AI-Powered Support
Meet Sage
Plansight's AI-powered support hub — combining intelligent answers, training videos, knowledge base articles, and interactive walkthroughs in one embeddable widget.
Plansight's AI-powered support hub — combining intelligent answers, training videos, knowledge base articles, and interactive walkthroughs in one embeddable widget.
Sage is Plansight's embedded AI support assistant — a single widget that gives every logged-in user instant access to help, training, and support without leaving the app.
Users ask questions in plain English. Sage searches the knowledge base, finds relevant articles, and generates a clear, conversational answer — personalized with the user's first name.
A fully branded, searchable article browser built from our UserGuiding content. Two-column category tiles, live search, and inline article viewing with embedded videos.
Curated Wistia video channels — organized by topic. Users browse channels, watch videos inline, and get step-by-step visual guidance on any workflow.
Interactive Supademo walkthroughs that guide users through Plansight workflows step-by-step. Hands-on learning directly inside the app.
When a user asks a question, here's what happens behind the scenes.
Via text input or voice (built-in microphone with speech-to-text).
Connects to UserGuiding via MCP protocol. Searches all articles, filters by relevance score (minimum threshold of 25).
Top articles + embedded video transcripts + the user's question are sent to Claude Haiku 4.5. Sage responds in natural paragraphs, using the user's first name, and recommends relevant videos or walkthroughs when appropriate.
Thumbs up/down feedback is captured on every answer. Related articles are shown as clickable links that open inline.
Who asked, what they asked, whether Sage could answer, the feedback, relevance scores — all logged for analytics. Conversations are saved per user, so they can pick up right where they left off next time.
When Sage can't answer, users see a simple "Get help" link — one click, not three. Two clear options:
A persistent "Need help? Contact our team" link is always visible below the input — users are never trapped.
Sage pulls from and pushes to the tools your team already uses.
Powers all AI responses. Trained on your knowledge base content, video library, and walkthrough catalog. Responds as "Sage" with a warm, professional personality.
Direct connection to your knowledge base via MCP SSE. Searches all articles in real-time, fetches full content including embedded videos. No scraping — live data.
Video channels embedded natively. API integration for video catalog. Channels load directly inside the widget with full Wistia player controls.
Interactive walkthroughs embedded via iframe. Step-by-step guided tours for complex workflows, playable directly inside the widget.
Support ticket creation with contact association, urgency levels, and auto-captured page URLs. Chat history logging on contact records.
Serverless PostgreSQL database for conversations, analytics, tenant config, and all persistent storage. Encrypted at rest (AES-256), SOC 2 Type II certified.
Real-time notifications across dedicated channels. Monitor all conversations, get alerted on negative feedback, track resource requests, and see support tickets — all in Slack.
Four dedicated Slack channels keep the right people informed at the right time.
General Monitoring
Every question and answer flows through here. See who's asking what, whether Sage could answer, and what articles were referenced. Your pulse on how users interact with support.
Thumbs Down Alerts
When a user gives a thumbs down, the team sees it immediately — the question, the answer that didn't work, and who was affected. Direct signal for content improvement.
Training Content Pipeline
When users request a new article, video, or walkthrough, it shows up here with their description of what they need and what they were trying to accomplish. A direct content roadmap from your users.
Urgent Issues
Support tickets with urgency level, description, the user's current page URL, and contact info. The support team sees it the moment it's submitted — no delay.
Sage serves multiple roles across the organization.
Deflect routine questions with AI answers. Track what users are asking to identify knowledge gaps. Use resource requests to prioritize new content. Reduce ticket volume while maintaining quality support.
New users get instant access to articles, videos, and guided walkthroughs. Sage recommends the right training resource for whatever they're trying to do. Self-service onboarding that scales.
Analytics show exactly what features users struggle with. Unanswered questions reveal documentation gaps. Feedback data (thumbs up/down) identifies where help content needs improvement.
Tickets arrive with full context — what the user asked Sage, what page they were on, urgency level. Fewer tickets overall, and the ones that come through are higher quality.
Every interaction is tracked and queryable. Built-in endpoints provide:
Simple, maintainable, and ready for enterprise deployment.
Backend: Node.js / Express
AI: Anthropic Claude Haiku 4.5 via API
Knowledge Base: UserGuiding MCP (SSE)
Widget: Vanilla JavaScript — no frameworks, no build step
Database: Neon PostgreSQL (serverless)
Deployment: Vercel (serverless edge)
What your developer adds to the Laravel Blade layout:
<script>
window.PlansightChatConfig = {
apiUrl: 'https://plansight-chatbot.vercel.app',
tenantId: 'plansight',
user: {
name: '{{ auth()->user()->name }}',
email: '{{ auth()->user()->email }}',
userId: '{{ auth()->user()->id }}'
}
};
</script>
<script src="https://plansight-chatbot.vercel.app/chatbot.js"></script>
That's it. Two script tags. User identity is injected from Laravel's auth system.
Every detail is designed to feel native to Plansight.
Implies wisdom and guidance. Ties into the green brand palette. Warm and approachable.
