Most health systems begin their AI journey with a single narrow use case and hope it scales. Far fewer get past the pilot. The hard part is proving value in one department convincingly enough that surgeons, schedulers and IT leaders all want the next deployment.
Minneapolis-based Allina Health took that path deliberately. The not-for-profit system serving the Twin Cities market employs roughly 28,000 team members and works with about 6,700 employed and affiliated providers across 82,000 annual surgical procedures and 6 million clinic visits. In 2022, it started with two specific problems inside a single hospital, block underutilization and robotic access. Four years later, that narrow beginning has grown into an enterprise partnership spanning perioperative coordination, inpatient capacity, coding automation and custom co-developed AI solutions.
During a Becker’s Healthcare webinar sponsored by Qventus, William Evans, senior vice president and chief medical group operations officer at Allina Health, Donna Sajady, business operations manager of surgical and perioperative services at Allina Health and David Atashroo, MD, chief clinical and AI officer at Qventus, discussed how the health system generated $7.3 million in added annual surgical contribution margin, recovered nearly 700 hours of robotic capacity and expanded automated care operations across the continuum.
Below are four takeaways from their discussion.
1. Results create demand
In 2022, Abbott Northwestern, Allina’s flagship and largest hospital, with 35 operating rooms, faced a familiar contradiction. Block utilization was low while demand for access was high. Robotic surgeons were the most vocal, insisting they could not get robot time even though utilization data suggested otherwise.
“It was clear we had challenges. What we had developed internally was not addressing it,” Ms. Sajady said.
Rather than pursue a broad transformation, the team scoped the initial deployment to two functions: block management and robotic availability. Results arrived faster than expected — an additional 3.5 cases per operating room, 36% growth in the robotics program and an 18% increase in spine cases. Block utilization moved from the mid-60% range into the 72% to 78% range.
Ms. Sajady attributed the early traction to three things: a tool her provider groups could adapt to their own workflows, surgeon champions who carried the message to end users, and targeted training delivered in phases. That proof point changed the pace of everything that followed. Allina had planned a staged rollout, but the team ended up managing enthusiasm rather than resistance as clinics at other hospitals asked when they would get access.
2. Personalization matters
Dr. Atashroo said generic outreach is where most capacity efforts break down. When a scheduler receives a blast email listing 20 open slots and only two are plausible for her surgeon, she learns to treat the messages as noise.
Instead, Qventus scores every surgeon against every available opening using hard filters layered with booking behavior, room and equipment preferences and the health system’s own strategic growth priorities, then reaches out in priority order.
“It is only marketing available time to a surgeon’s actual schedule pattern. If he is in clinic on Wednesday, it’s never going to let them know that there’s Wednesday time open,” Ms. Sajady said.
The same logic governs block release. Dr. Atashroo noted that roughly half the available opportunity sits in partial blocks, not full days. Hourly machine learning models predict which portions of a surgical block are unlikely to be used, factor in the surgeon’s booking patterns and time remaining, calculate the appropriate incentive and draft the outreach.
That tailoring shows up in practice. Oncology surgeons, who book on short notice, are not nudged at the three-week mark. Total joint surgeons, who book far in advance, are prompted at the 21-day requirement. The same history also feeds block review, giving Allina specific documentation of which nudges were accepted, denied or ignored and whether the time was ultimately used.
3. Schedulers need integration
About 25% of Allina’s providers are independent, and closer to 40% at Abbott Northwestern at the time of implementation, practices whose default system is their own ambulatory EHR, not the health system’s.
The mechanics matter as much as the access. Case fields auto-populate instead of requiring duplicate entry, removing an error-prone step that Dr. Atashroo said costs five to seven minutes per case through the traditional path. H&Ps, orders and consent forms are uploaded, read via OCR and written back into Allina’s Epic environment, and requests land in the scheduling queue without additional work from OR schedulers.
Employed physicians get the same one-click launch inside Epic, with a machine learning model predicting case duration and block time displayed alongside prioritized open time.
Allina never mandated adoption. “We did not mandate it. That wasn’t the approach we took and we didn’t need to. People quickly adopted and wanted it,” Ms. Sajady said.
An unanticipated benefit was waitlist functionality. Clinics with complex cases in heavily blocked rooms can flag a specific date or time window and be notified automatically when a case cancels or a block releases, rather than calling scheduling repeatedly to check.
4. Co-development helps clinicians
Allina also became Qventus’ inaugural AI Solution Factory client, a co-development model created after health systems told Qventus that they wanted to invest in a platform partner that could build solutions for their specific use cases and existing processes. Engineers embed in the partner’s day-to-day operations, leadership brings a wishlist of operational gaps, the two organizations align on opportunity size and feasibility, and Qventus builds with continuous client feedback.
Mr. Evans called it a career highlight. To surface candidates, Allina ran an internal AI hackathon that generated about 100 ideas from across the organization; with five engineers available, the list had to be narrowed. “Within 24 hours, we had a working prototype” that improved patient care, team safety and efficiency, he said.
Allina’s 2025 surgical results — $7.3 million in added contribution margin, nearly 700 additional hours of robotic capacity and a sixfold return on investment — reflect a sequence of decisions rather than a single deployment. The same platform recently absorbed the operational complexity of a new surgery tower, running parallel scheduling rules, blocks and guidelines for the existing 35 ORs and the incoming 38 without manual intervention.
“The technology is here. It’s working, it’s adding value now,” Mr. Evans said. “In my mind, this really does need to be a core competency for all systems. I think if you do not embrace this technology now, you’re quickly going to be left behind.”
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.
