Built-in AI: Onboarding reduced from 45 minutes to 2 minutes

We redesigned the customer dashboard of an American cyber risk management platform, placing an embedded AI agent at its center. Onboarding was reduced from a 45-minute assisted process to under 2 minutes in self-service mode, and a tool previously run once a quarter became a system for daily decision support. Board members, the CFO, and the CISO use the same interface — each in their own way.

Challenge

The platform was a mature cyber risk analytics solution, with a Fortune 500 customer base. As the product evolved, however, three barriers emerged:

• High barrier to data understanding: interpreting metrics required specialized expertise — a common pattern in tightly regulated, analytics-driven sectors such as banking, finance, and compliance.
• Low daily retention: most users logged in only once a quarter to generate periodic reports — the platform wasn’t part of their daily work.
• High implementation cost: new customers needed about 45 minutes of setup with the customer support team before they could use the system independently.

When agentic, conversational AI became a market standard, the client tasked us with bringing this interaction model into the user dashboard. The biggest challenge was that one interface had to serve vastly different audience groups:

• Board member needs a concise, self-explanatory view that speaks for itself without delving into details.
• CFO looks for business context and the story behind the numbers.
• CISO requires full conversational depth to precisely diagnose vulnerabilities and act immediately.

Our approach:

The key product challenge was how to integrate LLM models into a classic dashboard without forcing users to switch between two different modes of work. We decided against building a separate chat alongside the platform to avoid creating two parallel, competing interfaces. We opted for a hybrid model: the familiar point-and-click interface for everyday tasks combined with the ability to ask a natural-language question when automation is needed. This approach preserves existing user habits—letting them gradually get comfortable with AI assistance while providing guardrails at every step.

At the heart of the portal is a four-level model of interaction with the AI agent (Aria):

Level 0 – Management: What's happening? 

The first view after logging in. A classic dashboard with key cybersecurity metrics that doesn't require direct interaction with AI. Buttons beneath the indicators instantly display pre-generated data summaries. It's a concise, self-sufficient view designed for the executive team.

Level 1 – Finance: Why is this happening? 

An analytical layer dedicated to the CFO and finance leaders. Every metric can be drilled down using ready-made analyses or an interactive AI chat. Contextual query suggestions (prompt chips) trigger complex prompting under the hood, delivering expert-level insights without requiring users to craft prompts themselves.

Level 2 – Execution: Do it for me. 

An operational space for the CISO and SecOps teams. Action buttons turn into concrete commands (Create, Identify, Prepare). Instead of offering instructions and advising what the user should do, the agent autonomously plans a multi-step workflow and executes it using integrated tools.
 

Level 3 – Any query.

An open text field (free-form prompt), available at any time for users with a clearly defined goal.

This layout is intentionally non-linear. In practice, users often loop their actions (from exploration, through in-depth analysis, to a concrete action), adapting the tool to their own decision-making process.

Solution

Aria is a fully native AI agent built into the fabric of the platform. It operates directly within the interface, branding, and authentication system, combining the analytical depth of source data with a conversational layer. Users don't need to rely on external tools. A system that previously required analyst assistance has become fully self-service—allowing users to ask questions, validate hypotheses, and implement actions in one place. Identified risks are no longer merely passively reported; they are actively mitigated directly from the dashboard.