#1D3A44 Dark Green · #B4E0D8 Light Green · #FB513B Accent Orange
Barlow for headings and buttons. Inter for body text. Clean, professional, readable.
Sage speaks like a knowledgeable coworker — warm, professional, and genuinely helpful. Uses the user's first name naturally. Writes in flowing paragraphs, not robotic bullet points. Acknowledges when something is complex and walks users through it.
While processing, Sage cycles through playful status messages: "Digging through the docs...", "Connecting the dots...", "Rummaging through articles..." — with an animated spinning logo. It makes the wait feel alive and intentional.
Built with insurance industry compliance in mind.
A dismissible banner at the top of every chat session warns users: "Please don't share personal health information (PHI), Social Security numbers, or other sensitive data in this chat."
Anthropic's API does not train on customer data. All conversations are logged locally (moving to encrypted Supabase). HTTPS everywhere via Vercel. Plan for BAA with Anthropic for production.
User info comes from Laravel's authenticated session — not stored in the widget. Only logged-in Plansight users can access Sage. Auth0 handles authentication.
Deployed on Vercel (serverless) with Neon PostgreSQL. SOC 2 certified stack. Auto-deploys from GitHub. All conversations, tokens, and costs tracked in real time.
Sage isn't just trained on Plansight documentation — it understands the employee benefits industry.
A custom knowledge base of industry content gives Sage deep understanding of benefits concepts, funding models, compliance requirements, and contribution strategies. When a broker asks a question, Sage can explain both the industry concept and how to handle it in Plansight.
HMO, PPO, EPO, HDHP — plan types, design elements, deductible structures, Rx tiers, network considerations, and rating factors.
DPPO vs. DHMO, annual maximums, frequency limitations, vision allowances, and carrier landscape.
Basic vs. voluntary life, AD&D, STD/LTD, elimination periods, own-occ vs. any-occ, imputed income, and EOI requirements.
Fully insured, community-rated, level-funded, self-funded, experience-rated — when to use each and the tradeoffs.
Percentage-based, defined dollar, buy-up, salary-based, spousal surcharges, opt-out credits, and tier structures.
Tax-advantaged accounts — eligibility, contribution limits, rollover rules, ICHRA, QSEHRA, and strategic pairing with HDHPs.
Accident, critical illness, hospital indemnity, pet insurance — supplemental benefits that fill gaps in major medical coverage.
Employer mandate, affordability thresholds, essential health benefits, metal tiers, reporting requirements, and compensation disclosure.
100+ terms every broker should know — from deductibles and coinsurance to stop-loss, HIOS IDs, and situs state.
The industry knowledge base is maintained as markdown files that can be updated anytime. As Plansight adds features or the regulatory landscape changes, Sage's understanding grows with it. Your team can add new topics, update compliance information, and refine industry content without touching any code.
Sage uses Claude Haiku 4.5 via the Anthropic API — fast, accurate, and extremely affordable.
Claude Haiku 4.5 Pricing
per million tokens (input / output)
Estimated Cost Per Question
half a cent per question
Each time a user asks a question, one API call is made. This sends a system prompt (~500 tokens), the user's question (~50 tokens), 3–5 relevant articles (~2,000–4,000 tokens), and the video/walkthrough catalog (~500 tokens) as input. Claude responds with ~300–500 tokens. At Haiku 4.5 rates, a typical question costs about half a cent. Token usage and costs are tracked in real time in the Admin dashboard.
| Monthly Volume | Input Tokens | Output Tokens | Estimated Cost |
|---|---|---|---|
| 500 questions/mo Small team, early adoption |
~2.5M tokens | ~250K tokens | ~$3/mo |
| 2,000 questions/mo Growing user base |
~10M tokens | ~1M tokens | ~$12/mo |
| 5,000 questions/mo Full rollout |
~25M tokens | ~2.5M tokens | ~$30/mo |
| 10,000 questions/mo High-volume enterprise |
~50M tokens | ~5M tokens | ~$60/mo |
With prompt caching enabled, real-world costs could be 50–70% lower than the estimates above.
Only the Ask Sage AI chat tab incurs API costs. All other tabs are free.
Current status and planned enhancements.
Plansight's support experience has two complementary layers. They work together — not as replacements for each other.
Contextual help that follows you
Hubs are UserGuiding's native sliding panel — they automatically update content based on which page you're on. On the Dashboard, you see Dashboard help. On an RFP, you see RFP resources.
AI-powered help across everything
Sage is a custom-built AI assistant that lives on every page, all the time. Users ask questions in plain English and get instant answers drawn from the entire knowledge base — not just the current page.
A user working on an RFP might open the Hub to follow the step-by-step RFP checklist. But when they hit an unexpected question — "Can I change the situs state after sending?" — they ask Sage, who searches the entire knowledge base and gives an instant answer.
Hubs handle the structured, contextual learning path. Sage handles the unstructured, on-demand questions. Together, users never get stuck.
Sage gives your brokers the support they need, the moment they need it — so they can focus on relationships, not searching for answers.