Evaluating AI at RSNA 2026?
Ask What Work It Actually Removes.
RSNA is one of the best places to compare medical imaging AI side by side. You can see new models. Talk directly with vendors. Explore new clinical and operational use cases. And get a sense of where the market is heading.
But with so much AI on display, it can be easy to leave with a long list of interesting technologies and not enough clarity about which ones are actually worth taking further.
So rather than asking only:
What can this AI do?
Ask:
What would change if we actually adopted it?
Here are seven questions worth taking into those conversations.
What problem is this AI actually solving?
Start with the problem, not the model.
What is difficult today? What clinical, operational, or workflow issue is the AI intended to address? Who experiences that problem? How often does it happen? And what does the current workaround look like?
This helps separate technology that is impressive from technology that is genuinely relevant.
Ask the vendor:
"If we did not buy this AI, what problem would still remain?"
If that answer is unclear, the value proposition may be unclear too.
Who actually uses it, and when?
AI demonstrations often focus on what the technology produces. Ask who is expected to act on that output.
And at what point in the workflow does that happen? Understanding the user and timing can reveal whether the tool fits naturally into an existing process or creates a new one that someone has to own.
Ask the vendor to be specific:
"Who interacts with this in a normal working day, and what do they do differently because it is there?"
That answer is often more useful than another feature list.
What changes in our current workflow if we adopt it?
Ask the vendor to show the before-and-after. Not just the AI screen.
- What happens today?
- What happens after deployment?
- Which step disappears?
- Which step changes?
- Which new step is introduced?
- Who needs training?
- Who needs to check or validate the result?
- What happens when the AI does not produce an answer?
This is where you can start to understand the real workflow impact. A useful AI tool should have a clear place in the process.
If the workflow becomes harder to explain after the AI is added, that is worth noticing.
What does implementation really involve?
A successful demonstration is not the same as a successful deployment. Ask what needs to happen between the booth and routine clinical use.
What systems need integration?
What data needs to be available?
What validation is required?
How much configuration is involved?
What governance is needed?
Who supports the implementation?
Also ask about time:
"What does a typical path from contract signature to routine use actually look like?"
The implementation burden can be just as important as the capability itself.
What evidence should we expect before scaling it?
A pilot can show that the technology works. That does not automatically mean it should be rolled out more widely. Ask how the vendor recommends measuring success.
Depending on the use case, that might include clinical performance, adoption, time saved, reduction in manual work, turnaround impact, workflow consistency, operational efficiency, or quality measures.
The key is to agree on what success looks like before the pilot begins.
Then ask:
"What exact evidence would tell us this is worth scaling?"
That turns the pilot into a decision process rather than simply a technology trial.
What happens after the pilot?
Many AI projects begin small. One site. One specialty. One workflow. One group of users.
The harder question is what happens when the organization wants to go further. Ask what changes when the AI is rolled out across more sites, more users, higher volumes, different workflows, or different specialties.
A pilot that works well in one controlled setting may still create challenges at scale.
Ask the vendor:
"What usually becomes harder when customers move from pilot to broader deployment?"
That question can surface practical issues that may not appear in the demonstration.
What would make us stop using it?
This is a question vendors may not hear often enough.
Not every AI project will deliver what was expected. Clinical priorities change. Better technology appears. Adoption may remain low. The workflow may change. The business case may weaken.
So ask what happens if the organization decides to stop. How portable is the data? What happens to generated outputs? Are there contract restrictions? How difficult is it to remove the integration? Does the workflow become dependent on proprietary components?
The takeaway:
A good AI decision should include a path in and a path out.
Do not leave RSNA with only an impressive demo
AI demonstrations are designed to show what technology can do at its best. Your job is to understand what happens after the demonstration.
Before you leave the booth, try to answer:
- What problem does this solve for us?
- Who actually uses it?
- What changes in the workflow?
- What does implementation require?
- How will we decide whether it is working?
- Can it scale beyond a pilot?
- How difficult is it to change direction later?
Those answers will tell you much more about whether the AI is worth pursuing.
Want to go deeper on workflow?
Our view is that clinical AI should reduce unnecessary work rather than create another layer for radiologists to manage.
Is your environment ready?
Evaluating the AI itself is only part of the decision. Your imaging environment also needs to support the data, connectivity, and workflow.
Bring your AI questions to RSNA 2026
If AI is part of your RSNA agenda, bring the use case you are evaluating, the workflow you are trying to improve, or the pilot you are trying to scale. At AdvaHealth, we can start with the problem and work forward from there.
