🦄 Making great presentations more accessible. This project aims to enhances multilingual accessibility and discoverability while maintaining the integrity of original content. Detailed transcriptions and keyframes preserve the nuances and technical insights that make each session compelling.

Overview

📖 AWS re:Invent 2025 - Advancing Patient and Business Operations Insights with Gen AI (IND212)

In this video, Shane O’Connor and Kendra from Huron Consulting Group demonstrate how generative AI is advancing healthcare analytics in patient experience and business operations. They showcase their AWS architecture using Amazon Bedrock, Nova LLM, and Redshift to analyze unstructured data from patient rounding surveys and revenue cycle notes. The platform achieves 90% accuracy in sentiment analysis, processing 10,000 notes weekly, enabling early detection of patient dissatisfaction before HCAHPS survey results arrive 3-4 months later. They preview future capabilities including real-time transcription of patient rounds and integrated analytics connecting patient experience to financial performance, emphasizing coaching opportunities for staff improvement.

; This article is entirely auto-generated while preserving the original presentation content as much as possible. Please note that there may be typos or inaccuracies.

Main Part

Introduction: Huron Consulting’s Approach to Generative AI in Healthcare Analytics

All right, welcome everybody. Shane O’Connor here, joined by Kendra. We also have one of our solutions architects in the audience with us today. We’re going to be talking about advancing patient business operations analytics with generative AI.

Today we’ll go through our stance on opportunities for generative AI use cases in healthcare data, compare our architecture and approaches, talk about some of the results that we’re seeing from our first pilots, and then give you a little preview of our roadmap.

So just by way of introduction, Huron Consulting Group is a global consulting firm with a really heavy presence in both education and healthcare industries. Kendra and I both work in different sides of healthcare, but as we’ve kind of come together to prepare for this, I’ve noticed a lot of commonalities in how we can be using our AWS architectures and the new capabilities with generative AI to start extending the services we’re able to deliver to our clients.

So Kendra, do you want to give a little background on patient experience and the Huron Rounding tool? Sure. Huron Rounding is a structured evidence-based approach to improve patient experience because it creates meaningful and repetitive interactions between leaders, caregivers, patients, and families, and improves communication, accountability, and outcomes across hospitals and health systems. In addition, it uses consistent leader rounding, typically by nurse leaders, managers, or executives, to strengthen relationships with staff, identify barriers to performance, and quickly escalate issues for action.

And within our business operations segment of healthcare consulting, we focus on improving client financial performance through changes within their revenue cycles, supply chain operations, pharmacy, workforce, and HR operations. We really aim to be their transformational partner through difficult change management opportunities. And as you can imagine, data is very much at the center of our DNA and how we orient ourselves towards what’s going to be most impactful for our clients.

So what generative AI in AWS has enabled us to start thinking about is moving beyond the structured data that we’re used to, which is still very much a part of how we deliver, and getting into more ambitious opportunities. I like to say it’s a click down to sort of the real story of exactly how our clients can start to take action on the financial opportunities we’re guiding them to. And that takes the form of really starting to use LLMs to understand the unstructured data pieces that we have access to and combine that into our delivery approaches.

Patient Experience Analytics: HCAHPS Performance and AWS Architecture Implementation

As we look into opportunities for generative AI and how it’s impacting patient experience through analytics, the platform analyzes survey feedback from rounding sessions to identify key patient experience drivers influencing HCAHPS performance. By moving beyond basic sentiment analysis, it delivers actionable insights that support timely staff coaching, enhanced service quality, and strengthen overall patient satisfaction.

HCAHPS data is a survey that is sent out to discharged patients from hospitals that participate in the Medicare inpatient payment system. The higher the score from the HCAHPS survey, which means that you had a positive patient experience at the hospital, the more funding the hospital gets. So it is important to get positive HCAHPS data back. The downfall to this is that it does take three to four months before the data is received back, and by then missed opportunities for early intervention to issues have been lost.

And I think we can move to our architecture slides now. Currently within the rounding platform we run manually entered questions and answers through a sentiment analysis. We use three of Amazon’s services. We use Amazon Bedrock, the Nova LLM model, and also Redshift. The EventBridge is just a glorified schedule that triggers the step function for sentiment daily, which invokes Redshift stored procedures.

