Igniting Success For AI Pilots in Healthcare

Healthcare providers, from Top 500 hospital systems to small specialty practices, are looking to utilize AI. Their goal is to speed operations and improve patient care, but all too often they wind up with failed pilots or stalled initiatives.

The Concord Connect™ Ignition Program, a white-glove implementation methodology, is designed to help healthcare organizations seamlessly transition from manual document processing to AI-driven, Straight-Through Processing (STP) where structured and unstructured data are brought together into a single, intelligent workflow.

Rupali Katole, Senior Program Manager, Concord Connect, and Project Ignition Program Lead, explains how the Ignition Program works, the questions to ask, and why early buy-in from staff is essential to achieve AI pilot success in healthcare.

Data shows that only three in 10 AI projects in healthcare actually launch. What are the main reasons the other seven don’t succeed?

There are several key issues that create healthcare AI adoption barriers, causing AI projects to stall or fail:

• Lack of Clear Objectives: Many organizations start AI projects without a well-defined goal. Without understanding the specific problem to solve, projects can lose direction quickly.
• Poor Data Quality: At the outset, healthcare data is often messy or incomplete. Extensive data cleanup is required to ensure the AI model works effectively. Without this crucial step, projects can or will falter.
• Adoption and Change Management: Even the best AI models won’t make an impact if staff aren’t prepared to change longstanding workflows. Having a plan to boost staff comfort and drive adoption, with elements that include ongoing training and support, is critical. That includes ensuring staff are aware that every workflow contains a “human in the loop,” in addition to the automation.
• Unrealistic Expectations: Sometimes organizations expect technology to instantly solve deep-rooted workflow issues. This disconnect between expectations and reality can lead to disappointment and abandonment of the project.

If you had to pick one factor that most often stops an AI project, even after a significant investment of time and money, what would it be?

The most common factor is a rush to solve too much at once. A better approach is to analyze, then prioritize, the most pressing pain points and workflows. From there, create defined, measurable goals. Technology alone doesn’t drive adoption. There needs to be a phased, incremental approach to deploying workflow solutions to achieve results that can be built upon in subsequent stages. Piloting an AI program in a small, manageable way and gradually rolling it out is critical. Without these “baby steps,” organizations struggle to achieve adoption.

How does the Ignition Program address the barriers you just mentioned?

The Ignition Program recognizes that every healthcare organization is unique, with its own workflows and software environments. No two systems are exactly alike. From the beginning, we engage both super-users and day-to-day staff in the implementation process. Rather than a one-size-fits-all solution, we partner with our customers to build a custom model as well as complementary workflows that use the model and streamline patient-information workflows. This hands-on, collaborative approach ensures the solution fits their workflow and drives better user adoption.

The Ignition program has five phases. Why was it important to break implementation into these specific steps?

Breaking the process into phases and focusing on scalability makes a project more manageable and less overwhelming, as well as set expectations for the client. Each phase is designed around the roles and responsibilities of the people involved, ensuring that the right stakeholders are engaged at the right times. For example:

• Strategic Alignment/Discovery: Ensures the correct problem is being addressed. On the customer side, a dedicated Project Manager, IT Director, Clinical Record Specialists and other key team members analyze document workflows and define high-level objectives. They are joined by a dedicated team from Concord that includes an Implementation Project Manager, Data Engineers, System Engineer, Developers, and other experts.
• Data Readiness & Technical Feasibility: Verifies the solution will work in the real-world workflow based on the technical capabilities and integration available. The AI team works with super-users to assess data quality and test technical feasibility with real-world samples to build a workflow that:
  • uses classification to identify document types and route to the correct teams
  • Extracts key patient and clinical information for matching to the correct patient record or workflow
  • Design workflow rules that can trigger notifications, routing, and downstream integration.
• Targeted Pilot Deployment: Demonstrates technical feasibility and allows for live testing before full rollout. The system is launched in a controlled, live environment, so models can be applied, and confidence thresholds established for confident and measurable outcomes. These thresholds determine whether data is automatically processed or whether staff validation is needed. This results in setting and meeting expectations.
• Operational Launch & Optimization (Go-Live): Focuses on user training and change management to drive adoption. This is the transition to full deployment, where a user feedback loop is created to fine-tune performance, retrain models, and ensure seamless integration into the environment. Here is where teams receive targeted training and knowledge transfer, building confidence in and familiarity with the solution.
• Sustain & Scale: Keeps the solution performing at agreed-upon standards. Ongoing support, continuous model improvement, and evolving human-in-the-loop workflows to expand automation and build new, interoperable STP networks.

This structure provides experts during five clear-cut and well-defined stages and gives the organization clear guidance on who to engage at every step of the process.

Can you share some real-world results from healthcare organizations that have used your solution?

Absolutely. Here are a few examples of workflow improvements:

Faster document processing, staff time savings

At a major children’s hospital and research center, a 19-document-type classification, referral-subspecialty model with entity extraction achieved significant accuracy at the outset, far beyond the provider’s expectation of at least greater than 70% during the initial phase. About one minute per document is being saved with full automation, with between 1,000 and 3,000 documents being processed weekly, resulting in time savings between 16 and 50 hours per week through touchless processing.

Manual lookup eliminated, indexing streamlined

With our FHIR Lookup capability, records administrators at a large cancer-treatment provider now automatically pull specific patient data from more than 99% of 14 types of incoming documents, including medical records, labs, prescriptions, referrals and billing, to be matched with the correct patient chart in the provider’s oncoEHR. This eliminated manual lookup and streamlined document indexing across eight workflow queues, and into the EHR, for up to 2,000 documents daily.

Rapid deployment, immediate time savings

A primary and specialty care system had very aggressive implementation timelines for records administrators, and saw onboarding completed in just two weeks on a 12-document-type classification model with 15+ extraction entities, including an Epic EHR integration. Document types include diagnostics, clinical labs, cardiology testing, specialty consults, admin forms, patient services forms, and medical records. Meeting those tight deadlines allowed them to make significant changes within their tech stack to improve operations in a very short window of time. The system receives 500 documents per day, and processing time was decreased by 2-3 minutes per document, and 50% of manual work was eliminated.

How does the Ignition Program help Concord continue to enhance its expertise, and help its customers improve their operations?

Every implementation teaches both Concord and our customers something new. We are learning from new document types, workflow challenges, and customer needs. These insights feed back into our product development, helping us build even more accurate and specialized models for future customers.

For instance, a recent implementation had more than 10 workflow queues, each with 35–40 workflow rules. This required constant collaboration at the customer’s desired, rapid, pace, to achieve deployment deadlines. The lessons learned here benefited both future projects and product development.

If you were advising a healthcare leader interested in using AI to improve system workflows, what’s your best piece of advice?

Start with the impact you want to achieve. Clearly define your scope and set measurable outcomes. Listen to your staff and identify their pain points, no matter how small they may seem. Understand how those individual roadblocks can add up to expensive inefficiencies.

Gaining clarity around the problems to be solved, will guide the alignment of new workflow processes into your technology stack that in turn drive scalable implementation that meets your goals.

Any final insights to share?

Recent data from MIT’s NANDA initiative found that 95% of generative AI pilots failed to deliver measurable ROI — and the problem was not in flawed models, but rather in poor integration and misaligned priorities. Importantly, the same research found that vendor-led AI projects are more successful than internal builds.

That’s why it’s essential when buying an AI product to partner with implementation experts who know your domain, someone with proven success. It makes a big difference. Always start small with a pilot and scale up as you see results. Let the workflows and AI serve your team, not the other way around.

Learn more about the Concord Connect Ignition Program here.