Automating Healthcare Site Verification with OpenFn and AI

Every month, an organization that provides technical assistance to a network of physicians receives thousands of applications from health workers around the world. Each applicant lists details such as their name, ID, and affiliated healthcare site. Before any applicant can be approved, the team must verify that these healthcare facilities actually exist and are legitimate.

This verification process; checking hospitals, clinics, or organizations across multiple countries was entirely manual. Staff members had to search for each site online, read websites, cross-check official directories, and then decide whether to approve or reject the application. It was slow, repetitive, and hard to scale.

The Problem

• Thousands of applications monthly, each requiring a site legitimacy check.
• Manual verification was consuming significant staff time.
• Many applicants were from countries with inconsistent or incomplete registry data.
• The organization wanted to explore how AI and automation could accelerate this process without compromising accuracy.

Implementation Approach

The idea was to combine OpenFn’s workflow automation with a large language model (LLM) to simulate a “research assistant” for the review team.

We built an OpenFn workflow that:

1. Read applicant data directly from a Google Sheet, which included:

• Applicant ID
• Application Name
• Site Name
• Country

1. Sent each record to an LLM (ChatGPT’s 3.0 Deep Research model) with a structured prompt.

• The model was instructed to:
• Confirm whether the listed healthcare site exists.
• Identify its online presence (official website, Facebook, none).
• Evaluate the site’s credibility based on available information.
• Return a structured JSON response summarizing findings, status, and confidence level.

1. Updated the same Google Sheet with the verification results—automatically tagging each application as:

• Pre-Approved (legitimate site found with high confidence)
• Declined (no evidence of legitimate facility)
• Needs Review (insufficient or conflicting evidence)

By structuring the AI output as valid JSON, we ensured the workflow could programmatically parse results and feed them back into the review system with no manual intervention.

Workflow Delivered

Here’s how the workflow ran end-to-end:

1. Trigger: A new batch of applications was added to the Google Sheet.
2. Data Extraction: The workflow parsed relevant columns and created a list of applicant objects.
3. AI Verification: For each applicant, the LLM performed a guided web search and produced a structured output (status, confidence, reasoning, and funding source).
4. Result Aggregation: All AI outputs were collected into an array of verified records.
5. Write-Back: The Google Sheet was automatically updated, saving hours of manual checking.

This approach effectively turned OpenFn into a bridge between structured applicant data and an AI-powered reasoning engine capable of doing lightweight due diligence on healthcare facilities worldwide.

Learnings & Reflections

1. AI is powerful, but guidance matters.
   Carefully designing the prompt and JSON schema made the difference between a vague answer and a reliable, parseable response. This meant having explicit field names and enforcing strict JSON formats.
2. Human-in-the-loop remains essential.
   The AI could triage applications quickly, but ambiguous or low-confidence cases still required a human reviewer.
3. Transparency builds trust.
   Including the model’s reasoning and any reference links in the output gave reviewers a clear audit trail for every decision.

See It In Action

Watch this walkthrough to see exactly how the OpenFn workflow connects Google Sheets to ChatGPT's Deep Research model, processes applicant data, and automatically updates verification results.

Final Thoughts

This prototype showed that AI-augmented verification is a realistic way to handle large-scale applicant screening, especially for organizations working across multiple countries and data sources.

By combining automation with structured reasoning, we reduced the verification workload, introduced consistency in reviews, and created a foundation for future improvements like registry lookups and domain whitelisting.

What started as an experiment in connecting Google Sheets to an LLM turned into a practical example of how OpenFn workflows and AI can complement human judgment in global health operations.