Oct 28, 2024, 8:00 PM ETInterview, Fireside Chat

How Glean CEO Arvind Jain Solved the Enterprise Search Problem – and What It Means for AI at Work

Core Vision & Market Shift: Glean operates on the belief that 80% of current knowledge worker tasks will be automated by AI assistants within five years.
Product Evolution: Glean has transitioned from an enterprise search company into a general-purpose work assistant and an AI application platform.
Business Performance: The company reports year-over-year revenue quadrupling and utilizes a "search satisfaction" metric, aiming for 80% success rates in sessions where users implicitly validate results by staying on the destination page.
Strategic Approach to AI: Arvind Jain, CEO and former Google Search engineer, argues that building a robust AI application requires solving search fundamentals first (data integration, governance, ranking) before layering on Large Language Models (LLMs).

Data Integration: Glean's first technical layer consists of deep integrations with common enterprise systems (Salesforce, Jira, Confluence, SharePoint, ServiceNow) to aggregate fragmented data.
Governance & Permissions: Unlike public web search, Glean employs a "Governance Engine" to enforce fine-grained access controls, ensuring users only retrieve information they are explicitly authorized to see.
Knowledge Graph: The platform builds a semantic knowledge graph mapping relationships between employees, roles, departments, and documents to contextualize search results beyond simple keyword matching.
Ranking Signals: Glean's ranking algorithm utilizes implicit behavioral signals, including:
- Document popularity at the organizational or team level.
- Recency of engagement (e.g., documents actively viewed in the last 1–2 weeks).
- Interaction history in communication tools (e.g., Slack threads where a document link was shared and "thumped up").
RAG Implementation: Glean utilizes Retrieval-Augmented Generation (RAG) architecture where a retrieval engine fetches relevant context from the knowledge base before feeding it to an LLM, ensuring answers are grounded in private company data rather than public training sets.
AI Native from Day One: The search system incorporated language models as early as 2019 for semantic understanding, predating the current generative AI boom.

Operational Shifts: Customer behavior has shifted from active information retrieval to AI-assisted validation:
- Engineering: Used as a troubleshooting tool to access historical bug fixes and context from previous team members.
- Customer Support: Agents validate AI-generated answers in real-time rather than manually searching for information, accelerating response times.
- Sales: Salespeople use Glean to generate 360-degree customer previews and receive real-time answers during meetings to handle objections without escalating to sales engineers.
- Cross-Functional: Universal adoption occurs in finding subject matter experts based on query context.
Agentic Reasoning Strategy: Glean is moving toward agentic workflows by exposing a workflow engine for users to manually construct automation steps (e.g., "write a weekly status report"). The system learns from these manual workflows to eventually automate complex, multi-step tasks.
Developer Platform: Glean offers APIs that abstract away the "boring 90%" of infrastructure (ETL, data pipelines, governance), allowing developers to focus on business logic and AI application logic.

Competitive Advantage: Glean leverages its existing search infrastructure ("Act One") to secure data access and trust, creating a "moral authority" and easier path to selling AI platforms ("Act Two") without requiring new data onboarding.
Model Strategy: Glean does not plan to train massive foundation models (e.g., GPT-4 scale) but trains smaller, specialized models on individual customer corpora to learn specific lingo, acronyms, and internal context.
Market Sentiment: The founder notes that while the market is crowded with competitors following the ChatGPT release, Glean's unique position is its early investment in the hard infrastructure of enterprise search, which remains a high-barrier-to-entry problem.
Future State (5-Year Vision): The anticipated shift is from reactive query-based AI to proactive assistants that anticipate user needs, manage schedules, and execute tasks autonomously, effectively doubling the productivity of knowledge workers.
Founder Advice: Jain advises founders to ignore competition, focusing instead on solving difficult, specific business problems where the technology provides a distinct advantage, noting that the vast market size means there is ample room for multiple successful solutions.