We register the Nova Micro model, which shows up in Redshift as a function that can be called. If you are familiar with the LLM model in general, the concept of the prompt within the model passes the question and answer text, which instructs the model to return one of four sentiments based on each answer and question.

By way of comparison, I think the business operations architecture functions in much the same flow. We are receiving raw data from our clients, sometimes in the form of flat files, sometimes other data sources. We bring that and store that in S3, move it to tables in Glue Catalog, and then we begin to really curate the data.

Through transformations in Redshift powered by DBT, we stack all these different disparate data sources and perform the hard work of modeling them together. This allows us to assemble more of a holistic story of, for example, the path that a claim is taking from billing to resolution and all of the different activities that are done to drive that forward. On top of that, we’re able to then facilitate access to this for our consultants and our clients in different ways.

Sometimes that’s going to take the form of simply writing out the processed LLM response or summary of the data to a table that we can then visualize in Amazon QuickSight. We’re also able to connect that to low-code apps so that we can more ad hoc make these Bedrock calls and perform the summarization of all of the data sources together to give a holistic view of everything that we are receiving.

Advancing Capabilities: From Sentiment Analysis to Automated Transcription and Business Operations Integration

As we look to elevate patient experience by understanding improvement opportunities, we’re going to be analyzing the survey responses to uncover underlying emotions, enabling rapid detection of dissatisfaction and improvement opportunities that drive higher patient satisfaction and foster that lasting loyalty. We look to identify patterns in patient interactions to guide improvements and streamline workflows and enhance care delivery. By turning that feedback into actionable insights, healthcare organizations can strengthen patient relationships, improve care quality, and optimize operations and align growth strategy with real-world patient needs.

Here’s an example of a real patient experience survey template questionnaire, and so I took out a couple of themes here around scheduling and access to care and provider-staff interaction. We pulled out a couple of different underlying sentiments where the patient expressed frustration with scheduling processes, noting that the automated system was difficult to navigate and they would have preferred to actually just talk to somebody directly. However, they did find that their pre-appointment details were helpful and clear, and once at the facility they were pleased with how quickly they were seen and their interactions with the staff.

So we pulled out two sentiments, one negative and one positive, and we are able to spot patterns in patient interactions that help guide the focus improvements in workflows for smoother care. Moving beyond just running manually entered questions and answer text through sentiment analysis, we are looking to the future of rounding AI. This is a recording and transcription process where the person conducting the round will be able to turn on the recording, set the device down, and walk away, and the application will seamlessly document the round in the background. This takes away from the impersonal approach of having the nurse’s face in a device or a piece of paper and the exhaustive note-taking after the fact.

We are going to offer this in three areas where we offer a real-time streaming process where the round questions are asked and answered and documented instantly. In areas of the facility where there are low internet connectivities, a batch processing option will be designed that will keep and track all of those key details that were captured during the rounds and still run them through the AI process. It just won’t be instant as you saw with the streaming process. And if all else fails, they can always record a voice memo on their phone and upload the file after the fact.

The AI process is so sophisticated that the person conducting the round has the flexibility to ask the questions in their own words and in any sequence, which takes away from having to read a script from a piece of paper or from a laptop. This is completely configurable by location, by subject, and by template. I know that I have preferenced a lot for this session for patient experience. However, using this capability in staff rounds, staff interactions, and using that for coaching opportunity is going to be huge. In addition to the rounds being transcribed, key details such as issues, recognitions, and notes are captured.

The system takes all key details from the rounds, including these three things, and documents them and sends them for resolution. We will continue to provide a sentiment analysis per question as we do today with the manually entered answers that we run through sentiment, but now we will be providing an overall sentiment of the round so the person conducting the round can see how the patient was feeling at the end of the round.

The round transcription is securely stored and it can be reviewed for further opportunities in the future. The upside to this as well is as the round is being documented, we will be providing an explanation as to why AI chose the answer it did. We understand that AI is going to be new to a lot of our service providers, so we want to educate them on why AI is choosing the answer it does. We will track all updated answers. The users have full autonomy to update any answers AI has chosen for them, and they have full control over them.

