What 55,000 employee questions reveal about benefits decision support

For years, digital benefits experiences have focused on access — making information available and expecting employees to find what they need. But as benefits grow more complex, access alone doesn't solve the problem.

Alight's early pilots of the AI-powered Alight Assistant — a feature integrated across the Alight Worklife^® platform — point to a different model. During the 2025 annual enrollment period, 8.5 million people engaged with AI chatbots across Alight Worklife, and users asked nearly 55,000 questions through the Alight Assistant alone. The volume confirmed something we'd suspected but hadn't seen at this scale: employees want guidance that fits their situation, delivered at the moment they're making a decision.

Here are five things those behaviors taught us.

1. Relevance builds trust — and repeat engagement proves it
   
   Employees didn't just try the Alight Assistant — they came back. One client saw an 187% year-over-year increase in AI assistant usage, a pattern that held across the pilot group. When answers deliver clear value, employees start to treat the tool as a first resource rather than a last resort.
   
   That value comes from responses tailored to the employee's situation, written in simpler terms, and delivered at the right moment. Employees could build on prior questions without starting over — a conversational flow that mirrors how people think through benefits decisions.
   
   The questions themselves followed predictable patterns:

• • Comparing plan options
  • Understanding Health Savings Accounts (HSA) and Flexible Spending Accounts (FSA)
  • Determining eligibility
  • Verifying provider coverage

These same topics surface long after enrollment closes — during life events, provider changes, and mid-year coverage questions. That's why the Alight Assistant is designed as a persistent feature within Alight Worklife®, available whenever employees need help with health benefits, savings accounts, eligibility, or coverage.

2. Context turns information into action
   
   Employees care about one thing: what applies to them. Generic answers miss that mark and often add confusion.
   
   The best AI responses pulled real-time context from Alight Worklife to tailor each answer to the person asking. That context included:

• • The employee's current coverage and what they're eligible for
  • The exact page or task they were working on at that moment
  • Plan documents, benefits guides, and plan design details

When answers matched the employee's unique situation, self-service worked. At one pilot client, web-to-call escalations dropped from 39% to just 6%. Employees found what they needed and skipped the phone call.

3. AI-generated answers outperform static responses
   
   Across pilots, AI-generated answers consistently earned higher satisfaction ratings than traditional static responses. 86% of employees gave the experience a thumbs-up, and 89% of those positive ratings came from AI-generated answers rather than pre-written content.
   
   Negative feedback, by contrast, clustered around the limitations of static answers — situations where the response couldn't adapt to what employees actually asked. As noted above, employees care most about what applies to them. That expectation is changing how they view benefits support: they want answers that fit their situation, not generic responses pulled from a document library.

4. Timing shapes the experience
   
   When support appears matters as much as what it says.
   
   The pilot cases showed that well-timed interactions — aligned to key decision points — increase engagement. At the same time, prompts that feel intrusive can disrupt the experience. The strongest results came from a model that stays present when needed and steps back when not.

5. Content quality determines AI quality
   
   AI performance depends on what's underneath it. A response from the Alight Assistant is only useful if it's correct — and correctness at scale requires structured, employer-specific plan content as the foundation.
   
   The Alight Assistant includes clearly defined guardrails to maintain accuracy and reliability, with fallback paths that preserve continuity when the AI's confidence in a response is low. This content-first architecture also benefits the people supporting employees directly: employees using AI-powered Alight Assistant achieved a 40-second reduction in average handle time compared to those without it.

What the data points to next

The enrollment pilots produced results that extend well beyond the enrollment window itself. Across the pilot group:

 
These numbers describe more than an enrollment success. They describe a shift in how employees expect to interact with their benefits — through direct, plan-aware answers delivered in the moment, across every stage of the benefits journey.

The Alight Assistant was built for exactly this: an AI-powered layer within Alight Worklife^® that meets employees wherever they are — during enrollment, through a mid-year life event, or on a Tuesday afternoon when they need to check whether a provider is in-network. 

What started with 55,000 questions is now shaping how AI-driven support extends across health, wealth, and wellbeing — reducing complexity and delivering answers that apply to the person asking.