Communication between people and machines is complex. Today, we use dropdown menus, prompts, hints and outright commands to let machines know what we want. In the world of lending, however, conversations are restricted. The main instrument used is the form. It can be physical or digital. But generally, it is deterministic in design. The lender needs your data. You have an option to choose from a menu of loan offerings. Then the loan will be processed. I submit that this is doing the lender as much as disservice as it is doing so to the borrower. But this can be addressed relatively easily today, compared to even a couple of years ago. Here, I will focus on what can be done and what the impact would be on the bank and the borrower. We intend to go back into the most natural form of engaging. Natural language(NL) enables a lay person to communicate in writing with a machine. Allowances are made for creole language(if that is widely used), multiple languages can be leveraged, spelling and grammatical mistakes are not hurdles and slangs are ok. This allows a potential borrower to have a conversation with a bank of his/her choice. Today, in our lab, we are working on NL as a key tool to address the complexities of multilingual Asian markets, especially rural and small-town communities. Let us take an example. Simon runs a small ceramics kiln and sends his products to a large city. He needs credit to expand his capacity. If he were to follow the general route of applying online, he may need to download a form, fill it up, scan it and send it. If he has better luck, there may be an online form. Then he has to attach proof of address, proof of income, asset list(if needed) and so on. This can faze someone running a small, creative and logistically demanding business. Let me add here that this can be the same issue for a graphic designer or a coder looking to grow. Sometimes, proactive loan managers call and ask detailed questions. Unfortunately, in many markets, the branch visit and face to face conversations seem to have diminished. In any case, not everyone may live close to the branch they bank with. In the world of digital banking, you would find it hard to relate to a specific branch and it's staff. These are real problems. People who need financing are not bankers and in many cases, do not have accountants and attorneys on call. But what if Simon could just type out in a web application the following: "Hi, I am looking for a 75,000 dollar loan, ballpark. Need to expand my kiln. I am a potter. I bank with you guys. Can someone please help me?" But if you observe in your interactions carefully, such a query is routed on a number of occasions to a "friendly AI assistant", likely a legacy of the Robotic Process Automation era. Simon will get a canned answer and at most a list of things he can do or options he can tick. Maybe there is a person to call at the end, but he is already at the starting line of the race to exhaustion. This need not be the case. First, by the month of December 2025, lenders and other financial service providers should start to lose their RPA applications. Second, not using Natural Language to it's fullest potential is only a loss to a financial institution. Think of how much data Simon has shared already. He has told the bank his name, his need, the amount, the purpose. He has said that he banks with them. Wow. That's a lot of data first base. Now, the bank could either get someone on the phone to him or suss out a bit more in a friendly way. But that won't come in a templated back-and-forth. The bank engagement platform has to engage proactively and flexibly. Simon is of this era. He gets how to use prompts and enter into chats with LLMs. Let's use that to design an interaction that is meaningful. So, maybe the bank's engagement machine responds thus to him. "Hey, Simon. That's great. We can offer you something if you can just scan your ID and send it to us. A quick check and we will go to the next step very soon". He is an existing customer. His ID should be enough to run a check. And as the machine guides him through one step after another, he will tell you more about himself. And for the actual loan officer, there are two big benefits. One, he can process multiple applications at the same time without spending his limited physical capacity. Yet, he can keep a live eye on all. And then there are the insights. He will have told the machine not only who he is, but his underlying motives for financing, how his business is doing, what sort of growth he is trying to get too, any tough times going on right now or expected soon. How is the machine getting trained so that it can maximize the yield from this conversation without signalling intrusion? That is key. How are the insights getting crunched in real-time and thrown into the mix with traditional credit scores to generate reputational ranking, psychographic profiling , demand intensity and potential of risk? The words used by Simon and the context of the words are all-important. Let us say the bank hears him say, "I am negotiating a shipment of 20 X Dynasty vases for next month". Ok, big statement opportunity and potential. But also a good reference to look again at event risk and crunch the numbers. Context matters. Reputational ranking, psychographic profiling, demand intensity and contextual event risk. There may be more but let's say we can start here, using both Natural Language Processing and Machine Learning. It is time to shift lending from apps to conversations.
