By Mike Baccala, Assurance Innovation Leader, PwC and Ed Ponagai, Principal, Finance Consulting, PwC
Earlier this year, we made 8 predictions about artificial intelligence (AI) in 2018. They’re all of particular importance to the finance function because—unlike autonomous cars and cancer cures—AI is ready for actual production in finance. Indeed, the finance function is one place we had in mind when we wrote: “AI will come down to earth—and get to work.”
By “get to work,” we mean that AI can drive improved productivity and decision-making today. Fifty-four percent of executives say AI solutions have already increased productivity, according to our research. Even more agree that AI will eventually augment their data-driven decisions. It’ll take an adjustment on the part of a finance team, as we noted in our last post, but the payoff will be right in front of them.
Artificial intelligence has a bright future in the modern finance function. Finance teams spend a lot of time wading through the data stored in a myriad of systems, from ERP and payment processing to business intelligence and financial reporting. Finance staff spend a lot of their day performing mundane tasks — from processing transactions, to exporting reports from multiple systems and reconciling the data.
AI can make all of that smarter. It can perform many transactions automatically — for example, by “reading” the content of an invoice in an email and pulling the relevant information into a payment processing system. That AI will never have the experience and institutional knowledge of a veteran corporate controller. But by running them through a large number of transactions, the algorithms can be trained to make smart inferences, helping finance to automate reconciliation processes or improve cash flow models.
AI can also alert the CFO to emerging trends and anomalies, not only making the function more efficient, but also improving risk controls and data-driven decision support. Banks, for example, can use machine learning and agent-based modeling to determine what an optimal balance sheet would look like for the coming quarter. By setting AI-informed targets related to different sections of their balance sheet, while also considering liquidity ratios for the overall business, they can optimize operational decisions such as credit card promotions or prices — all while staying within the institution’s appetite for risk.
“Big data” was added to the Oxford English Dictionary in 2013, preserving in history the importance of data as a transformative force for organizations. Yet many corporations have been underwhelmed with the ROI on their big data initiatives.
Big data’s real payoff will come through artificial intelligence. Data is the key to making AI effective. The more data that is standardized and labeled (and “cleansed” of unintended bias) the better. For example, machine learning software can be trained to replicate the thinking and decision-making of expert auditors as they review a general ledger. It can examine every transaction, user, and account to find unusual transactions, without bias or variability. This is possible when a provider has access to comprehensive data sets and experienced auditors to train the algorithms.
Similarly, a company’s business units might have experienced data analysts with their own FP&A systems, each with its own proprietary formats and data definitions. An AI that could make sense of all those systems could examine more scenarios and improve its forecasts. But it would need access to the business units’ data in standardized form. Just think of all the hoops analysts jump through right now to pull data from multiple ERP systems, and manipulate it so it can be fed into their FP&A spreadsheet analyses. Multiply those headaches by a factor of ten and that’s the big data challenge finance teams have faced until now.
An even smarter system could then bring external data into the planning process. A home improvement retailer, for example, could improve store forecast accuracy by factoring in diverse data sources, such as interest rates and housing starts, or weather patterns and traffic data. More correlations can be examined, and better predictions surfaced. Oracle EPM Cloud applications already do some of this and we expect more capabilities in the future.
But it comes down to having the right data in the right shape. “If someone boasts to me, ‘Our enterprise is working with 10 petabytes of data,’” Andrew Moore, Dean of Carnegie Mellon’s School of Computer Science told me in an interview, “I will yawn unless I see there is good indexing of the data.” In order to generate business benefits, data must be well indexed—that is, defined in data models that serve a business purpose. AI is both a beneficiary of data models as they are improved, and a tool that can be used to give more structure to big data. Thus, even if big data investments haven’t yet achieved their hoped-for ROI, bigger returns are still in the pipeline as more AI comes online.
Enterprise application suites aren’t going to disappear as more AI gets switched on. Indeed, enterprise vendors are all working to add their own AI capabilities in the cloud. So expect AI not to replace existing systems, but rather to enhance or sit among them.
Oracle, for example, has embedded artificial intelligence into its cloud applications for finance, human resources, supply chain, manufacturing, commerce, customer service, sales and marketing professionals. Companies can also use Oracle AI Platform Cloud Service to access existing data sources, develop machine-learning solutions and integrate them into applications.