Open enrollment is the one HR project where a dozen owners, a stack of carrier deadlines, and a very sensitive audience all collide. I’m living it right now. We are running open enrollment at Warp this week.
If you missed the first two, I covered onboarding and employee relations. Both use the same four-part prompting framework: context, the ask, what good looks like, and guardrails. Every prompt below follows it.
Where AI belongs in benefits
Some of you are probably asking whether it is safe to use AI for benefits at all. My answer: yes, with the right guardrails and systems. Here is how I think about it.
Use AI to draft, organize, and model. Benefits means a lot of documents and a lot of conversations, and summarizing is one thing AI is very good at. Use an approved tool, and use placeholders for anything personal.
Stay in the driver’s seat. AI should never make the call. You wouldn’t let it decide a hiring outcome, and you shouldn’t let it pick a plan. It can help you think through a decision. It should never tell an employee what is right for their health or their family.
Keep health information out. Benefits touches PII and PHI. Never upload a diagnosis, a claim, or anything you wouldn’t be extremely careful with. General plan documents and company-level numbers are generally fine. We use the enterprise version of Claude at Warp, which comes with data protections that other tiers don’t have, so check with your IT and security team on what your company allows.
Lean on your broker for compliance. AI is great at building checklists and organizing information. For compliance notices, filings, and claims situations, lean on your benefits broker or advisor. If you’re not sure what a broker does, here’s a breakdown.
Prompts for open enrollment
Prompt 1: Build the open enrollment project plan
Open enrollment is a great AI use case because it has a dozen owners and carrier deadlines that can’t slip. Many companies don’t have a broker building this level of plan for them. This prompt puts that power back in your hands. It works backward from your dates and hands you a week-by-week timeline plus a tracker with owners and due dates.
Here’s the prompt. The company is fictional, so swap in your own details.
I'm the HR lead at Northwind Labs, a 38 person remote-first company with employees in CA, NY, TX, FL and WA. Our health coverage renews Jan 1, 2027. Employee enrollment runs Mon Nov 2 to Fri Nov 13, and our carrier needs final elections by Fri Nov 20. Today is Oct 19.
Build an open enrollment project plan working backward from those dates. Give me:
A week-by-week timeline starting the week of Oct 19. A tracker table with columns for task, phase, owner, due date, status, and dependency.
Include compliance notices and plan documents, finance sign-off, payroll deduction setup, employee materials, a manager briefing, a reminder cadence, a post-enrollment audit for missing or incomplete elections, and a payroll check before January paychecks. Put compliance first. Keep the tracker to 15 tasks or fewer. Use roles as owners, not names. Flag anything that depends on my carrier's actual deadlines, and tell me which notices I should confirm with my benefits advisor.
Three things in this prompt do the heavy lifting:
- Work backward from your deadlines. Give the model the constraint that matters. It fits the plan to your calendar instead of a generic one.
- Put compliance first. The model only prioritizes compliance because you told it to.
- Flag anything that depends on the carrier. This tells the model there is uncertainty in what it produces. A date it can’t know is a date you verify, not one you trust.
When I ran it live, the output surfaced follow-up questions for me to confirm and mapped out ownership for each step: who confirms the carrier deadlines, formats, and windows; who collects the plan documents and SBCs (the benefits advisor); and who owns communications (the HR team). If you already have an enrollment process with set owners, feed that context in and the tracker updates to match.
Prompt 2: Turn the plan into a live tracker
The plan is useful, but honestly it’s long. So I don’t stop at a static document. I turn the tracker into something I can work from:
Turn the tracker into a live interactive spreadsheet that auto-updates days until due with a status dropdown, a days-until-due column, and a count of overdue tasks.
One thing I love about these tools is that they no longer only summarize. They build assets: first documents, then spreadsheets, now small apps you can host for your team. I’d encourage you to test the limits of what yours can create.
From there, I download the spreadsheet and upload it to our issue tracker, Linear, so I can assign every step to an individual. If your team lives in Google Sheets or email instead, the spreadsheet works there too.
Prompt 3: Build the employee enrollment guide
Once you have a plan, the next job is communication. Nobody is going to read a 40-page SBC. Employees need to understand the plans, why you designed them the way you did, and which one fits their life.