From a business operations side, we’re thinking of very similar things. It’s integrating again disparate data sources to a curated history of financial performance and then using large language models in addition to our typical metric analysis to highlight areas for improvement in things like payer relations, staff performance, and denials management. What we try to do here is facilitate the ability to look either at an atomic level of a claim or a patient history, or more at a level that’s kind of analogous to population health, where we’re looking for trends across payer populations, across your service lines, and pointing out things like the key issues, the denial patterns, what seems to be going on within staff handoffs, that deeper level of insight paired with our structured data.

That points to opportunities where we can help our clients navigate change management. What’s very important to us is to bake this all towards a score and kind of a way for our clients to start to understand their opportunities differently, be that in staff effectiveness or in how they’re managing their payer relations. One way that we’ve kind of taken this the furthest is really in that staff management, ensuring there’s quality actions being taken to drive all accounts to resolution.

There’s a big thing that we focus on with our clients, which is the effectiveness of workflow from the time a claim is billed to when we’re getting fully paid and resolved. What you’re looking at here is how we can write these results out ultimately to tables that are read in Amazon QuickSight, which give again that staff level performance and an overall picture of the score of our rating of how effective the actions being taken are. Again, that’s all coming from us being able to use the architecture to go after these free text notes and unstructured pieces of data that just a few years ago we would have had to review manually to provide that kind of additional insight.

Results, Impact, and Future Roadmap: Achieving 90% Accuracy and Expanding Coaching Opportunities

As we look to our results and what is going to make this feature successful, sentiment analysis is currently at 90% accuracy. We started at about 75% and have worked up to 90%. We understand, especially within healthcare, that we need to get to as close to 100% as possible. As we continue to evolve the model, adapt, and the machine learns, we hope to get closer to that 100%.

To capitalize off of this new data that we will be receiving, moving towards creating smart rounds using real-time sentiment analysis to identify patients at risk of bad experiences, to provide targeted service recovery interventions is going to be key. We can start to prioritize those patients that had previously a bad experience, make sure that their issues that they previously had seen with the hospital have been resolved, and we are targeting their core problems.

As mentioned, this will have a direct impact on the hospital revenue through optimized HCAHPS funding. As stated, HCAHPS funding takes about 3 to 4 months to get back. So with us being able to detect poor patient experience earlier on, we’ll hopefully increase those HCAHPS scores that the hospitals are seeing, which will in return impact their funding.

From a business operations side, we’re starting to just gain a lot more scale with the amount of data that we’re able to process. We’re at 10,000 notes that are reviewed on an automated cadence per week now, while we maintain around a 90% accuracy.

We’ll be able to expand that accuracy greatly as we move to broader sections of our client base. It’s not really just the staff effectiveness and the quality pieces. This is also identifying net revenue opportunities and denials trends at a deeper level than we were able to before as well.

As we look to the roadmap and what our future is, we will start to aggregate the sentiment data through business operation outputs, emerging themes and opportunities through our BI analytics and platforms, which we do use Amazon QuickSight for. We’re looking for opportunities for enhanced coaching, connecting the results to real-time coaching opportunities for staff to improve patient interactions and staff satisfaction to follow up quality. I just want to emphasize that coaching piece that’s really core to who we are. Landing all of these new data and insights just to a dashboard really gets us about halfway down the journey with our clients.

What Huron is always aiming to do is then pair with our clients on the strategy, whether it’s with clinical staff or business operations staff, to help them get better, to help them actually seize these opportunities. Certainly, the ability to go deeper into what the opportunities are by having language models summarize free text for us gets us to the place where we’re strategizing with our clients a lot sooner. So the coaching piece will continue to be a very high priority for us.

Then we’re also just excited as we start to connect our data. Currently, we are on similar architectures but not totally connected in some of the same AWS accounts. As we start to fold things together more, we think there’s opportunities for our consulting offerings to grow, starting to tell stories and find opportunities that are in the hidden interconnections between this data. How does patient experience start to translate to market share and market capture that our clients are going for? How do things within patient sentiment reflect possibly lingering net revenue opportunities that are going to happen when it comes time for full collections by our healthcare clients?

So, thank you all for coming. Hope you enjoyed this. Kendra and I will be around after and happy to talk with anybody about opportunities that you see in healthcare.

---

; This article is entirely auto-generated using Amazon Bedrock.