FirstThought / Clinical expert discovery
Finding the person
who knows.
- Role
- Founding Member & Product Design Lead
- Period
- 2019–2022
- Product
- ML-powered expert-network platform
- Reach
- 25-person analyst team · 7,500+ expert candidates vetted annually
- Team
- 2 Engineers · Compliance & Partnerships · Head of Research · Business Development
Why I helped start FirstThought.
We brought machine learning to expert discovery for clinical research. We sourced information from ClinicalTrials.gov, PubMed, and other scientific sources. As one of the founding members and the product & design lead, I defined the roadmap and was the sole designer across the product, its identity, and the research operation.
Manual exploration
Finding the repetitive work
we could automate.
Research Managers were piecing together expert evidence one source and one page at a time. I began with ClinicalTrials.gov to understand that work firsthand and identify which steps were repetitive enough for the product to take on.
We wrote a simple JavaScript scraper, ran it on individual pages, extracted the raw text, and experimented with structuring the fields that could point to a relevant expert; Automate the collection and organization work, allow the Research Managers to spend more times with face-to-face client engagements rather than spending time in a CRM system.
Manual exploration
My technical prototype / Iterate on extraction and structure
- ClinicalTrials.gov
- Visit an individual page
- Run my JavaScript scraper
- Extract raw page text
- Structure useful fields
- Can we identify an expert?
What the prototype exposed
Visiting pages, extracting text, and structuring fields were repeatable tasks that did not need to consume a Research Manager’s attention.
What guided the product
Automate the repetitive work, then present the evidence clearly so Research Managers could focus on expert fit, client context, and judgment.
Product strategy and roadmap
Building the platform
one constraint at a time.
We began with a CRM. The rest of the roadmap followed the work: what was slowing the research team down, and what would help them next? That took us from expert records to discovery, then email and SMS.
1 · Foundation · 2019
CRM
Built the foundation for managing experts and the research operation.
- Expert records
- Research activity
- Outreach tracking
2 · Discovery
Automated discovery
Connected the CRM with our scraper and data-enrichment tooling.
- Clinical and scientific data
- Expert discovery
- Phone-number discovery
3 · Workflow
Email integration
Connected staff email through Microsoft Power Automate.
- Staff communications
- Email workflows
- Operational integration
4 · Engagement
SMS outreach
Twilio for text outreach and automated follow-up replies from our system.
- Multichannel outreach
- Automated follow-up replies
- Less manual work for staff
Why this order?
The team needed somewhere to keep expert records before we could improve how those records were found and filled in. Once discovery improved, another problem came into focus. We could find the right person and still struggle to hear back.
Build around the next constraint
Each phase addressed the next constraint in the research workflow. Integrating email brought staff communications into the operation. Adding SMS extended that workflow to another channel and let the system send follow-up replies automatically, removing a manual task for research staff.
Experience → hypothesis → decision → outcome
Bringing a lesson from patient feedback
to expert outreach.
An earlier lesson proved useful here. At Quality Reviews, I had seen text messages draw stronger responses than email. FirstThought still relied heavily on email, and I wanted to find out whether the same lesson would travel.
We added Twilio so researchers could reach experts by text within the existing workflow. The system also sent follow-up replies automatically, giving staff one less task to handle by hand. After adding SMS, response rates increased 1.5×.
Product walkthrough
From a research question
to a consultation-ready expert.
Follow the workflow from a new project through expert discovery, screening, and client review.
Workspace dashboard
The dashboard gives the research team a starting point for active projects and the work that needs attention.
Create a project
A new research question becomes a project with a shared brief and a place to organize the search.
Expert search & discovery
Researchers explore potential experts and narrow the pool against the project’s needs.
Candidate review
Candidate review brings the relevant people into a working pipeline for closer evaluation.
Expert profile
Each profile puts background and supporting evidence in context so researchers can judge fit.
Outreach & screening
The team coordinates contact and screening while keeping candidate progress visible.
Client shortlist
A focused shortlist gives clients a clear set of experts to review for the consultation.
Project workspace
The project workspace keeps the research and engagement overview together as the work moves forward.
Outcome
The product shipped and reduced research from several days to a few hours for a 25-person analyst team. A complementary vetting workflow processed more than 7,500 expert candidates annually, and adding SMS outreach increased response rates 1.5×. The company ultimately expanded faster than its funding could support and ran out of funds.
What I learned
How I learned to build from zero.
I learned how much a small, working experiment can clarify. The scraper gave us evidence to discuss; the roadmap gave us an order in which to act. Working with our PM and two engineers meant making decisions concrete enough to build, then learning from what we had shipped.
I also learned that product design includes knowing what technology is available to help. Power Automate gave us a way to connect staff email to the workflow; Twilio let us add text outreach and automate follow-up replies. Understanding those possibilities helped me shape the roadmap and have better conversations with engineering. Part of my responsibility was to look beyond the interface, recognize when an existing service could solve a problem, and work with the team to decide whether it belonged in the product.