From bicycle to rocket: what HUS proved about specialty care automation

Sosiaali- ja terveydenhuollon ATK-päivät 2026 event was organized by FCG in May 2026. In the event, we heard Meri Utriainen and Johanna Pakarinen from HUS presenting on what patient flows can be automated in specialty care. I want to share my report as someone in the audience on what they covered. This is one of those presentations that combines a real production case, honest numbers, and an operational learning that translates directly to every healthcare provider running similar volume work. From the healthcare automation point of view, there’s three things from their talk worth breaking down: the challenge they set out to solve, the real numbers from two years of production, and the analytical takeaways for what this means at scale.

Full disclosure before going further: Digital Workforce is the contracting partner on the HUS breast cancer follow-up solution discussed below. When a customer presents production results from a solution we built with them, our job is to listen to what they have learned.

The challenge: high-volume repeating work

From the patient flow point of view, breast cancer follow-up is HUS’s largest patient group. After the active treatment phase ends, these patients stay in specialty care for follow-up for up to ten years on a risk-based schedule. As Meri described it, this is a group of more than 6,000 patients followed for one to ten years. And inside that follow-up, the same process repeats over and over: imaging, blood tests, treatment feedback, symptom phone calls, digital channel touchpoints, appointments.

This kind of high-volume repeating work bottlenecks easily. The HUS team was very honest about the trigger point. The bottleneck grew significantly during COVID. From the operational point of view, the clinical need was clear: they needed more effective operational control and stronger anticipation of capacity across the whole follow-up population. And in parallel, they wanted to bring the patient into their own care, which is something people are already used to in every other area of their lives.

Ambitious target: One decision for the entire patient follow-up

The ambition Meri described was a really specific one. Instead of a professional revisiting the question “what is the next step for this patient?” every year, the goal was one order from the professional that covers the entire ten-year follow-up. The automation then runs that follow-up reliably in the background, year after year, only surfacing the patient’s case to a human when there is a clinical decision to make or an exception in the process. That is not a small ask. From the workflow design point of view, you are replacing episodic decision-making with a single configurable long-cycle workflow that has to hold up across thousands of patients, multiple subgroups, and ten years of edge cases.

The real numbers: impressive impact at scale

The solution has been in production for two years. Here are the numbers Meri and Johanna shared.

• 6,909 patients are now in the automation.
• In 2025, 52% of patients chose symptom-based follow-up over a scheduled appointment when they got their care feedback. That is a meaningful freeing up of resources on the professional side.
• One of the most interesting numbers from the whole presentation: interference demand has dropped by 47%. The HUS team was explicit that they expected the opposite: that as appointment touchpoints went down, patient anxiety would generate more inbound calls and questions. The data showed the inverse. When patients get their imaging on time and know what is coming next, they need less reassurance contact.
• In the three weeks observation period in from May 2026, the automation handled 6,768 tasks, and only 83 of those (1.2%) were escalated to a professional.

Looking at the wider results across suitable patient flows: more than 95% of patients in suitable flows can be moved to automated solutions. More than 95% of task in those flows are handled reliably and automatically. More than 95% of structured findings turn out to be neutral and lead to no intervention. And 50% of patients prefer symptom-based contact over scheduled controls.

The numbers are the takeaway. This is two years of live production at HUS, with the operational data to back every claim. From the credibility point of view, that is the difference between a use case and a proven use case.

Repeatable solution to scale out

I want to spend the most time on this section, because what makes it important is that the solution has the structural ingredients for replication, and the HUS team was honest about both what worked and what they would do differently. There are patterns from their talk that map directly to where healthcare automation is going.

The “one order, ten years” framing

Replacing episodic decision-making with a single configurable workflow is the unlock. It is what makes the same logic layer applicable across patient groups. HUS already named the next domains they are considering to scale this approach into: drug surveillance follow-up in dermatology and neurology, imaging-based follow-up for other cancers and for gene-mutation carriers and meningioma monitoring, and pathways that cross the specialty care to primary care boundary like coronary patient reocurrence prevention. From the scaling point of view, the same logic layer carries. You build the long-cycle follow-up workflow once. You parametrise the variations for each illness, each subgroup, each care plan. The multipliers on patient amounts here are not small.

The 47% interference demand drop is the counterintuitive finding worth studying

Most healthcare automation business cases lead with cost saving or speed. The real long-term economics often come from second-order effects nobody modeled in advance. The HUS team was open about this. The human resource liberation from the solution was actually larger than they expected. As Meri put it, they could have planned the change management to harness the full potential earlier. That is exactly the kind of operational insight that turns a proven use case into a transferable playbook. Every customer running similar volume work should hear that learning, and then design their own rollout with the change management in place from day one.

The deliberate separation of automation from AI

Johanna referenced their CEO’s line: when we can bicycle, why jump in a rocket. The breast cancer follow-up solution is algorithm-based process automation. But it is not AI, a medical device, nor does it make clinical decisions. It handles the logistics in the background. That is the right pattern, because it lets the AI investments go to the places where AI actually brings value: referral triage on roughly half a million referrals per year, imaging interpretation where AI is already in production, pre-visit data collection, language model support for structured documentation. From the technology selection point of view, start from the process need and choose the right tool for each step. AI is one component in the care pathway.

From the geographical scaling point of view, this proof case makes growth conversations grounded. This case is what proof market means in practice. Production cases like this one are excellent showcase for healthcare markets in the world, including UK NHS and the US. The NHS faces the same workforce pressure and the same productivity mandate that HUS is solving for. The same care pathway logic carries. The patient amount multipliers across geographies are even larger than the multipliers across patient groups inside Finland.

Wrap-up

First of all, the healthcare productivity challenge is real and getting bigger across every Western market. Workforce pressure, cost pressure, an ageing population, and rising demand for high-quality care are not going away. HUS is solving for this with a combination of automation and selective AI. And they have the production numbers to back the approach. Two years of operation, 6,909 patients, 6,768 tasks handled in three weeks with only 1.2% escalated.

The numbers from a single patient group already represent meaningful scale, but the real story is the multipliers ahead. Apply the same one-order-covers-ten-years pattern across the next patient groups Johanna named, and the impact compounds: drug monitoring, imaging-based follow-up across more cancer types and other long-term conditions, pathways that cross the specialty and primary care boundary. From the patient amount point of view, breast cancer follow-up is already large. The other long-term illness groups are larger.

And finally, the operational learning HUS shared openly is the kind of insight that makes a proven use case actually transferable to other providers. HUS could have planned the change management earlier to harness more of the resource liberation. The technology is one part. The process design is one part. The change management to make sure the freed-up clinician time is redirected to medically meaningful patient contacts is the part that determines whether the business case actually lands.

The bicycle and the rocket are both useful. HUS just showed us what a well-built bicycle looks like at scale, in production, with the numbers to prove it. That is what proven looks like. And that is the pattern we are building on as we scale care pathway automation from Finland into the UK and US healthcare markets.

The author, Juha Nieminen, is the global head of Healthcare business at Digital Workforce and a member of its Management Team. With more than two decades in sales leadership and business development across healthcare, IT, and other sectors, he focuses on bringing intelligent process automation to health and social care organisations. He holds an M.Sc. in Industrial Engineering and Management and is based in Helsinki.