Artificial Intelligence (#237)
/The topic of this podcast episode is the impact of artificial intelligence on accounting. Key points made are noted below.
The Nature of Artificial Intelligence
Artificial intelligence is the simulation of human intelligence processes by machines, which includes learning, reasoning, and self-correction. AI is usually applied to expert systems, speech recognition, and machine vision.
Artificial Intelligence in Collections
So where can this fit into an accounting system? I’d be a bit concerned about having an AI actually keep the books, since you’d somehow have to monitor the journal entries it was making. It could come up with some pretty strange financial statements. However, it could work well in the collections area. First, consider the state of the technology in the collections function right now. At best, you have a database that tracks which receivables are overdue, who to call and when to call them, and which keeps track of the results of each collection call. In essence, it’s a database that supports the work of the collections staff. It also provides a certain amount of automation, like automatically dialing customers.
How could an AI improve on the situation? Consider what it would be like if an AI takes the place of the entire collections staff. You’d have to interface the AI with the collections database, and add a realistic voice synthesizer. Then have the AI call customers directly. Let’s go through how this would work.
The Mechanics of Customer Contacts
As an example, the AI automatically calls a customer. The customer claims not to have received whatever you shipped to him, so he’s not paying. The AI pulls up the delivery information from the third party shipping firm and sends the information to the customer instantly, while it’s still on the phone. This means the customer could be barraged with all possible forms of transmission. The information could be attached to an e-mail, or a text message, or even sent by an old-fashioned fax. Or all three at once.
Then the AI interprets whatever the person is saying, obtains some kind of a commitment from the customer, and transcribes the entire conversation into the collections database. And the AI could be having similar conversations with multiple customers at the same time.
This is the ideal situation. Now, what if the AI can’t understand the customer or the customer is being unusually stubborn. In this case, the call kicks out to a real collections person, who handles the call from there. But these instances should be relatively rare, especially as time goes on. The reason is that the AI is always learning, so it builds a bigger database of experiences over time. Eventually, it should have dealt with even very low-probability situations, and will know what to do.
Third Party Service Provider
Now, consider what it would be like if the AI was maintained directly by a third-party software provider, so the system is not maintained by the company. In this case, the supplier may be running the same AI for thousands of phone calls every day for all of its customers. This means the system keeps getting better, and at a fast pace. The AI could end up learning from millions of phone calls, so it recognizes all kinds of accents and knows how to deal with every possible situation.
So, why would it make sense for a software company to develop a collections AI? Consider the business case. The software company could make a sales pitch that it will take over the collections function from its customers. Entirely. In exchange for an annual contract, clients could eliminate their entire collection departments, which could be a huge savings.
Let’s say that the supplier sets its prices at half what the in-house collections function was costing its clients. Would it be profitable for the supplier? Well, what’s the supplier’s cost? It’s almost all fixed cost, with a variable cost component for a backup collections staff that takes over when the AI can’t handle a call.
Cost Structure for Artificial Intelligence
The fixed cost is a mainframe for the AI, a backup power system, and the costs for a few thousand phone lines, plus developers to monitor the whole thing. This costing structure means that the breakeven sales level is fairly high, but it’s nearly all profit for the supplier after that breakeven point is reached. So not only would this business be profitable, but it could create a billion dollar business, because it would make sense for a lot of customers.
Advantages of Artificial Intelligence
The reason I’m focusing on collections as a target for AI is that it doesn’t directly impact the accounting system. Instead, it just mimics a person in dealing with customers. And using an AI for collections is appealing for companies, because they can not only save a lot of money, but also the AI may be better than humans at collecting on overdue invoices. That means an AI could potentially accelerate the cash flows of a business.
Artificial Intelligence in Credit Analysis
Could AI be used anywhere else? I think so. Assigning credit to customers would be a good place for an AI. The system could access online credit reports, review the information, and issue a credit limit within a few seconds. That has two benefits. First, an AI always follows the credit granting rules, so it always issues the same amount of credit under the same circumstances. Which is to say that it can’t be persuaded by the sales manager to grant a larger credit line. The system would also learn from the ongoing bad debt history of the company, so its credit granting capabilities should improve over time.
There’s a good market for someone to create a credit-granting AI, since it could potentially be sold into hundreds of thousands of medium to large-size companies.
Artificial Intelligence in Auditing
Credit and collections are the best two opportunities. In addition, there might be a use for it in auditing, but only if there’s a way to input a lot of client financial information into the system. For example, what if a really large audit firm, like Ernst & Young, fed the financials for all of its clients into a central AI, which could then churn through the data and flag possible fraud situations? The AI could learn over time from the masses of financial statements, and probably develop a terrific skill level in fraud detection.
In this case, I don’t see AI saving money for audit firms through cost reductions. It can’t really replace auditors, since a large part of the job is having face time with customers. Instead, by pushing the fraud analysis angle, an audit firm could reduce its risk of being hit with shareholder lawsuits if it misses a client fraud situation.