Overview
The Burning Glass Institute — a research organization focused on workforce analytics and skills development — needed a tool to help enterprise clients understand how AI could transform their operations. The ask: design an experience that would take a large organization with no clear AI strategy and produce a tailored, actionable AI opportunity landscape in a single session.
The users were senior leaders at companies prepared to invest significantly in AI adoption. They were experts in their business — not in AI. That distinction became the central design problem.
The Problem
Enterprise organizations know they need to act on AI, but most don't know where to start. Without a structured framework, the question "where should we use AI?" is nearly impossible to answer from scratch.
Any tool designed to help would need to bridge a significant knowledge gap — and do it quickly enough to be useful in a consulting or sales context.
Iteration 1: The Chat Interface
The first design explored a conversational approach: an LLM-powered chat that asked open-ended questions to surface business challenges, then generated a ranked list of AI use cases in response.
The logic was sound — conversation feels natural, and an LLM could theoretically adapt to whatever the user shared. In practice, the experience put too much burden on the user. A blank prompt asking "Do you have a specific area in your business you'd like to leverage AI?" gave enterprise leaders no scaffolding to work from. Users who weren't AI-fluent didn't know what to say — and the quality of the output depended entirely on what they typed.
The client's feedback was direct: it felt too complicated. For a high-stakes audience used to structured decision-making, an open-ended chat created friction rather than clarity.
The Pivot: Structuring the Input to Unlock the Output
The core insight from Iteration 1 was that the problem wasn't the LLM — it was the interface asking users to do AI strategy cold. The redesign flipped the model: instead of asking users to articulate their needs in freeform text, the tool would collect structured data first, then let the AI do the interpretation.
This meant rethinking the entry point entirely.
The Structured Flow
Iteration 2: Structured Onboarding + Strategy Formulation
The second iteration introduced a three-step onboarding flow before any AI strategy work begins.
Step 1 — Company Profile: Users enter their company name and URL. Simple, fast, and signals to the system what industry and context it's working with.
Step 2 — Select Peers: Users identify up to five peer organizations for benchmarking. This grounds the output in competitive context rather than generic recommendations.
Step 3 — Establish Workforce Profile: Users select their active business areas, then confirm or edit their current workforce composition — headcount and average salary by role, across functions like Customer Service, Risk, Compliance, IT, and Operations. They can also upload HRIS data to populate this automatically.
By the time users reach the strategy screen, the system already has enough structured context to generate meaningful, company-specific recommendations.
Formulating the Strategy
The strategy screen gives users two paths: "Describe My Idea" (if they already have an AI use case in mind) or "Help Me Choose" (if they want the system to surface opportunities based on their stated priorities). A third entry point — Explore Peer Benchmarking — lets them see what comparable organizations are doing.
The "Help Me Choose" path asks one open-ended question — the company's top strategic priorities over the next 2–3 years — and responds with a set of suggested company strategies, each with specific AI applications attached. Users select a strategy, optionally refine it, and define their success metrics (revenue growth, cost to service, cycle time) before proceeding.
This structure transformed the chat's open-ended burden into a guided decision that even AI-unfamiliar leaders could complete confidently.
The Outcome, and What I Learned
The Output: An Actionable AI Landscape
Once onboarding and strategy formulation are complete, the tool generates a full AI opportunity landscape organized around three dimensions:
Workforce Impact shows how AI adoption changes the workforce composition — which roles grow, which decline, and by how much — with an adjustable adoption assumption slider (Conservative to Aggressive) and estimated cost savings and headcount optimization figures.
Workflow Changes maps AI capabilities to specific work activities and tasks, showing whether each task is augmented or automated, and which roles are affected. Users can drill into any role for task-level detail.
AI Vendors & Tools recommends specific technologies and vendors tied to each use case, with a visual network map of skill and technology relationships and an AI Technology Vendor Plan for investment prioritization.
The experience continues into a dashboard that gives organizations an ongoing view of their AI portfolio: an Opportunity Mix scatter plot (value at stake vs. labor intensity), Talent & Skills impact, internal mobility opportunities via Sankey diagram, and a Progress & Throughput kanban tracking use cases from Ideas through Scaling.
Design Decisions Worth Noting
Dual entry paths in strategy formulation — "Describe My Idea" vs. "Help Me Choose" — respected the range of sophistication in the user base. Some clients arrived with a specific use case; others needed the system to lead. Both paths converge on the same structured output.
Workforce data collection up front made the workforce impact output genuinely personalized rather than generic. The "Upload HRIS Data" option lowered the effort barrier for larger organizations with existing HR systems.
The adoption assumption slider on the workforce impact screen was a deliberate choice to make AI projections feel honest rather than prescriptive. Leaders could explore conservative vs. aggressive scenarios without committing to a single forecast.
Reflection
The most significant design decision on this project wasn't a UI choice — it was recognizing that the interface had to compensate for what users couldn't reasonably be expected to know. An open-ended chat works when users have clear intent. For enterprise AI strategy, intent needs to be built, not assumed.
Restructuring the input — through onboarding, peer selection, workforce profiling, and structured strategy paths — gave the LLM what it needed to generate high-value output, while giving users a path they could actually follow.
