AI in Accounting and Finance: A Q&A with Aaron Harris
I recently sat down with Sage CTO, Aaron Harris, to discuss key use cases where AI and machine learning can help the accounting industry, developing trends, and what to do with the data that AI provides. Check out what Aaron had to say…
Q: What do you think is important to today’s finance leaders?
A: Today’s finance leaders want more time to focus on strategic work. They want to empower their teams with better technology to automate repetitive, routine tasks so they can shift their team’s view from looking at the past to looking at the future. We have a pretty provocative view that breaks from traditional accounting and finance solutions. Accounting and finance solutions should embed powerful analytic capabilities, enabling interactive exploration in the system of record. However, that’s not enough. We want solutions to push “active insights” to finance leaders. “Active” means the solutions push the insights when they’re discovered so leaders can make decision. And the insights shouldn’t necessarily be answers to questions that are well-defined. Ultimately, we can use modern technology to continuously scan business activity looking for emerging opportunities and risks finance leaders may not have thought to explore.
Q: What are some future trends that you see developing in the finance world?
A: As we think about the future, I think we’ll see some really provocative statements on how innovation can transform the accounting industry, that will get CFOs and finance leaders and their teams out of these periodic, redundant, frankly low value processes, and focused on the future.
I envision a world where the books are always ready for reporting, where you always have real time, complete, up to date information about the performance of your business. If you can achieve that nirvana of continuous accounting - as I like to describe it - you can see what's going on in your business. Continuous accounting isn’t valuable without confidence that the data's gone through some assurance processes, that anomalies have been identified, and that exceptions have been reviewed. We’re already using machine learning technology to test the accuracy of the data that you're reporting.
Q: That all sounds great; how do we get there though?
A: Getting to this future requires the power of multi-tenant cloud computing. That's really the first step of this. The next step is harnessing that power to develop new capabilities that are designed on the basis of artificial intelligence (AI) and machine learning.
That said, trust and ethical use of data is an absolute priority. We’re mindful that our use of AI must be transparent and ensure we meet customers’ privacy needs; for example, ensuring that all data is anonymised. That’s why we’re embarking on customer research into their concerns and priorities for AI, evolving our ethical principles, as well as ensuring we have the right framework in place to deliver them.
I’ve tried to help our customers, and the industry at large, understand in very practical ways what AI and machine learning will do for the industry. There's a lot of hype, and a bit of anxiety; some of it unwarranted. There's a sense in the industry that people need to embrace AI, and they need to prepare for it. But what we've found is customers have been really wanting to know in far more practical terms, "What does it really mean? What do I really need to do to prepare for it?"
Q: What are some of the key use cases where AI and machine learning can help the accounting industry in practical ways?
A: I think there are six future use cases:
- Continuous analytics and performance monitoring: With all that rich data flowing through the system in real time, we will be able to analyze that data continuously with powerful computing capability to develop data models.
- Anomaly detection: We will find those anomalies in real time, and alert you from that sea of thousands of transactions which ones probably warrant a human going in and looking at what could be inaccurate, irregular, or simply fraudulent.
- Continuous security monitoring: In the same way that we will continuously monitor the business to find anomalies in the data, we think we could also look for anomalies in activity that may be evidence of a malicious actor trying to gain unauthorized access, or perhaps an employee trying to do something irregular.
- Recommender systems: To anticipate the user experience of the future, we look to get away from the basics today that are defined by lists and menus and bookmarks and shortcuts, and get to user interfaces that adapt based on the ways the system learns the user likes to work. What are the activities they care about, that they spend time with? It becomes a much more dynamic, reactive user experience as we learn more about the behaviors of the user.
- Process automation: We want to apply machine learning to automate the long tail of manual activity that still happens within your teams.
- Conversational AI and bots: In some cases, employees refuse to log in to the accounting system to complete simple tasks like purchase approvals. Conversational AI will allow these individuals to complete those tasks using the tools they already know and love, like corporate communications platforms, digital assistants, etc.
Q: What can we do with all the data that AI provides?
A: You can develop powerful collective intelligence capabilities. Most people understand that collective intelligence means that if you're part of a peer group, there's a couple of hundred other companies that look just like you. We look forward to being able to do benchmarking and help you understand where you are performing versus your peers for the KPIs that matter to you.
What will be really powerful though, is when you can actually analyze all that information among your peer groups and develop the relationships that drive performance. For example, to one day be able to give recommendations and say, “Businesses that have greater value renewals tend to have shorter cash collection cycles. If you want to drive high value renewals, then you need to focus on your cash collections, because that's what we can see in the performance across your peer group, and where you can improve your overall performance.”
That means getting to leveraging that massive amount of data across the customer base, across the organizations, to help them collectively to improve their performance.
Q: As a developer, you spend a lot of time researching and gathering customer data to help improve the next product iteration. What is something that’s surprised you?
A: It's kind of funny to me – If you ask a customer, “We want to build out some forecasting abilities using machine learning. Where do you think we should focus?” You expect them to say cash flow forecasting, right?
What was surprising to me about this whole process of really working with customers is: that really is the answer. Any way you slice it, everybody wants better visibility to cash flow. We interviewed one customer of a well-funded private company. They've raised a lot of money, they're growing like crazy, and they've got a really sophisticated accounting team. The CFO spends two hours every day looking at cash.
Q: What are the signs of AI project failure? How do you mitigate them and find a solution?
A: Artificial Intelligence projects can fail for all the same reasons that any other software project fails, even though the risk profile for AI projects tends to be higher. At Sage, we broadly divide the risks into six categories: technology, market, team, legal, financial, and business model.
However, AI projects have an added layer of risk that is often underestimated, especially amongst executives that aren’t educated in AI. When building an AI service, it is mostly unknown whether it is at all possible to create the desired output from the available data set. The signs of failure here are clear: there is no signal in the data, model accuracy is low, the delivered output isn’t as good as the chosen benchmark, etc.
In the Sage AI Labs, the way we mitigate the above risks is by running a disciplined agile process, combined with a highly cross functional team that is in constant contact with customers and experts to validate outputs. This can only work well if all the basics have been taken care off, such as seamless data access, security and data governance, and a frictionless self-service “you own it, you run it” platform.
We also work to continuously adjust our process over time and have learned from our early mistakes – such as letting siloed data scientists go dark for months or prematurely optimizing a model without ever testing it with a customer.
We improve everyday and there’s so many great things to come. It’s an incredibly exciting time to be a part of this industry.