Five Practical AI Tips For Better Workflows 

Every behavioral health leader feels the same pressure building at the same time: more documentation, tighter compliance expectations, and teams that are already stretched thin trying to keep up. AI keeps entering the conversation as a possible answer, but whether or not it make a meaningful difference in the day-to-day reality of your practice is tricky to figure out 

It’s a challenge our friends at The Crossroads, Inc., were facing, as their Clinical Program Director, Dillon Stewart, shared in a recent webinar. For Dillon and The crossroads, their biggest breakthrough with making AI meaningful in practice came from focusing on a single, familiar problem—documentation pulling clinicians away from patients—and asking how to solve it in a way that actually sticks. 

Dillon was given space to work within Kipu Intelligence to build a massive set of beneficial changes to The Crossroads’ workflows that reduced administrative drag, improved documentation quality, strengthened reimbursement, and gave staff time and confidence back in their work. 

Based on what Dillon shared, here’s the five tips that can help leaders like you make your AI solutions more practical and build big positive impacts on care. 

The fastest way to make AI useful is to aim it squarely at the point where your operations are already straining. In Dillon’s case, that pressure showed up in a place every behavioral health leader recognizes immediately: documentation consuming time that should be spent with patients. 

Instead of stepping back to design a sweeping strategy, he leaned into that tension and followed it. Where are clinicians losing time? Where does the work slow down or break apart? Those questions became the starting point for everything that followed, and they led directly to the first meaningful use cases. 

In practice, Dillon started by deploying AI where the time loss was most visible. He rolled out the group note assistant so staff could dictate or jot quick shorthand during sessions and generate structured notes afterward. He then customized prompts behind biopsychosocial assessments and progress notes so the system pulled the right details, emphasized risk and medical necessity, and produced payer-ready language. 

As those pieces proved reliable, he expanded the same approach across templates by refining prompts, connecting outputs to treatment plans, and embedding the tools directly into everyday documentation so clinicians could stay in the session and let the system handle the structure. 

Dillon approached this by working directly inside the moments where documentation was already happening and reshaping what those moments looked like. 

In group sessions, staff no longer had to choose between being present and capturing everything. They could dictate or jot quick shorthand during the session, then generate structured, individualized notes for each client afterward. What used to require a second pass, often at the end of a long day, was handled as a natural extension of the session itself. 

From there, he extended the same approach into assessments and treatment planning. During a biopsychosocial, staff could focus on the conversation while the system captured and organized the information in the background, guided by prompts he had already configured. Those outputs flowed straight into treatment plans and progress notes, carrying context forward and keeping everything aligned so clinicians could move through the work without stopping to stitch pieces together. 

As Dillon put it, “The services we’re providing is not the documentation. The documentation is just recording the event that took place.” That shift in thinking shaped how everything was built. 

The result is a workflow where clinicians stay engaged with patients while documentation happens alongside the work There’s no extra system to log into, no separate process to manage. It’s the same workflow, just lighter, faster, and more consistent. 

Dillon’s progress with AI accelerated when he started treating prompts as something that could be built, tested, and improved over time, much like any other core part of the system. 

Early on, his prompts were simple. A sentence or two asking the system to summarize or extract key details. They worked, but the outputs were inconsistent and often missed what mattered most for clinical accuracy or payer expectations. So he started refining them. 

He expanded the instructions, added specificity around risk factors, medical necessity, and tone, and then tested the outputs against real scenarios. When something didn’t look right, he adjusted the prompt and ran it again. Over time, those prompts became more structured, more detailed, and far more reliable. 

In practice, that meant building out prompts for each part of the workflow. A treatment plan prompt could span multiple pages to ensure goals, barriers, and success measures are consistently defined. Each one was designed with a clear purpose and refined based on how it performed in the real world. 

As Dillon described it, “The more instruction I gave it and the tighter that instruction was, the better the output turned out to be.” That realization turned prompting into a repeatable process instead of a one-time task. 

As Dillon expanded his use of AI, he focused on building continuity across the entire care journey. He started by feeding outputs from one part of the workflow directly into the next, so information captured during a biopsychosocial assessment carried forward seamlessly rather than sitting isolated in a single document. It carried into the treatment plan, shaping goals and identifying risks. Those same elements then showed up in progress notes, where clinicians could track movement against those goals without having to restate or reinterpret the same information each time. 

In practical terms, that meant a clinician could run a session, dictate or summarize it, and have the system tie that interaction back to specific treatment plan goals, highlight barriers to progress, and reinforce medical necessity. The connection was already there because the prompts were designed to look for it. 

That continuity shows up quickly in the places that matter most. Documentation becomes easier to review, audits move faster, and payer conversations become more straightforward because the rationale for care is consistently supported from start to finish. 

As clinicians used the system, the uncertainty that often comes with documentation began to fade. They could focus on the interaction in front of them, knowing the details would be captured clearly and shaped into documentation that held up under review. 

It wasn’t long before notes became more consistent, consistency translated into stronger quality, and that quality showed up in fewer denials and more reliable reimbursement. Just as important, the day felt different. When documentation stops weighing on every interaction, clinicians have more energy for patients and more control over their time, something that’s increasingly hard to come by in a field where burnout continues to climb. 

Across organizations that are getting real value from AI, the trajectory looks similar. Progress starts in a focused corner of the workflow, proves itself quickly, and then expands as confidence grows. Documentation becomes faster and more consistent, clinicians spend more time in sessions, and leaders gain clearer visibility into quality and reimbursement performance. 

What stands out in Dillon’s experience is how much of that progress came from working within a system that already supported those workflows. Because Kipu Intelligence lives inside the EMR, he wasn’t stitching together separate tools or asking staff to change how they worked. He was able to shape the documentation process directly by designing prompts, refining templates, and embedding intelligence exactly where decisions and documentation were already happening. 

Leaders who see the strongest results create space for that kind of progress. They identify someone close to the work who’s willing to test, refine, and keep pushing until it clicks, then support that effort with time, feedback, and visibility. 

If that capability exists in your organization, the path forward is already within reach. The next step is simply to start where the friction is highest and let the results build from there.  

About the Authors

Travis Moon Travis Moon, Kipu's Content Marketing Strategist, is a seasoned leader in healthcare IT and content marketing, specializing in the behavioral healthcare sector. He develops impactful, data-driven campaigns that support healthcare professionals and enhance patient outcomes. With over a decade of experience, Travis has led strategic content initiatives for major healthcare organizations, including the launch of data visualization tools and thought leadership campaigns.