When we talk about AI, the conversation almost inevitably turns toward chatbots that divine what we want, smart speakers that serve up information in a conversational tone, or cars that can see pedestrians and park themselves.
That’s all exciting stuff, no doubt, but for many businesses, reality sets in when they see how their accounting department goes about processing invoices and purchase orders. It’s for the most part still a low-tech world far removed from reaping the benefits of artificial intelligence.
Clerks in the AP department rip open envelopes and FedEx pouches filled with paper invoices, purchasing orders and receipts. Others electronically stream into their in-box. They manually enter them, assign the right GL code, and then send them on for approval and payment. A similar “black box experience” awaits partners and vendors who want to find out when they get paid.
But look again, and you’ll see that AI is already making inroads into one of the most traditional areas of a company. Done well, expert systems powered by machine learning can ingest invoices at high volume — no matter if they’re in electronic format, printed or just scribbled on a scrappy receipt. Software, in theory, doesn’t care what you put in front of it. It applies rules to decipher amounts, accompanying notes and issuing entities.
We call those inconspicuous helpers expert systems because they’re able to learn from their accomplishments and errors, as well as watch what their human minders are correcting and become true experts. Modern AP systems are already capable of applying their smarts to all kinds of documents you throw at them, comparing new invoices with old ones and making decisions on the fly on if a document should go into bucket A or B, or if it should be flagged for human review because something doesn’t match up.
A Magna International representative leads a demonstration of a semi-autonomous vehicle in Holly, Mich. Magna is a contract manufacturer of cars.
Bloomberg
But just how new is AI? You might object that optical character recognition (OCR) has been automating categorization for decades — telling a letter “O” from a number “zero,” for instance. But we’re talking about something fundamentally better that’s quickly going mainstream in terms of availability and affordability. Machine learning has trickled down from high-end tasks to mundane things such as AP automation in a small or mid-size company. You can buy fairly powerful AI as a service today, no matter what your budget, and just plug it into your workflow.
Algorithms similar to what makes autonomous driving in self-driving cars possible enable an enterprise to catch erroneous invoices or zap dupes. More to the point, state of the art software proactively scours every document for entire words or concepts. And thanks to the cloud, the software improves the more documents you feed it. Artificial neural networks, sometimes called incremental growing neural networks, or “gas,” exist that can learn new data without degrading the network’s performance or forgetting old inputs.
It’s the same approach to machine learning that companies like Google successfully apply to train its services on ever-growing datasets in search, translation, maps or tagging objects in photos. That way, good old paper invoices can be captured and converted into structured data to immediately analyze them. A system that learns like an eager apprentice knows which companies you do business with, what their terms of payment are and so forth. In sum, it creates what I like to call financial intelligence. It comes in handy to manage your cash flow and keep tabs on stores, geographies or vendor clusters.
The quietly growing potential of AI in fintech Investors have taken notice of the potential that machine learning and AI hold for fintech, including its subset, the accounting function. According to CB Insights , just the first half of 2016 saw more than 200 AI deals totaling close to $1.5 billion. Fintech is one of the core areas where intelligent algorithms can make a difference, not just in terms of flashy bots for wealth management but to modernize the staid AP function.
And we have just begun to scratch the surface of what AI can do for AP automation. Imagine a financial workflow where AI can automatically compare the information provided on a purchasing order, receipt or invoice with information available on the web, to validate addresses and phone numbers, even other company metrics. Such a true expert system will not only do the math for any amounts contained in documents, but more importantly it will catch suspicious or fraudulent documents and report them to the human supervisor.
It’s also only a matter of time before learning systems can understand the handwriting of frequent vendors and automatically translate it into structured data as if an invoice has arrived in digital format, something that traditional OCR was never capable of. Even better, AI will be able to sit at the company-wide intake source and channel the documents accordingly. Contracts will go to legal or HR, work orders and invoices to accounting and so forth. We are fast approaching a world in which human experts only do the real value-add such as vetting and reviewing, enabling them to deal with rising volumes in less time and with far fewer errors.
Don’t be distracted by the AI fireworks in 2017 that seem far away from day-to-day business. With AI and machine learning as a service, any company can dial up the performance of its accounting team on many levels right now. And what’s more, it can get better at it all the time — long before an autonomous truck will drop off the latest batch of paper invoices at your office.
Laurent Charpentier is the chief operating officer and chief innovation officer of Yooz North America.