How AI can offer businesses greater control over their payments [Q&A]

We none of us particularly like handing over money, and that applies equally to businesses and individuals.

Payments are essential to the commercial world, up to now though they've always been a quite labor intensive to process. But what if it could be automated? Would you feel comfortable handing over control of payments to AI?

We spoke to Lloyd Humphries principal product manager for data and analytics at e-invoicing and accounts payable automation specialist Tradeshift to discuss these questions and more.

BN: What advantages does using AI to handle payments offer?

LH: One of the biggest obstacles standing in the way of AI adoption is quantifying the value that any proposed initiative will bring to the organization. With payments, there's no such problem: the benefits are clear cut, measurable and can be dramatic -- both in terms of how quickly they materialize and how much money can be saved.

To appreciate how profoundly AI can improve payments, it helps to understand how much of a hidden cost center they are for businesses. Right now, account payable (AP) teams in a typical organization spend around 25 percent of their time chasing and correcting invoice exemptions, since around four-in-five invoices require some form of manual intervention. This is tedious work, but it's always been seen as one of the costs of doing business. Even so, it comes at quite a price, since it costs an average of $11 and takes eight days to process every single invoice.

This is where AI can make such a transformational difference. When organizations digitize payments and support processing with the right AI and automation technology, they can slash the costs of processing invoices by a factor of five and bring down the average processing time to under three days. If that were all the value AI could bring, it would be compelling enough; however, these benefits should only be seen as the first fruits of payments automation. Automating payments isn't just about making processes faster and more accurate (important as that is): it should be the first step towards making the finance function more strategic and building stronger, more valuable relationships with suppliers.

BN: Why should enterprises trust AI with control over making their payments?

LH: This question cuts right to the heart of so many AI initiatives, especially those that involve sensitive data or strategic processes. We're keen to avoid developing black-box technology -- we trust users can deal with some of the complexity involved and are willing to put the data, the performance metrics, the projections and the trade-offs in front of them so they can make informed decisions.

Artificial intelligence is a lot like human intelligence: it doesn't come out of the box with all its faculties fully developed. It takes time and training before businesses should trust it to perform tasks with less human oversight. That's why we recommend businesses start small and teach the technology how to perform simple tasks. Once the AI is performing that task with an acceptable error rate (or markedly better than human operatives), that’s when it can be entrusted with the task.

While it's straightforward to automate the entire payments process right from the start, Tradeshift knows that organizations would never hand over the keys to artificial intelligence until they are comfortable with the technology. That might sound obvious, but we're the only technology provider that enables users to fine-tune the degree of automation, dialing fault tolerances up or down depending on the organization's goals or the specific task in hand. Other than Tradeshift's Ada AI, all the solutions currently on the market are a question of on-or-off rather than being something you can influence. And that has the effect of making trust in the whole application a binary.

BN: Which parts of the process lend themselves best to automation?

LH: If a task can be completed by a human with a machine then it can, in theory, be done by a machine alone. In any process, however, deciding which elements to automate will always differ from business to business.

Rather than seeing it as a good fit for some tasks but not for others, it's much better to see AI as a journey. A good starting point is to take a task that's time-consuming and error-prone when performed by people, but that also has minimal consequences should the technology make a mistake; for example, line matching between invoice and purchasing order, or entering the right coding categories for each invoice. Even in the early stages, with humans verifying the AI's accuracy, you can expect fewer errors with automation while also enjoying significant savings, which will only increase as the technology becomes more adept and highly trained.

BN: How can historical transaction data help to improve current performance?

LH: Businesses should see data as the fuel that enables AI to become smarter, faster. While you can train AI without it, it's much better if you can feed in historical transaction data which provides invaluable context and clues for the technology to chew over and learn from. The more data you can give the AI, the faster you can automate certain processes, and even do away with the approvals process entirely on certain recurring or low-value transactions from trusted vendors.

When considering the overall value of information, it's also important to think outside the bounds of your organization. Artificial intelligence can be a powerful weapon against fraud, especially when it is fueled by large quantities of data from multiple organizations around the world, by analyzing huge volumes of instances (such as historic security breaches) to identify patterns that would normally not be found by the human eye.

BN: Where does this fit into the wider business intelligence picture?

LH: One of the most compelling aspects of payments AI is that it gives us a glimpse of a future where business intelligence benefits every employee. BI is often painted as a complex, highly-specialized layer requiring discrete investment and talent: in other words, it has to go through a specialist team of data scientists and analysts before it can deliver value to people who can put the insight into action.

That's a fair assessment of BI as it stands. The next step in its evolution, however, will be to bring value to everyone in the organization, delivering them with insight that helps employees think strategically as well as with tasks in the here-and-now. Payments AI is starting to show us what this future might look like, where automation and digitization generates a wealth of insights and in a form that anyone can use to forecast demand and spend, identify patterns in global activities, and access real-time and highly-detailed metrics right. Rather than having to train people to use new systems, or have insight siloed in a part of the business that doesn't have a clear understanding of operations or strategy, payments AI points to a future where BI stops being seen as a standalone function. Instead, it will be woven into the systems we use every day; it will be 'free' and require little or no training, so anyone from senior strategist to front line worker can identify patterns that will help power the business' future success.

Image credit: AndreyPopov/depositphotos.com