This post is part two in our three-part series that defines AI Agents by WorkFusion. In part one, which you can read here, we explained that our AI agents are defined this way: AI Agents are digital co-workers that decide, act, and communicate. WorkFusion AI Agents are also pre-built, explainable, and controlled.
The purpose of this post is to explain the meaning of these three core AI Agent functions and to provide insights about why WorkFusion customers benefit from them being pre-built, explainable, and controlled.
#1: AI Agents Decide: The WorkFusion AI Agents make decisions and recommendations that were previously made by people. Because each individual AI Agent is designed to perform a specific process, such as KYC, CDD, sanctions screening alerts review, etc., they are goal orientated. The decisions they make are not made in a vacuum of predetermined rules. Instead, they have environmental awareness from the perspective of the inputs they receive and the interactions they have with people and other systems. As they interact, they apply adaptive learning to continuously improve.
#2: AI Agents Act: They integrate with other internal and external systems, including data systems, to put decisions into action. Rather than simply connecting to multiple systems to share data, they frequently transform data (such as reformatting an address) and smartly connect with systems, tools, and sources along the process. This enables them to support multi-step execution and fully perform an end-to-end process.
#3: AI Agents Communicate: WorkFusion AI Agents can perform agentic collaboration across other AI Agents, people, and other reporting systems.
More than a decade of progress translates to unique value
Because WorkFusion has incorporated multiple AI and automation technologies into our platform since 2013—and focused heavily on the financial services industry for just as long—we provide AI Agents that are ready to run out of the box and that have AI guardrails already in place. We call these additional attributes ‘Pre-Built, Explainable, and Controlled.’
Pre-Built: WorkFusion customers derive great speed and ROI benefits by leveraging AI Agents that arrive with out-of-the-box, pre-trained AI models. They don’t have to code or train the models, yet at the same time, they can take them further if they choose to do so. They can customize the rules, dictate their desired thresholds for confidence levels (such as the confidence level they can flexibly set for determining a false positive), and use AutoML for things like document tagging and training. Also included are WorkFusion’s analytics. So, it’s not just about having an AI Agent that can do a piece of work, but it’s also about performing the overall process—having the proven mechanisms in place to manage how customers’ AI Agents perform and meet their specific needs to make their business users effective.
Explainable: Every step and decision made by WorkFusion AI Agents is explainable, with multiple levels of transparency, often leveraging scoring and prioritization techniques. Within this realm of transparency, we include our use of large language models (LLMs). This allows for WorkFusion AI Agents to incorporate OpenAI, ChatGPT, BERT, Mistral, and other LLMs into the processes they perform. Customers can also incorporate their own LLMS, essentially giving themselves a private LLM.
When an AI Agent uses GenAI, it is always in conjunction with WorkFusion’s other models that act as a check on potential problems. As an example, consider an AI Agent that performs adverse media screening alert review. If that agent typically provides 75 percent straight through processing (STP) on alert decisions, an LLM can be incorporated to give its own decisions. When WorkFusion decisions and the LLM decisions agree, the confidence level in decisions rises to 100 percent, enabling full STP. In the event they disagree, WorkFusion capabilities for human-in-the-loop reviews come into play. AI Agents commonly collaborate with people at one or more steps in a process to ensure that situations requiring nuanced analysis involve people who are typically analysts or investigators. This type of explainability, combined with agent-people interactions and checks, puts operations teams at ease.
Controlled: WorkFusion AI Agents are not fully autonomous, and thus, never out of a customer’s control. As noted above, agents can make decisions, but oversight can be built into every workflow, if desired. As noted above, human-in-the-loop technology enables people to collaborate with our AI Agents throughout each process. The only exception to that occurs when a customer desires 100 percent STP. However, that is rare in the financial services industry.
The decision to mainly remain semi-autonomous has always been a purposeful decision by WorkFusion. To be completely AI-driven means introducing unacceptably high levels of risk for customers who operate in highly regulated industries like banking and financial services. Having a system with multiple checks and human-in-the-loop technologies helps to avoid AI hallucinations and other issues that could compromise results. With AI Agents’ ability to provide incredibly high rates of automation, we see no reason to compromise the system-wide guardrails that our AI Agents follow. This viewpoint leads to proven results that auditors and regulators embrace, enabling customers to reap the full benefits of AI-driven process automation.
Stay tuned for part 3 of this 3-part series, which we will publish soon. In the meantime, schedule your own personal demo to learn more.
AI,ai agents,AML,Banking & Financial Services,Digital Workers,FinCrime