I build the guide in Claude Design, attaching the Summary of Benefits and Coverage (SBC). You can pull plan data from your carrier’s portal, or ask your broker or advisor for it. Here is the prompt:
Create a one page "Open Enrollment at a Glance" guide for Northwind Labs using the plan data I pasted. Audience: employees, mostly remote, many first-time enrollees. Include a timeline strip for Nov 2 to Nov 13, two plan cards (Plan A: Gold PPO and Plan B: Silver HDHP) with deductible, out-of-pocket max, and visit costs, a "do this week" checklist, and who to contact. Warm, modern, plain language at an 8th grade reading level. Use [brand colors / our design system]. Add a footer: "Illustrative only. Check your plan documents for full details." Do not invent any numbers, data, etc. Confirm with me if you have any questions or gaps to fill in.
Three things to notice:
- Use the plan data. The model works from your facts, not from its training data or a vague memory of what health plans usually look like.
- Set an 8th grade reading level. You are explaining something inherently complicated and sensitive. Plain language is the point.
- Never invent numbers, and have it ask you. Telling the model to confirm gaps with you keeps it honest.
This is the same idea as the welcome guide from the onboarding prompts. A few months ago, a page like this meant waiting on a design team or spending hours nudging squares around in PowerPoint. Now a people team can draft it in about thirty seconds. At Warp, I set up our organization’s Claude Design with our design system, which includes our logo and fonts, so everything the people team makes looks like it came from our brand.
Claude Design also lets you queue several requests at once, so you can keep asking for changes while it builds. I asked for a friendlier look with our branding and moved on to the next prompt.
Prompt 4: Make it interactive
This is the prompt I’m most excited about. How many of us have heard “I’m thinking about growing my family, will this plan work for me?” The toggle answers that without a single email to HR.
Make an interactive version. Add a toggle for "Just me", "Me + partner", "Me + children" and "Family" that shows the cost per paycheck for each plan, using the table in the data I pasted. Show the difference between the two plans in plain words. Do not recommend a plan.
All of it comes from a 20-page SBC full of complicated data, modeled into scenarios an employee can click through. Notice the last line again: Do not recommend a plan. The tool informs. The employee decides.
From there, you can share the guide with your broker for a second look to verify the numbers, export it in a few formats, or take it into an app builder like Lovable or Vercel and host it as a live page. Your employees can self-serve during open enrollment.
Additional AI prompts for benefits
The first four prompts cover the enrollment itself. These two are for the bigger decisions that come before it. Both follow the same pattern: real context, a clear ask, and a boundary that keeps you in charge.
Practice your renewal negotiation
Renewals are where the numbers get uncomfortable. Mercer projects an 8.2% rise in health benefit cost per employee for 2027, the biggest increase since 2003. A small company can get a much bigger number from its carrier. This prompt lets you rehearse the call before you are on it.
Play the account rep from [carrier] on a renewal call. I'm an HR lead at a [38]-person company pushing back on a 14% increase. Be realistic: cite medical trend, claims experience, and network changes, and only concede when I ask a well-formed question. After 6 exchanges, stop and coach me. Which questions worked, which did I miss, and what should I request in writing?
The coaching step at the end is the real value. You come away knowing which questions moved the rep and what to ask for in writing.
Compare a traditional plan to an ICHRA
When headcount, states, and budget are all changing, the question is no longer which plan. It is which model. This prompt builds a decision framework and, importantly, does not hand you a verdict.
We have [N] employees across [states], with an age range of [range] and an annual benefits budget of about [$]. Lay out a decision framework for choosing between a traditional small group plan and an ICHRA. Include cost predictability, employee choice, admin burden, multi-state complexity, and what happens as we grow to [future headcount]. Build 3 scenarios with the assumptions stated, and list what data I need from my broker to validate each one. Don't give a verdict; give me the trade-offs.
Asking for the data your broker needs to validate each scenario turns the output into a checklist for your next conversation. For the full picture on the ICHRA side, read our ICHRA guide for startups.
Next up: HR reporting
This series runs every few weeks and is built around what you want to learn. The next session covers AI Prompts for HR Reporting.
My one rule for your upcoming open enrollment season: AI drafts and checks, and you stay in the driver’s seat.
