Smart machines are here to stay. Not to replace finance executives, but to make their processes more intelligent. In fact, artificial intelligence (AI) is blazing one of the fastest adoption curves of any new technology since the World Wide Web in the mid-1990s.
True, its use is not yet mainstream in finance. However, The Hackett Group research shows a big spike in intent to adopt AI technologies, particularly the approach known as machine learning (also called cognitive computing). Those who believe AI is an overhyped distraction, or something that only data-rich enterprises like Google and Facebook can use practically, are putting their organizations at risk. Competitors will proceed and leave them behind.
Machine learning in the finance context
According to The Hackett Group’s 2018 Key Issues Study, finance organizations expect to see a 1.7 times increase in the mainstream use of machine learning over the next two to three years, albeit from a low base. But 44% already report piloting or limited adoption of it, and 66% expect to have pilots or limited adoption by 2021 – a 50% increase.
Source: The Hackett Group 2018 Key Issues Study
Machine learning is most often deployed in advanced analytics applications. Companies want to profit from the insights they believe are hidden in their data and use smart machines to figure it out. How they approach the process can be categorized into four related families, based on output type:
- Predictive economic forecasting, which is based on detection or discovery, and leverages machine learning’s pattern-recognition capability. With it, companies can discover potentially valuable information or categorize content amid the general clutter of Big Data.
- For example: A multinational company is using an AI-enhanced forecasting model to bolster the accuracy of its economic forecast. The company relies on a third-party tool that scans hundreds of social media networks globally for over 100 keywords that it deems important for potential future financial and economic events. It identifies the number of occurrences of these words and analyzes the intensity of the conversations around them, based on the number of posts related to each. Using that intensity, it continuously learns about the level of importance of the conversations. Using a “heat scale,” it ranks the probability of a series of possible events – for example, the likelihood of new elections or financial crisis in any given market within the next three, six, nine, or 12 months. Those probabilities are fed into finance’s predictive forecasting model for cash-flow and investment decisions.
- Demand forecasting, which is derived using machine learning for statistical association of observed outcomes with defined patterns.
- For example: At one telecommunications company, finance leverages machine learning to assess the probability of customers buying certain products at certain retail stores at specific locations, based on a trove of data about past purchasing patterns and assumptions about future demographic changes. By putting the data in a Hadoop environment and running it through smart machines, the finance team can discover what kinds of products it should hold in inventory in each store. Providing the individual stores with information about what to stock significantly reduces inventory cost and improves store profitability.
- Predictive credit-risk management, a solution that leverages a machine’s ability to pull together data from multiple sources in multiple formats and detect patterns and potential risks by learning from the data.
- For example: A large global resources company wanted to reduce its exposure to counterparty credit risk (and reduce the time it takes to make decisions by making them more objectively). Finance deployed an AI predictive model that pulled information from a variety of internal and external sources. These included the company’s own database of customers’ historical payment patterns, external data on industry economic situation, and macroeconomic as well as publicly available financial data on each customer. Using these diverse inputs, the predictive model produces an intelligent risk rating and assigns a probability of default to each counterparty. When finance combines that probability with the size of the counterparty exposure and its own recovery-rate assumptions, the model’s risk rating can be used to support decisions about the amount of credit to extend to each customer and on what terms.
- Process execution data mining, or how intelligent systems can help finance monitor other systems’ performance and alert professionals to any breakdowns or bottlenecks. The staff can then quickly resolve performance issues and maintain steady and timely processing.
- For example: One home-improvement company uses process mining to identify breakdowns and bottlenecks in real time in processes such as account-to-report. The staff can then attend to them and resolve them quickly. The way process mining works is by pulling event logs from all IT systems, running them through an analytics engine, and providing a real-time visual map of the processes. The application then prioritizes issues based on which ones are having the biggest impact, and provides prescriptive recommendations on how to fix them
Today’s artificial intelligence systems deploy advanced machine-learning software with extensive, self-adapting behavioral algorithms. These machines are literally getting smarter, improving their skills and usefulness based on their access to large data sets. Machine learning will continue to evolve its presence in finance, enhancing the role of finance experts.
For more on this topic, see AI: The Potential And Implications For Finance Leaders.