As AI agents handle more complex use cases, a broader industry desire has emerged for more dynamic agents that can reach a goal, without following a script or sacrificing compliance.

For example, a caller filing an auto insurance claim is likely to share a lot of context at once, so your agent needs to move fluidly between capturing what’s been said, asking about what’s still needed, all while staying within its guardrails to assign a claim number.

Today, there are two ways to build complex agents: (1) a single prompt that allows wide ranging conversation but more loosely follows business processes, or (2) a multi-state agent with defined transitions but hampers the ability to have truly fluid conversations.

To address this gap, we are introducing Goal Oriented Agents. Instead of scripting the order of every question, you tell the agent what it needs to accomplish and the prerequisites to get there with goal nodes. It adaptively gathers the information it needs to reach that resolution, instead of following a rigid set of rules. More flexibility means a better chance your agent reaches its goals.

The result is an agent that’s conversational, compliant, and dynamic.

Handle Complex Intake and Qualification

As agents take on harder work, Goal Oriented Agents let brands scale that complexity up, without losing control of guardrails or the direction of a conversation. This unlocks more complex use cases, such as Medicare enrollment, roadside assistance, mortgage qualification, and HIPAA-compliant medical intake.

Complex intake and qualification are some of the strongest use cases we’ve seen for Goal Oriented Agents. For example, a caller filing an auto insurance claim is likely to share a lot of context at once: what happened in an accident, where, and who was involved.

A more rigid agent built on conditional logic tends to handle this kind of call with repetitive follow-up questions, often asking for details the caller already gave just to make sure each field gets captured.

A Goal Oriented Agent checks everything the caller has said against its goals and prerequisites before asking for anything again, and it keeps an empathetic tone the whole way through. It can take in all that context without losing track of what it ultimately needs: in this case, enough information to assign a case number.

How It Works: A Dual-Model Architecture

Goal Oriented Agents handle complexity with a dual-model architecture, including a director and an actor:

  • The actor is a smaller, fast model that drives the conversation, optimized for natural, engaging dialogue.
  • The director works in the background, checking progress against the goal requirements.

As soon as the caller finishes speaking, the actor and the director start working in parallel. The actor replies right away, so the conversation keeps a natural pace. Meanwhile, the director checks whether the goals for the current stage have been met. From there, it either suggests how the actor should guide the conversation or routes the call to a new stage with updated instructions for the actor.

To build a Goal Oriented Agent, you define what each goal node needs: the specific goals you want by the end of the goal node, the prerequisites to get there, or both.

  • Goals can be collected variables, like the caller’s name, or tools, like scheduling a callback.
  • Prerequisites can be data fields, like date of birth or location for a qualifying sale, or plain-language conditions, like “the caller mentioned a death in the family.”

Once those goals and prerequisites are met, the agent can take an action or move on to the next goal within the node, bringing the conversation closer to resolution.

At the end of every turn, the director checks progress against both. When a caller shares extra context in a single answer, the director can check off several goals and prerequisites at once. In the example above, Dan’s first message covers the vehicle damage, photo evidence, and injuries in one turn, so the director marks all three as collected, and the agent moves straight to asking for his name.

With his name captured, the director sees that the location and policy number are still missing, so the actor asks where he is next. Dan never has to repeat himself, and each question moves the call closer to a case number. One model holds the conversation while the other makes sure the agent stays on track.

A Flexible Agent That Navigates Complexity

Agents that stick rigidly to a script lose the back-and-forth that makes a conversation feel human. Goal Oriented Agents follow the caller’s lead, asking questions in whatever order fits the conversation. The director keeps that flexibility in check, holding the agent to its goals and guardrails on every turn.

That’s the same principle behind how Regal approaches AI generally. Our priority isn’t just teaching customers to use our tools, it’s making sure the technology solves the problem and successfully completes the complex tasks at hand.

Have a press release to share? Contact our team today.

Have an article or a piece you would like to contribute? Get in touch with the editorial team.