How to set up knowledge base for wellness program enrollm…

Gather the documents and information your employees use to learn about, decide on, and enroll in the wellness program. A strong knowledge base answers the questions that normally come through email, intranet searches, or HR tickets, so your AI agent can handle them around the clock.

What to prepare:

• Program overview and benefits – description, covered services, provider information, and how to use the program.
• Eligibility and enrollment windows – who qualifies (full-time, part-time, dependents), enrollment deadlines, and waiting periods.
• Step-by-step enrollment instructions – the exact workflow: how to log in to the benefits portal, select the wellness plan, confirm selections, and submit.
• FAQs – documented answers to the most common questions your HR team receives, such as cost, covered activities, reimbursement procedures, and what happens after enrollment.
• Policy and fine print – any rules around mid-year changes, life-event exceptions, and plan restrictions.

Organize this content into one or more files (PDFs, text documents, or a public intranet page with a URL). The clearer and more specific the source material, the more accurately the agent will answer.

1. Start a new agent in Chatref

Log into app.chatref.ai. If you are new, your account includes $50 in free credit with no time limit—enough to build and test your knowledge base before going live. Create a new AI agent and give it a name that employees will see (e.g., “Wellness Program Assistant”). The agent will be the chatbot that draws answers from your knowledge base.

2. Add your wellness program content

Under the agent’s settings, navigate to the Knowledge base section. You can upload files directly or provide a URL to a page that contains your enrollment information. Supported formats include PDFs, text files, and public web pages. Upload the documents you prepared in the previous section: program overview, eligibility rules, enrollment steps, and FAQ.

Chatref reads this content and indexes it so the agent can retrieve the right passages when a question arrives. No categorization or tagging is required—just add your files. If you maintain an internal benefits page already, pointing the agent to its URL is the quickest way to keep it up-to-date; any changes on the page are reflected in the agent’s answers after a quick re-scan.

3. Let the agent learn your content

After adding sources, Chatref processes them automatically. The system extracts the text and builds a searchable knowledge base. This usually completes in a few minutes. Once the status shows “Ready,” your agent is grounded in your own program information and will not guess or pull from the open web.

4. (Optional) Adjust the agent’s voice and tone

While the agent is grounded in your docs, you can also set a custom welcome message and choose a style that matches your company’s internal tone (for example, friendly but professional). This is done under the agent’s Settings tab and is entirely optional. The core accuracy comes from the source content, not from prompt engineering.

Before sharing the chatbot link with employees, test the agent’s answers against realistic enrollment questions. Use the built-in playground (available on your agent’s dashboard) so that tests do not cost any credits.

Run through these types of questions, which mirror what employees actually ask during enrollment periods:

• “How do I sign up for the wellness program?”
• “Am I eligible if I work 20 hours per week?”
• “What is the deadline to enroll this year?”
• “Can I add my spouse after the enrollment window closes?”
• “Do I have any out-of-pocket cost for the program?”

Verify that each answer is:

• Accurate – the information matches what you uploaded.
• Complete – the answer includes the necessary steps or conditions, not just a fragment.
• Grounded – the agent cites its source or shows it is pulling from your content (look for the “source” indication in the response).

If an answer is off, go back to the knowledge base and add the missing information or rephrase unclear parts. Because the agent retrieves content directly, fixing the source document improves future answers immediately.

The agent gives an outdated enrollment deadline

This happens when the knowledge base contains old dates or multiple conflicting PDFs. Remove any source that has stale information, upload a single current version, and re-test. Keeping one master file and updating it each enrollment period avoids this.

Answers reference the wrong plan or location

If your organization runs multiple wellness programs or regional variations, separate documents for each (e.g., “US Wellness Program” vs. “UK Wellness Program”) help the agent stay on-topic. When employees specify their location or plan name in a question, the agent can retrieve from the correct subset.

The agent says it does not know

This is usually a content gap. The knowledge base might not cover the exact scenario asked—for example, a question about a specific life event that is not mentioned in your policy document. Add that detail to your FAQ or policy file and re-test. In production, the agent can be configured to escalate unknowns to a human, so no question is lost.

Answers feel too generic or missing practical steps

Generic answers often come from documents that describe the program benefits but lack the actual enrollment workflow. Include an explicit “How to Enroll” section that lists each mouse click or portal navigation step. The more operational detail in the source, the more actionable the agent’s response.

Unclear eligibility guidance

Eligibility rules can be complex (multiple employee types, dependents, waiting periods). Present them in a clear bulleted list within the knowledge base document, using plain language. Avoid legal jargon unless necessary; the agent will reuse the same phrasing in answers, so plain text leads to clearer responses.

What causes wellness program enrollment chatbot problems for Corporate Wellness Programs?

Most problems stem from incomplete or outdated knowledge base content. If the source materials lack the full enrollment workflow, miss recent policy changes, or mix information from multiple years, the chatbot will produce inaccurate or contradictory answers. Another frequent issue is not updating the content when enrollment windows open—employees get last season’s deadlines. Finally, if the agent’s retrieval does not handle nuanced eligibility questions (e.g., for COBRA participants), adding explicit Q&A pairs for those edge cases resolves the gaps.

How do I improve wellness program enrollment chatbot for Corporate Wellness Programs?

Refine the source documents: keep them updated each enrollment cycle, use clear language, and structure them with distinct sections for eligibility, steps, exceptions, and contacts. Test each enrollment season with a set of standard employee questions and add any uncovered scenarios to the FAQ file. If your program spans multiple regions or employee groups, create separate knowledge base sources rather than one large document, so the agent stays precise. Also, review the conversation insights in Chatref to see the actual questions employees ask—this data highlights the exact topics where your knowledge base needs expansion.