Healthcare is a complex field
Working with health tech companies means stepping into a deeply specialized field. Unlike other industries, healthcare is filled with unique terminology, constantly evolving research, strict regulations, and a complex stakeholder ecosystem. These challenges require more than just surface-level understanding. They demand structured knowledge acquisition.
Take, for example, the difficulty of understanding oncology. Cancer treatments involve an interplay between pharmaceutical innovations, regulatory approvals, clinical trial phases, and the workflows of oncologists, pharmacologists, nurses, and insurance providers. Each of these stakeholders has different concerns: clinicians want treatment efficacy, pharma companies focus on drug development and approval, and payers assess cost-effectiveness. Feels complex already? Well, as an example, this barely scratches the surface. Oncology alone is an entire universe, with distinct worlds within it—diagnosis, prognosis, treatment, and more. Entering a space without a structured knowledge process is a recipe for misalignment and wasted effort.
This is why we’ve been experimenting with ChatGPT capabilities to accelerate how we build contextual understanding when conducting a deep dive into a new field. Now that Deep Research and Projects are widely available, I’ve taken this even further, making AI a more active collaborator in our desk research process.
Our process for Leveraging ChatGPT in Desk Research
This is a documentation of how we are experimenting with ChatGPT (the Plus version) for Desk Research currently. We are still refining the process and it’s not intended to be followed as a guide, but rather to share our experience and spark some conversations about it.
We keep it all organized into a ChatGPT project
When starting our desk research activities, we configure a ChatGPT Folder for the project. The best part about these folders is that they maintain project-specific context while allowing separate conversations on different topics. In the instructions section, we add a prompt for adding context to it. It contains the customer goal, the project goal, the project context, and our team role.
The thing I’m missing the most from this functionality is sharing the folder with my colleagues. It will be much better if we can all collaborate over the same project, each one adding conversations about the different aspects we explore.
Request Documentation from the Customer and process it
Ask for materials like pitch decks, Canvases, whitepapers, published research, and regulatory documentation. This helps identify how the customer frames their problem and approach.
We annotate key terminology, stakeholders and processes involved, phrases highlighted as differentiators or moats, competitors, and questions. We then use ChatGPT’s search functionality to clarify specific terms and concepts, and widen our view from the customer perspective.
Example: If a client is working on digital therapeutics for diabetes, we might search for recent advancements in CGM (Continuous Glucose Monitoring) or regulatory pathways for software as a medical device (SaMD). Even when a topic isn’t a core part of the client’s focus, staying informed on state-of-the-art trends and potential disruptors helps us stay aware and enrich conversations with subject matter experts.
Draft an Initial Product Canvas
At Arionkoder, we use a streamlined version of the Lean Business Model Canvas, focusing on four key questions:
- Problem Statement: What core issue are they solving?
- User Personas: Who are the key stakeholders?
- Value Proposition: How they’re framing the solution and what makes their approach unique?
- Revenue Streams: How does the business model sustain itself?
We create a first draft with what the client has shared and what we learned, and then we use a session to check our understanding with them. We might have got some details wrong, but they always appreciate seeing our progress in understanding them.
The process finishes with going throught the project instructions and modifying the context with what we have learned.
Using Deep Research for Market Context
All of the previous steps are now used as context for generating a Market Analysis report with Deep Research.
For this, we activate the deep research feature and provide a structured prompt that looks like this:
Aspects to Cover
I need an in-depth market analysis on [specific market] to understand [key objective, e.g., emerging trends, competitive landscape, consumer behavior, growth opportunities].
The analysis should include:
- Market Trends: Identify emerging patterns, new technologies, and changing consumer behaviors.
- Competitive Landscape: Overview of key players, recent strategic moves, and differentiators.
- Customer Insights: Key consumer segments, unmet needs, and pain points.
- Regulatory & Economic Factors: Any new policies, financial constraints, or government initiatives influencing the market.
- Opportunities & Risks: Areas for potential investment, risks to watch, and expected future shifts.
We have already gathered some information, that is shared below:
Current Context: […]
Valuable Sources Identified:
- [Source 1: URL]
- [Source 2: URL]
- [Source 3 URL]
Output Format: It needs to contain an Executive Summary (Concise key findings), the Detailed Market Insights (Structured by the aspects listed above) and Data-Backed Observations (Include references where possible)
We refine this prompt based on each project’s needs. Deep Research will ask follow-up questions before generating the report.
In our experience, this is the most valuable and time-saving addition to our process. In previous projects when we had the time to do this manually, it consumed a week of work for a good enough report. ChatGPT’s Deep Research won’t provide an excellent report, but it creates this good enough piece in minutes.
But does this make us better thinkers, or just faster ones?
Exploring and curating learnings from the Market Context Report
The final step is human curation. We spend an hour reviewing the generated insights, new information, and sources, assessing their relevance to the project. If any insights seem unclear or incomplete, we go back and forth with ChatGPT, requesting related articles or additional context.
Once we’ve identified the most valuable findings, we share them with the client for alignment. After agreeing on their relevance, we incorporate them into our Miro boards and ChatGPT’s project context.
Key learnings
As research is an iterative process, incorporating Deep Research too late or too early in the process makes it less useful. A well-defined initial manual research phase prevents AI-generated insights from becoming overwhelming or irrelevant, but the more time you use for initial manual research, the less valuable and time-saving the report will be. That’s why a structured, timeboxed process helps with the initial part.
Human curation is essential because ChatGPT struggles with selecting truly valuable insights. When asked to curate, it often generates generic summaries or simply rephrases everything to avoid omission. Without critical human judgment to assess relevance and depth, the results can feel superficial. While Deep Research accelerates information gathering and synthesis, its true value comes when paired with human framing and interpretation.
Using AI this way when we enter new Health tech projects ensures that we start smarter, work faster, and collaborate better with clients, without losing the depth and nuance that human expertise brings to the table.