Responsible AI in Hospice Starts With Knowing What AI Should Do

The pressure to have an AI plan can put technology ahead of the reason for using it. A hospice leader sees another demonstration, hears what peers are trying, and starts looking for somewhere to apply it. Another leader sees the risks and decides to wait.

Neither response answers the question that should guide the investment: what does our hospice need to do better?

Responsible AI in hospice starts with choosing work that AI is suited to support, defining its limits, and establishing how people will verify and use the results.

The value of AI should justify the cost, complexity, and risk. That value might be more time for care planning, less avoidable rework, or earlier visibility into gaps that need attention. Financial performance matters, alongside clinical integrity and the experience of patients and families.

Define the work AI should do and when it should help

Start with a task your team struggles to complete consistently. Then examine what makes it difficult. If the problem is unclear ownership, adding AI may simply produce another output nobody acts on. A reminder or process change may be enough. AI becomes worth considering when the work calls for capabilities such as finding relevant hidden gaps and risks, across extensive documentation.

Consider what that looks like in three hospice workflows.

1. Before an interdisciplinary team (IDT) meeting, a nurse may spend time moving between visit notes, medication changes, and functional assessments to prepare a patient update. AI can assemble that information into a reviewable starting point, showing changes and conflicting observations. The team still interprets the findings, with less time spent assembling the update and more time available for care planning.
2. Before recertification, a physician needs to assess the patient's course across visits and benefit periods. AI can bring relevant evidence together and flag gaps for review. If notes from the same period describe different assistance needs, the physician needs that discrepancy brought forward. AI supports the review; the physician remains responsible for the eligibility decision and certification narrative.
3. For chart audits, AI can review 100% of available charts instead of a sample, helping compliance staff identify missing support or inconsistent data. Broader coverage matters as CMS uses predictive analytics and machine learning in program integrity and examines claims patterns to target hospice oversight. Your team still needs findings it can investigate and prioritize, with clear steps for resolving the underlying issues.

When the help arrives matters. Identifying a gap before a decision gives the team time to clarify conflicting observations or obtain needed information. Finding it during an audit leaves the team explaining a decision already made, with fewer options for addressing the underlying gap.

Operational consistency in hospice starts with addressing those needs while the work is still underway.

Evaluate AI by what it helps your team address earlier and how much downstream work it can reduce.

Require hospice-specific design and demonstrated performance

Your AI strategy may involve a third-party vendor, an internal build, or capabilities your electronic medical record (EMR) vendor provides or plans to develop. Apply the same standard to each path: does it solve the hospice problem you identified, can your team use it responsibly, does it have the necessary guardrails for accuracy, privacy and security, and when will it be available? A roadmap commitment may inform your planning, but it cannot relieve a workload gap your team faces today.

A solution not purpose-built for the unique workflows of hospice may combine observations from different periods, omit meaningful changes, or present conflicting documentation as a consistent picture. Your team then has to reconstruct the record to determine what it can trust, potentially creating more work and leaving important gaps unresolved.

Choosing an AI vendor partner

If you are evaluating a vendor-provided AI solution, start with its hospice foundations. Was it originally developed for hospice, with experienced hospice operators shaping its workflows, or adapted from another setting? Ask who contributed that experience and how it influenced the product. Require a demonstration showing how it follows the patient across visits and benefit periods, reconciles observations from different disciplines, and represents decline, stability, and improvement. Have your clinical or compliance team verify the output against representative source records, with appropriate data protections in place.

Ask about known limitations. Ask the vendor to show how it works with your specific EMR today: what information it accesses, where staff review results and make corrections, and who supports the connection. Evaluate the actual workflow, rather than assuming compatibility from a logo or planned integration.

Building hospice AI internally

If you build internally, your organization owns development, AI usage costs, validation, security, ongoing support, and updates. Budget for maintaining EMR compatibility as systems change. Compare the full cost of building and maintaining the solution with the cost of a vendor-provided solution, rather than comparing a vendor's price with the cost of an initial prototype. Also, make sure you can build a solution in a way that ensures data privacy and security.

Evaluating your EMR's AI capabilities

Your EMR supports documentation and daily operations. Specialized AI requires dedicated development and refinement. A complementary vendor can focus on a defined set of hospice problems while the EMR company develops its core platform, extending your capabilities without requiring the EMR to build everything itself.

Test a defined use case before expanding

You do not need to change the whole organization to find out whether AI helps. Start with one workflow and a small group of staff. As an example, consider a hospice testing whether an AI solution can reduce the work of preparing for IDT meetings. Before testing, establish how patient information will be protected and who will verify the results.

1. Define the improvement you expect. That might mean reducing preparation time while giving the team a more complete picture of each patient. Record current preparation time and agree on what information the team needs for its discussion.
2. Compare results in everyday use. Have staff use the solution to prepare for IDT meetings and check its output against the source records. Does it capture relevant changes and conflicting observations? Measure the total preparation time, including review and corrections, and ask whether the team arrives better prepared to discuss care.
3. Use the findings to decide what comes next. If preparation takes less time and the team consistently has the information it needs, expand to another team. Address omissions or added work before expanding, and continue checking whether the benefits hold as use grows.

Building a smarter hospice for 2027

Akssi, from Buds Technology, was built by hospice providers for hospice, with workflows designed specifically for hospice rather than adapted from another setting. It brings together patient information from across your EMR so teams can see the patient's course more clearly, identify gaps while there is still time to address them, and better support clinical decisions.

That is where responsible AI should lead: less time piecing information together, earlier attention to what needs action, and clinical judgment supported by a clearer picture of the patient.

As you plan for 2027, explore how Recert-Assist, Audit-Scan, and IDT-Prep can help your team strengthen recertification review, identify documentation gaps earlier, and prepare for more informed IDT discussions.