From Big Lists to Bite-Sized Action: How Siftwell Analytics Makes AI Actually Useful for Healthcare Operators

 

From Big Lists to Bite-Sized Action: How Siftwell Analytics Makes AI Actually Useful for Healthcare Operators

Every health plan has the lists. Thousands of members predicted to have high emergency department utilization. Tens of thousands unlikely to refill critical medications. Hundreds of thousands at risk for chronic disease complications. The predictions are accurate. The models are sophisticated. And the lists sit unused because nobody knows what to do with 50,000 names ranked by risk score.

In a recent episode of Category Visionaries, Trey Sutten, CEO and Co-Founder of Siftwell Analytics, a healthcare technology company that’s raised over 5 million in funding, described why most healthcare AI fails at the point of implementation—and how his company engineered a solution that operators actually use.

The Operator’s Frustration with Predictive Analytics

Trey understands the list problem intimately because he lived it as a health plan CEO and Medicaid CFO. Predictive analytics vendors would deliver impressive models with excellent accuracy metrics. Then they’d export a spreadsheet with thousands of members ranked by risk.

“I think a lot of people, a lot of technologists have entered the healthcare industry and thought that, you know, it was a better model, a better mousetrap that was going to do it. They wanted to talk about accuracy measures like area under the curve or something,” Trey explained. The problem wasn’t that the predictions were wrong—it was that they weren’t actionable.

A health plan might have 50 care coordinators and 12,000 members predicted to skip cancer screenings. Even with perfect predictions, how do you decide which 1,200 members each coordinator focuses on? What intervention does each member need? Why are they unlikely to get screened in the first place?

Traditional predictive analytics answers none of these questions. It ranks members by risk and walks away.

The Actionability Gap

This gap between prediction and action creates a fundamental problem in healthcare AI adoption. Health plans invest in predictive analytics, get accurate predictions, and then discover they can’t operationalize the insights at scale.

The result is what Trey calls “big lists that are rank ordered”—technically impressive but operationally useless. Care teams either ignore them entirely or randomly select members from the top of the list and hope for the best.

This is the problem Siftwell was built to solve. “The motivation really is to go from big lists that are rank ordered to bite sized chunks that people can understand, marshal resources around and really get results,” Trey said.

How Siftwell Breaks Down the Lists

Siftwell’s approach starts with predictions but doesn’t stop there. The platform uses multiple types of AI to transform predictions into actionable segments with clear intervention strategies.

Take the cancer screening example Trey shared: “We had a client that wanted to better understand how to connect more of their members to cancer screenings. We ran the analytics, identified the 12,000 that were unlikely to go, but we go deeper and we help them understand for different cohorts within that 12,000.”

Rather than presenting 12,000 names ranked by non-compliance probability, Siftwell segments that population into cohorts based on why they won’t get screened. One cohort might have an 80% chance of non-compliance related to transportation, affordability, and distance to screening facilities. Another cohort might have compliance barriers related to language access, health literacy, or cultural factors.

This segmentation transforms an overwhelming list into manageable cohorts with clear intervention strategies. The transportation-challenged cohort needs mobile screening units or ride vouchers. The health literacy cohort needs education materials in their primary language and community health worker support.

The Multi-AI Architecture Behind Actionability

Making predictions actionable requires more than just better predictive models. Siftwell combines multiple AI approaches to explain predictions and generate recommendations.

“We use machine learning to make predictions and then we really dig into those predictions using a couple other types of artificial intelligence to explain why those individuals are being identified for those predictions,” Trey explained.

This multi-layered approach matters because different AI techniques solve different problems. Machine learning excels at pattern recognition and prediction. But explaining why a prediction was made and what to do about it requires different capabilities—natural language processing to interpret unstructured data, causal inference to identify drivers of behavior, and recommendation engines to match interventions to barriers.

By orchestrating multiple AI systems, Siftwell delivers what operators actually need: not just who is at risk, but why they’re at risk and what intervention is most likely to change that outcome.

