From Process to AI: How Agent Finder Pinpoints Your Best Agent Opportunities

What’s the problem?

Implementing AI agents can seem daunting and organizations struggle to understand the best use cases. The challenge is picking a use case that has a strong ROI and can make an impact. But it also needs to contribute to an overall end-to-end process. There is no point implementing an agent in one part of the process that simply creates a bottleneck further downstream. Or worse, building an agent that doesn’t address the true root cause.

Build on solid foundations

What is very clear from all the research on AI success from industry analysts and consultants is that you need to have well-defined business processes to understand the best use case. But more importantly, to take advantage of AI Agents, you need to rethink your business processes. This is because your agents have skills that humans don’t. And your current processes were designed with the limitations of humans. We explored this in detail in this blog.

So a clear definition of the end-to-end business processes enables you to pinpoint the issue that needs to be addressed by an agent. This is very clearly spelled out in the video case study of our Case-to-Bug process. There were several areas where an agent could have been built to support the process. 

They used Agent Finder inside the process diagram to identify all the potential agents that could support the process. But the team could then evaluate each suggestion to understand the true problem and which agent to build. Find the video case study at the end of this article.

Agent Finder: How does it work?

Agent Finder analyzes a Universal Process Notation (UPN) process map, which acts as a blueprint of your business operations.

Using a set of pre-defined criteria, Agent Finder can:

• Analyze documented processes to automatically identify potential use cases for AI agents.
• Categorize the type of AI agent needed by considering the nature of the process and data involved.
• Provide reasoning and confidence levels for its recommendations.

The business requirement it writes helps you pinpoint exactly where agents can have the greatest impact and helps you to optimize processes before applying AI to them. Agent Finder’s recommendations are presented with links to the particular step in the process diagram so that you can see them in context. You can then choose to accept the suggestion, and a new requirement is automatically created.

Agent Finder did a job better than a dozen CTAs in a 90 minute workshop. Amazing.

Chief Technical Architect, Global Systems Integrator

Below is an image showing the process diagram and the right panel with the 3 agent recommendations. The mouse is rolled over the first recommendation, so it shows the detail and highlights the process step it refers to. If you click on the tick, it writes the business requirement and attaches it to the process step. 

Different types of agents

There are different ways that AI can help accelerate or improve a business process. A conversational agent or an AI-powered workflow may be more effective. The key is understanding that “agent” is an umbrella term for different kinds of AI-powered solutions. 

Agent Finder understands these differences when it makes its recommendations. In the example above, there are 3 recommendations; one recommendation was a Conversational Agent, and 2 were Flow.

3 ways to solve a problem:

• Conversational Agents: This is the most flexible type of AI agent, excelling at understanding natural human language and working with unstructured data. These agents can manage complex validations, guide users to accurate answers, and assist with flexible planning. A conversational agent can empower a junior team member with the knowledge of your best employee or act as a virtual coach.
• AI-Powered Workflows: These agents are characterized by a “flexible process” but rely on “fixed data”. They use AI to introduce adaptive logic and advanced reasoning within a structured data environment, making them valuable for automating complex logic that would be difficult to program traditionally. This type of agent could also produce written summaries of complex data.
• Flow (Pure Automation): This is for processes that are “fixed process” and involve “fixed data.” This type of automation doesn’t always need a full AI agent, but can be made faster and simpler through a workflow, such as a button that opens a form, which a workflow then processes.

Getting started

Getting started with Agent Finder is designed to be simple and intuitive. You just need a UPN process diagram to begin. Once you have a diagram open in your workspace:

1. Open the right-hand panel within the UPN process diagram.
2. Choose to accept the use cases that make the most sense for your business, and Agent Finder will create the necessary requirements for you.

This visual and interactive approach helps you quickly see and validate potential AI applications in the context of your actual business processes.

But then what?

After identifying and accepting a potential use case, you’ll have a set of new requirements. Now you can:

1. Evaluate Existing Agents: You may already have an agent that can handle the new task as a new Topic.
2. Create a New Agent: If a new agent is needed, you will create an Agent Instruction Diagram (AID) to visually map out how the agent will achieve its goals, defining its actions and guardrails.

The AI can automatically generate user stories and acceptance criteria directly from the diagram, which are then synced to development tools like Salesforce DevOps or Jira for the dev teams.

But I don’t have any process maps!

No problem. Configuration Mining can automatically generate UPN process diagrams from your Salesforce metadata, providing an on-demand blueprint of how your organization operates. Alternatively, you can use AI to generate a UPN diagram from a transcript, a document, or notes in just a few minutes. You can also start from scratch, or from a pre-built template in Elements.cloud, which is designed to be simple and easy for anyone to use.

Case-to-Bug example

One real-world example of this approach is the “Case-to-Bug” process at Elements.cloud.

The Problem: Customer support agents lacked the deep technical knowledge to write bug reports that development teams could use. This resulted in a “language barrier” and missing information, leading to endless back-and-forth communication and a growing backlog of unresolved issues.

The Solution: The team used a process mapping approach called an Agent Interaction Map (AIM) to create a conversational AI agent that acts as a virtual expert. This agent gives support staff the technical knowledge they need from a 45-page playbook without having to dig through the document. When a case is confirmed as a bug, a Salesforce Flow is triggered, which uses a prompt template to automatically generate a detailed and organized bug report from the case’s unstructured data.

The Result: The average bug resolution time dropped from over 23 days to just over 5 days. The quality of bug documentation dramatically improved, with the average score jumping from a dismal 0.8 out of 10 to a solid 8 out of 10. The transformation eliminated the problem of bugs getting lost in the backlog, leading to a 100% resolution rate.