will be incremental and may therefore appear unremarkable. Although the impact of these improvements at the individual level will be hard to measure and may be imperceptible, it will be significant in an overall sense, even in the short run. For example, a couple of minutes here and there multiplied tens of thousands or even millions of times over the course of a year will have a profound impact on productivity. Fewer errors as well as a richer and more accurate set of data will boost effectiveness as seen in customer service, billing, payments and organizational performance.
Although skepticism is certainly in order in evaluating vendors’ claims, organizations must develop their understanding of AI and its use cases (existing or potential) to be able to make best use of capabilities as they become available. AI is not just for data scientists, and finance and accounting departments must be a fast follower in adopting this technology. Beyond the positive impact on productivity and likely enhancements to the quality of their decision-making and work product, eliminating the drudgery of back-office work will enable organizations to attract and retain the best talent, especially the younger cohorts who increasingly will not put up with wasting their time on antiquated methods.