Product
·
6
min read
Product insights: Building a voice interface for health record systems

Jana Bakešová
Jun 16, 2025
Healthcare might not be the first field that comes to mind when thinking about heavy administrative workloads, yet doctors often find themselves buried in paperwork. At Applifting, we looked at this challenge from a new angle—developing a voice-controlled solution that frees up doctors' hands and lets them focus on what truly matters: helping people.
Easing the admin overload for medical professionals
The health industry is facing a staff shortage, with administrative burden overwhelming healthcare workers and pushing them to the brink of burnout. We were approached by a health tech company looking for the most efficient way to alleviate this burden.
Voice control seemed like a viable solution that would allow users to quickly navigate the system and perform administrative tasks while keeping their hands free.
At the heart of this challenge were key questions that shaped our approach:
When it comes to clinical tasks, is it more important to prioritize speed—allowing doctors to give quick, straightforward commands—or to maintain a more formal approach and tone of communication?
How might background noise in a busy clinic affect a doctor's willingness to use voice control?
How do everyday experiences with voice assistants like Alexa or Siri translate to this very specific work environment?
Mitigating risk with real user insight
To build the best possible voice interface, we first had to understand the needs of the users—the clinicians. We focused on a few specific use cases for the minimum viable product (MVP) based on the tasks clinicians perform frequently, voice command suitability, how competitors approached similar features, and technical feasibility.
When selecting the clinicians for testing, we considered their workload, how frequently they used the system, and the environment in which they interacted with it.
Talking to the clinicians was a very unique opportunity, as their time is limited and expensive. Hence, we did our best to be as prepared and concise as possible. By the time we had the interviews set up, we already understood how the chosen use cases worked within the system, prepared a high-fidelity prototype with a voice layer using a tool called Protopie to demonstrate future features, and an interview script that reflected our hypotheses.
After conducting a qualitative analysis of research outputs, our path forward became clearer. We understood the preferences, fears, and concerns regarding the workflows, as well as the situational implications. Most importantly, the riskiest assumptions were validated—reducing the chance of major setbacks later in development. What a great start to build the product!
Let's build a voice assistant!
While our development team was on a mission to bring the designs to life, we on the product team focused on shaping the assistant’s behavior using Dialogflow CX, a tool for building conversational experiences that also served as the logic layer. This setup gave us the flexibility to directly influence how the assistant responded to users.
The separate logic allowed us to train the language model independently from the development. We also experimented with the training phrases to give users some level of flexibility while ensuring the system correctly identified the proper flow.
This was easier said than done. Distinguishing between similar flows, handling different commands, and allowing for flexibility were all challenging tasks. We conducted a lot of experiments to tune the model.
Ready, set, release?
Not exactly. We intentionally did not wait until the assistant was polished and perfect. Once it was ready to use—despite known bugs—we organized usability testing with the people from our client's departments who are close to the customers. The usability testing allowed us to test the feature in its early form, assessing approximate adoption without worrying about the reputation risk.
New bugs, usability hiccups? Sure, not everything went smoothly. But to us, that kind of feedback is a win, not a catastrophe. It showed us exactly what to improve before releasing to the first clinic and gave us valuable ideas for future enhancements
For instance, one hiccup was a lack of understanding regarding the assistant's states. In human conversations, verbal and non-verbal cues help us know when to speak or pause. Similarly, a voice assistant must provide clear, visible signals to communicate its status. The usability testing highlighted the need for this clarity. We analyzed all the relevant technical states, matched them with user expectations, and added animations to make the interactions more intuitive and enjoyable.
User interviews, usability testing, usability improvements, bug fixes—you might think it was finally time to launch. Not just yet. We still had some hypotheses related to the clinicians and security that couldn’t be validated during the internal testing. So, one final testing round was organized before the release. Better safe than sorry.
Beta launch and handoff
With the beta release ready, we prepared walkthroughs, communication strategies for the clinicians, success metrics, methods for gathering feedback, bug reporting, and prioritization plans. Our client’s team also contributed by adding articles to the system's support page to guide users.
This concluded our effort, and we were ready to start beta testing and hand the project over to the client. At Applifting, we value transparency and independence. Therefore, we provided all the reports, documentation, and code to our client, allowing them to continue at their own pace.
Job well done
What an adventure—turning an idea into a real, usable product! Along the way, we shaped the assistant through continuous learning, user feedback, and iterative problem-solving. We tackled challenges head-on to ensure a smooth delivery, never losing sight of the clinicians' needs. Our client was an excellent partner throughout—collaborative, open, and deeply engaged in building something meaningful.