Why Operators Built It Differently

Siftwell’s focus on actionability stems directly from the founding team’s operational experience. When Trey gathered C-suite executives from his previous health plan to identify market gaps, the problem wasn’t prediction accuracy—it was resource allocation.

“We started to collect the C level executives and say, hey, you know, now that we’re figuring out what our next thing is, since we’re not all going to this combined organization, what is it that the market needs? Where is their holes?” Trey recalled. The answer was clear: health plans needed “this notion that you can match people with their needs at exactly the right time and the right level.”

This operator perspective shaped every product decision. Features aren’t built because they’re technically interesting—they’re built because they solve real operational challenges that Trey and his team faced in previous roles.

“What way more important to me is getting the results that I’m trying to get. This isn’t about how accurate your model is, it’s about your model telling me that an individual needs an intervention and so that their use of an emergency department or their readmission rates go down,” Trey emphasized.

The Bite-Sized Chunks Framework

Siftwell’s approach to actionability follows a repeatable framework that works across use cases:

Step 1: Predict the outcome. Identify members at risk for poor outcomes—missed screenings, medication non-adherence, avoidable ED visits, readmissions.

Step 2: Segment by barriers. Use AI to understand why different cohorts are at risk. What specific barriers prevent them from achieving better outcomes?

Step 3: Match interventions to barriers. For each barrier-based cohort, identify interventions most likely to overcome those specific barriers.

Step 4: Deliver actionable cohorts. Present care teams with manageable cohorts (hundreds, not thousands) paired with specific intervention strategies.

This framework transforms overwhelming lists into executable care plans. Instead of asking care coordinators to figure out what to do with 12,000 names, Siftwell hands them 15 cohorts of 200-800 members each, with clear intervention strategies for each cohort.

Measuring What Matters: Outcomes Over Accuracy

The shift from predictions to actionable insights changes how success is measured. Traditional analytics vendors optimize for model accuracy. Siftwell optimizes for outcome improvement.

“This isn’t about how accurate your model is, it’s about your model telling me that an individual needs an intervention and so that their use of an emergency department or their readmission rates go down,” Trey said. “That’s what I’m fundamentally about in my former operating roles and that’s what Siftwell is selling us.”

This outcome focus creates a different relationship with clients. Siftwell isn’t selling analytics software—they’re selling measurable improvements in health outcomes and cost reduction. The accuracy of the underlying models matters only insofar as it drives better outcomes.

The Competitive Moat of Actionability

Making AI actionable creates defensibility that pure prediction engines lack. Once a health plan builds workflows around Siftwell’s cohort-based approach, switching to a vendor that delivers ranked lists would require completely rebuilding operational processes.

The domain expertise required to segment by meaningful barriers and match effective interventions also creates a moat. “We’re taking really deep and battle tested managed care experience, combining it with these advanced Technologies to not only point out who they should focus on, but why they should focus on and even what they should do in order to better engage them,” Trey explained.

Competitors can build accurate prediction models. Replicating the operational knowledge of what interventions work for which barrier types requires years of managed care experience—exactly what Siftwell’s founding team brings.

Lessons for B2B AI Builders

Siftwell’s approach offers a framework for any B2B AI company struggling with adoption:

Prediction is necessary but insufficient. Users need to know what to do with predictions, not just have them.

Actionability requires understanding operational workflows. You can’t make AI actionable without deeply understanding how your customers actually work.

Segment by the dimensions that matter operationally. Risk scores aren’t operationally useful. Segments based on barrier types and intervention strategies are.

Use multiple AI techniques. Different problems require different AI approaches. Orchestrate them to deliver complete solutions.

Measure success by outcomes, not technical metrics. Optimize for what your customers actually care about achieving.

The gap between predictive analytics and actionable insights has plagued healthcare AI for years. Siftwell’s solution isn’t more sophisticated models—it’s building the complete system that operators need to turn predictions into improved outcomes. That’s the difference between analytics that impresses data scientists and analytics that transforms healthcare operations.