Qing (Joanna) Xia

Mapping is a critical step for a hospital customer to get onboarded to a clinical decision support solution I managed. The customer needed to spend around 300 hours to configure its hospital formulary into a specific template. Then the in-house team needed to spend another 200 hours to map the template to our proprietary identifiers. This often delayed customer onboarding, and it made the cost of in-house onboarding very high because the hospitals needed to allocate significant clinical and IT resources.

In order to expedite onboarding and reduce operating costs, I led the in-house clinical, UI/UX, data science, and engineering teams to develop a tool, using machine learning to automate both the template population and identifier mapping. The tool had two types of users, and the cross-functional team members had a different understanding of what defined success and the MVP scope. I organized workshops on personas, user journeys, and requirements to get everyone on the same page before development started.

The solution is hosted on Azure and accessed by a set of REST API endpoints, developed using the Flask framework. AI prediction services use the TensorFlow framework and the UI uses React, Redux, and Ant frameworks.

The CDS solution generates alerts about problematic medication decisions within hospital workflows, delivered as APIs embedded into hospital EMR systems. Both hospital customers and in-house teams (product, engineering, clinical, and customer experience/implementation) lacked ways to collect data about user experience (e.g., action taken toward alerts, alert volumes, and patterns) to improve alert relevance.

Because it was hard to satisfy all persona needs with the first iteration, I started by aligning with functional leadership and senior management to focus first on customers and our in-house implementation team. Then I collaborated with users to define their desired user journeys and requirements.

I led a team of around 10 people, including clinical, UI/UX, and engineering, to develop a module that collects the raw inputs and outputs of each alert triggered in the hospital workflows, encrypts and phones home, and massages into structured data for Power BI report generation. As a result, in-house teams could access usage reports through the Power BI dashboard while customers could view .xlsx files.

The usage reporting module is hosted on Azure and accessed by a set of REST API endpoints, developed using the Flask framework.

The clinical solution is API-based and embedded into hospital EMR systems and, thus, does not have its own UI. Therefore, sales found it challenging to convey user scenarios and clear value messaging to prospects because it was hard for prospects to visualize how the solution could benefit their day-to-day work.

In order to minimize time to market, I aligned with sales and customer experience teams on having a clickable mockup as the demo tool instead of a real-time interactive one. I convinced senior management to invest in and prioritize the project against others competing for UI/UX resources.

I led a team of six, including UI/UX, clinical, customer experience, and product marketing, to define minimum elements to mimic an EMR UI and key scenarios to show to prospects within a 15 to 30-minute sales session and to launch the demo tool with detailed scripts within six weeks. Given the potential for international internet connectivity issues, we delivered an offline demo tool and video, using UXPin.

The offline demo increased prospect conversion from showing interest to taking action to try our solution by 40%.

I led a team of three to design a new centralized credit approval process for a mid-sized Vietnamese bank. Before approving the rollout, the client's senior leadership wanted to see a successful dry run in a few branches and the new center of credit approval. However, the dry run processed hundreds of loan applications per day while the bank employees were not capable of using Excel to convert raw timestamps into step-by-step time spent.

In order to generate immediate results for the dry run, I developed a macro-based Excel model for the bank employees to use. All they needed to do was copy and paste raw timestamps for each loan application and click a button to display the time duration of each step for each application.

This tool enabled stakeholders to see, in a clear and visual way, the end-of-day time savings brought by the new credit approval process, and eventually led to senior leadership's approval of a full rollout right after the dry run was completed.