UiPath

Transcription:
Transcripts are generated using a combination of speech recognition software and human transcribers, and may contain errors. Please check the corresponding audio for the authoritative record.

Ryan Ma (00:08):
I hope that doesn't count against the time. It's great to be here everyone. Thanks for taking the time. I'm going to go through a very quick presentation of UiPath today. Just for those that aren't aware, we are actually not a FinTech. We are more of a horizontal automation provider in the space. That said, we have worked with many of the large banks and non-bank lenders across the United States and across EMEA. I see some familiar customers in the crowd today. But what we really want to focus on is some of the newest launches that we've put out over the last couple of months. The first one being our capability to orchestrate end-to-end business processes. So as you think about why you might want to pay attention to the next nine minutes, a portion of that is you have probably made large investments into systems of record, loan origination systems, servicing systems, and all of these are solutions that are very critical for your day-to-day business.

(01:04):
But they don't always have the end-to-end business process in mind. And as you think about processes, they normally traverse across multiple systems. The second piece of this is over the last six to 12 months, I would bet everybody in the room has heard the word either "AI agent" or "agentic." We have put out an agentic solution as it relates to being able to build AI agents. What we're starting to do is really commercialize these, make them into more financial services-based products so that we can bring a loan eligibility agent or a QA QC agent to the market to help drive those efficiencies that everybody's talking about as it relates to AI. So this is the backdrop. This is just a very quick look from our website, but what I want to do is get into our platform and talk first about process orchestration and then second around AI agents.

(01:55):
To start, what I have pulled up on the screen here is an example of what we call UiPath Maestro; this is our conductor. This is our orchestration layer that we have within our platform. A couple of key pieces to take away as you look at the screen: one is we're able to build and design these processes within our platform that are going to have a very familiar look and feel that many operation folks would be able to understand. This uses industry-standard BPMN to design the process. But we take it a level further beyond documentation to start to actually wire these processes up, to put an underlying automation, an underlying human-in-the-loop activity, or workflow that exists. And then also, now we're talking about AI agents as well—being able to actually trigger an AI agent if and when needed within an end-to-end process.

(02:41):
You'll see here, just as I zoom in very quickly, we'll be able to actually start to dissect some of the core pieces of this process that we've built in this loan origination process. We have upfront things like being able to read financial documents. You've probably seen some vendors already today that are doing document extraction. You're also going to see loan origination system updates. So we'd be able to use an automation to stamp back into your system of record. We don't become that system of record; we're just an orchestrator across your process. And then downstream, you'll even see in some cases some parallel processing where you have agents within this process that can run concurrently to make sure you move from step A to B to C as fast as possible. Now I won't belabor this too long; I want to actually get into what one of these agents might look like in practice.

(03:28):
What you'll see as I switch screens here is our agent builder technology. I mentioned we have the capability to use some of our out-of-the-box agents that we've built with some of our customers to do things like QA/QC and loan eligibility. We do things across the horizontal, so if you're looking at procurement, finance, or IT, all those types of operations are supported as well. But really, what's very special about AI agents and where we think there's going to be a massive amount of value is we're now able to handle more of those cognitive tasks that need to happen within a process. People have used automation traditionally to move data from point A to point B or to maybe send a communication. What an agent will be able to do is take unstructured input and start to make decisions around how risky an individual loan is that we're about to underwrite.

(04:16):
Is there any missing information on this application that is absolutely critical for us to move forward? That is all powered by the use of large language models in the backend. You'll see in this visual editor that I have here, we have a model on the left-hand side. You're going to have optionality to be able to choose an LLM of choice. You might use OpenAI, Google, Microsoft, or you might want to bring your own. All of that is supported. On the top, you'll see context. This is where we start to make solutions very specific to your business and your organization, because a model is extremely intelligent, but it is also able to reference operating procedures that are specific to your organization such that you can make sure the decisions that come out of these AI platforms and AI programs are grounded in the way you want your answers to be.

(05:05):
On the right-hand side, you'll see some tooling as it relates to being able to do data fetches and integrations. We can do everything from API-based integration to user interface-based integration. For example, triggering an on-premise robot to be able to actually scrape a user interface and get data that's required from a system that's on a mainframe or a third-party application that's not API accessible. What I'll show you here very quickly on the right-hand side is actually how we code the agent. Moving from workflow or traditional automation into the age of AI, everything that we're going to do in terms of instructing this agent is going to be free text. Here what we're doing is actually describing to this piece of software essentially: what is your role, what is your goal, and what are you responsible for achieving?

(05:49):

That agent, using the model, the context, and the tools you make available to it, will be able to start to make decisions, give output, and at this point be able to hand off to a human or automate straight through depending on what the process is. Again, what's really critical here is these are going to be cognitive-based tasks that were not automatable before. That's where we're going to drive that next incremental value as you think about where automation fits within your organization. Just to wrap us up and to round us out here, what I want to be able to do is actually show you what one of these runs looks like. In this example, what we actually provided to the agent was some input data that may have come from systems of record or other workflows in your organization, passing in things like a loan-to-value ratio and a massive JSON payload for document extraction.

(06:37):

What you'll be able to see actually is when we push this data into the agent, the agent will start and look through, use an LLM, be able to trigger tools to gather extra information as needed, and then ultimately get to a point where the agent is able to return—in this case—the risk score and the explanation for how it derived that answer. This is where cognitive work starts to become automatable when we can use these types of frameworks and embed them within a full end-to-end process. The last thing I will leave you all with is that we really believe that agents are not going to take over the world. They're not going to be their own workflows, but they're meant to be an accelerant to make sure that humans can do less of the repetitive clicks, keystrokes, and data copy-pasting, and really move more into how to mitigate risk or how to get this loan closed faster. This is how we're seeing our customers start to move the needle around SLA adherence, cycle times, and driving cost out of the business as well. I have 30 seconds left, so that's it. Thank you everybody.