From an AI interview to information useful for care

By using both sources side by side, we discovered we were giving nearly equal weight to two kinds of content that didn’t hold the same value for the professional.

The transcript could contain errors —for example, misreading a sentence or confusing speakers— that didn’t always carry over into the analysis. By keeping both pieces of content constantly facing each other, the interface invited comparison and put too much focus on transcript errors when what they actually needed to work on was the analysis.

This led us to rethink the hierarchy:

Since the AI already had the interview and analysis information, we proposed adding a second layer of information: an AI-generated summary that would serve as a starting point, which each professional could review and adapt.

We weren’t trying to reproduce each center’s particular format. The goal was to find a common structure that could be useful across different care contexts.

The question then shifted from only what information to summarize to also deciding how to present it so it would actually be easy to consult.

We took both proposals to a first test with professionals to compare which type of information was most useful.

The test helped us move toward a more direct, organized structure, closer to the type of information professionals needed to consult and later adapt.

But it also let us detect a different problem.

In that first proposal, Analysis and Summary already appeared as the two main outputs, while the transcript remained available as supporting information. However, the Summary tab stayed inactive until the professional approved the Analysis.

During the test we observed that professionals didn’t try to explore a tab they perceived as disabled.

The intention was to show that a step came after the Analysis, but the inactive state achieved the opposite: the Summary was present in the interface, but it didn’t help convey what would happen after approving.

This led us to change the interaction: Analysis and Summary needed to stay visible and accessible, even when the Summary hadn’t been generated yet.

This first validation therefore produced two decisions that would shape the next version: moving forward with a more structured, direct Summary, and keeping both sections accessible to make the path between Analysis and Summary understandable.

With the new architecture, Analysis and Summary moved to the first level of Life Story, while the transcript remained available as supporting information.

The Analysis kept the complete content organized by life stages and was the space where the professional could review and correct the AI’s work.

The Summary offered a shorter, more structured version, designed to quickly locate the most relevant information.

After the first validation, both sections remained visible and accessible, making it more evident that they were part of the same process.

Adding the Summary also expanded Life Story’s role within Amaia Conecta.

We added a dedicated Life Story section within each person’s profile. Once the professional had reviewed and approved the Analysis from Amaia Profesionales and the Summary had been generated, it could also be consulted from the tablet.

Content management stayed on the computer; Amaia Conecta worked as a place to consult the synthesized result.

This avoided moving all the complexity of review and editing onto the tablet, and kept a clear function for each environment.

The tests also revealed more isolated issues: confusion about what Regenerate did, low visibility of some editing tools, or differences in how certain flow elements were used.

We logged them as opportunities for improvement, but didn’t treat them as general problems when the evidence only appeared in one or a few sessions.

Despite the friction along the way, the sessions reinforced two fundamental decisions.

The first was the value of the Summary. When professionals reached it, they understood its role as a shorter version of the Life Story and found it useful to be able to consult and adapt it.

The second was keeping the professional within the generation process. During the tests we observed how they reviewed, corrected, and adapted the AI-generated content before approving it, confirming that professional validation fit how they actually worked.

The sessions showed that, once discovered, professionals understood the difference between the full Analysis and the Summary, and could decide when they needed to review, modify, or simply consult the result.

The value wasn’t just in shortening information. The Summary allowed what was learned during an interview to also remain available to other professionals who later cared for that person.