Developing Granular Pilots for AI Implementation: A Use Case
Introduction
In October of 2024, my team at Acclaim built something I’m proud of: the first HIPAA-compliant, AI-enabled document extraction tool on the Appian platform. We integrated it into our client intake system, and the results were immediate. Intake became faster, more accurate, and far less stressful for staff and families. Beyond Acclaim, this project gave Appian a powerful proof of concept, showing how AI can automate critical but time-consuming workflows across healthcare.
What initially stood out to me wasn’t just the technical achievement. It was the human impact. Intake used to feel like a wall. Clients endured months of waiting, staff survived repetitive, mindless data entry, and there were little errors that added up to major barriers to accessing services. Staff were exhausted, families were frustrated, and providers were stuck in a cycle of rework. By automating the background tasks, we gave people back something invaluable: time and clarity. Families moved through the process faster, staff stress went down, and clinicians were able to deliver care sooner, with a focus on quality.
This is what I mean by human-centered AI in the context of my organization. My goal isn’t to replace people, but to work along side them. It’s to develop tools that remove inhumane and mindless work that does not lead to satisfaction or fulfillment without ignoring that that work does need to be done. In fields like autism care, where turnover reaches 65% annually, even small improvements in workflow can ripple outward into retention, satisfaction, and better outcomes for kids and families.
Extracting essential information from autism diagnosis documents is just a single use case, but it’s a powerful one that demonstrates how other use cases can be approached and executed. It shows that, if we start with where people are struggling the most in their roles, we can turn mundane work into positive human impact by drilling down in complex processes to identify granular work that can be easily scoped, developed, tested, and measured for impact. Here’s how we did it for the intake process.
Scoping
The first step was to identify which bottlenecks in the value chain were reducing access to our services. We found that caregivers often dropped out during intake because of communication fatigue and a lack of visibility into the process.
We began by building intake applications to simplify data collection. From there, we explored where AI-enabled document extraction could help automate repetitive tasks — the copying, pasting, and retyping that often caused errors.
One of the most critical documents our intake team handles is the autism diagnosis report, which is required to obtain authorization for treatment. These reports are narrative, unstructured, and no two look alike. Yet each one contains essential information: testing methods, date of diagnosis, client age, practitioner, and other details needed for both authorization and treatment planning.
Our final scoping step was to define the variables we wanted to capture and turn them into requirements for the development team. We validated the list with both clinical and operations leaders to make sure nothing was missed. This step was just as important as the technology itself: stakeholder input was key for accuracy, completeness, and ultimately buy-in.
Development
With clear requirements in hand, we built the first version of the document extraction tool. At the time, it required expression editing in Appian and enough technical detail that a business user needed developer support. But even within a year, the landscape shifted. Appian’s AI Document Center now allows users to accomplish much of the same work with well-written natural language prompts.
We tested our tool on about 100 diagnosis documents in a HIPAA-compliant environment and found it consistently accurate. We usually say the model is 95%+ accurate, but the reality is that since implementation it has not produced a single error in our clinical operations.
We remain mindful of the risks. Hallucinations and errors do exist in other models and use cases. But so far, our implementation has been stable, safe, and effective. Human-in-the-loop verifications ensure that any issues that do happen in the future can be prevented through final human verification and acceptance.
Testing and Adoption
Once the tool passed internal testing, we integrated it into our intake system and rolled it out to staff. Adoption was immediate. Tasks that once caused delays and mistakes became afterthoughts. Productivity surged.
In the first month, our intake volume increased by a factor of fifteen. That fundamentally changed the speed at which we could onboard client families. Intake became transparent, consistent, and easier to navigate. Staff anxiety decreased, and families had a clearer path into care.
I’ve presented this work at Appian World in April 2025 and Gartner’s AI conference in June 2025. Those talks focused on the technical and organizational impact. What I haven’t spoken about as much is how this project served as a proof point for human-centered AI use cases. Once people saw AI wasn’t about replacing them, but about removing friction and tedium, they began to imagine other applications. Staff who helped design the tool started identifying new opportunities where AI could add value.
What’s Next
Right now, we’re continuing down the value chain of autism care, tackling the administrative friction that drives burnout for families, operations teams, and clinicians. Each new use case is designed to make work more humane and efficient.
AI is coming to every technology and service organization, whether they’re ready or not. But in healthcare, we can’t afford to wait for vendors or tools to catch up. Fragmented systems still create barriers that delay care. My team and I have been committed to building and implementing granular, high-impact use cases that remove non-clinical work from the hands of people while leaving space for the judgment and empathy that only humans can provide.
