Technology & Innovation

Artificial intelligence as a trusted business partner

Man using VR device while raising a hand for illustration

Businesses need help in prioritizing data, conducting and governing transactions, and figuring out recommended actions from an ever-increasing flood of financial, operational, third-party data, and situational context. These processes have become even more difficult in our Time of Corona, where non-essential business sites have been closed down. We have culturally shifted from an environment where roughly 5% of US workers worked at home full-time (Source: US Census 2017) and about 40% work from home occasionally (Source: Gallup 2017 study). Now, over 40% of US workers work from home full-time based on Amalgam Insights’ latest estimates as of April 2020.

This rapid decentralization of the workforce has led to the breakdown of informal aspects of decision-making and strategy such as unplanned discussions, sharing of body language and facial cues, the ability to physically visit campus-based operations, and the ability to support corporate decision-making processes in a consistent manner. In this environment, it becomes strategically important for companies to have a shared basis for contextualizing, evaluating, and augmenting complex decisions, which provides a practical starting point for considering the use of artificial intelligence.

Artificial Intelligence is a challenging corporate buzzword to understand. It has existed for decades both in the consumer world through science fiction and entertainment vehicles in the form of robots and disembodied super-human intelligence’s, as well as in the academic space where neuro-scientists and related researchers have been delving into the process of simulating brain activity. But our reality is that neither of these visions reflect the majority of artificial intelligence technologies used today or how they can practically help your organization in 2020. To provide some guidance, I will quickly describe four approaches to artificial intelligence and recommendations for how you should think of these approaches in supporting your finance, accounting, and business process management capabilities.

The Four Classes of AI

“AI” is a challenging term because it gets used to describe four separate approaches of simulating intelligence:

  • reactive machines that ignore the past,
  • limited memory models that use strictly defined data for a limited time,
  • theory of mind approaches that seek to imitate the brain’s ability to learn a specific task, and
  • self-awareness approaches where AI is basically sentient and self-guided.

If you watch certain movies and TV shows, you may think that deploying AI is focused on these third and fourth stages. But realistically, enterprise software and tools are focused on the first two stages, which provide a lot of practical value for automating and accelerating closed-loop processes, matching defined transactions, and providing short-term context for a specific decision. From a practical perspective, today’s versions of AI are predicated on building machine-learning models used to quickly analyze the data. These models are optimized by “training” the model with appropriate data that allows the model to associate specific data with specific results. This training process typically requires subject matter experts with experience either in creating the data or in manually doing the task in question to ensure that nothing is left out.

Reactive machines use predictive analytics to figure out what the next best step or next potential outcome is most likely to be based on your current environment. From a practical perspective, this capability aligns to budget forecasting, process automation, and governance to determine what needs to be done next, next-best-action recommendations across revenue and operations tasks, and eliminating both activities and process steps that are less optimal. Reactive machine AIs are becoming relatively common in the enterprise software world, but it is important to acknowledge what they can and can’t do. In general, they are designed to help employees with an immediate problem rather than to independently learn, improve, and gain new skills. This is, for instance, why the likes of Apple’s Siri and Amazon’s Alexa don’t improve quickly despite having millions of daily users providing feedback: these AIs are not built to improve themselves, but to provide an immediate artificial intelligence-based perspective or response to an immediate request. On the bright side, because these systems are limited, they can be set up and trained more easily than their more complicated counterparts. They can often be provided as embedded capabilities within a larger software platform rather than developed as a set of standalone services.

Limited memory AI use cases are rarer but are starting to be created for business use cases. For instance, a self-driving car will need to maintain a certain amount of memory regarding prior activity to contextualize what will happen next. Some sales and service training tools combine prior performance and tactics with readings of the salesperson’s emotions and language usage to provide a conversation-based context for what should happen next. These systems require real-time monitoring to know what the user just did, real-time access to current data, a solid training process for subject matter experts to teach the AI system how to use data correctly, and a suitable interface to either communicate the results to a person or to enact a response with the appropriate system or machine.

Theory of Mind AI would allow people to interact with an AI and have a relatively open back-and-forth conversation on a specific topic with an artificial intelligence with its own perspectives and ability to understand open-ended questions. Initial attempts include IBM Project Debater, an automated debating machine that has held its own against human debaters. But Amalgam Insights proposes both that this type of AI is still emerging and that the chat-bots and other conversational AIs that are currently available for purchase do not live up to this standard.

And finally, the concept of a self-aware and self-learning AI is interesting, but current technological progress is still years away from providing this type of artificial intelligence to the general public. Amalgam Insights warns that, in the near future, there will likely be more and more hype around “more human” AI. Be aware that these claims likely refer to a Stage 3 approach of Theory of Mind where the AI performs some fragment of intelligence for a specific task which would be referred to as Artificial Narrow Intelligence rather than a self-aware or broadly applicable intelligence that would be an Artificial General Intelligence solution.

With this definition of AI in place, how can your organization actually take advantage of the AI that exists from a practical perspective?

Recommendations for Adopting AI

First, think of AI as a way to standardize the treatment and analysis of high-volume transactions, multi-step processes, and decisions requiring data from multiple sources. It can provide an initial data cleanse to search for anomalies in the financial close and related accounting and finance transactions. AI can also provide a hypothesis around complex challenges with multiple data drivers, such as customer retention or marketing campaign success. Providing this standard to all relevant stakeholders allows everyone to start from a shared perspective in either tackling a problem or in using business data appropriately. This approach is especially important for the remainder of 2020 and early 2021 as remote work will continue to be the norm for 40% of the workforce, including a majority of managerial and executive workers, and companies will need to maintain shared methods and resources for tackling strategic challenges.

Second, use AI in an ethical and purposeful fashion. Before moving through with a project, first look at the AI from a business perspective including executive interests, governance and compliance issues, and the potential for the AI to unintentionally changed or altered from its stated goals through bad data or training. It can be easy to look at AI as a tool that simply gets work done quickly, but this can be a problem if the AI has not been configured to take relevant business perspectives, data, and concerns into account, first. For instance, an analysis of revenue by zip code may, in the process, effectively redline revenue sources by race or gender in ways that may not be legal for your organization. Make sure that your organization fully takes business ethics into account in using AI, since AI tools can scale broadly and can empower unexpected changes.

Third, keep in mind that AI tools are continuing to evolve. While we’re not going to see self-aware robots in the near future, AI is quickly becoming more powerful in its ability to support Limited Memory AI to provide real-time context and feedback, as well as starting to move towards a Theory of Mind approach that will allow for more intelligent and free-form discussions to explore data, documents, and business challenges directly with an application.

Although AI is a foundational pillar of the future of business and technology, our challenge is what author William Gibson has noted, “The future is here—it’s just not evenly distributed.” Although the worlds of finance and business often have some mathematic and statistical training, AI requires additional experience in data training, advanced mathematics, application development, natural language processing and generation, and streaming and video analytics across all departments and verticals to take full advantage of AI. As you start using AI, keep both the evolution of AI and these recommendations in mind to augment your team’s capabilities with AI’s ability to identify and optimize specific processes.