Key results
faster time-to-insight for internal teams by eliminating the need to write queries or search dashboards.
hours/month saved for the analytics team by offloading repeat ad hoc requests
increase in self-serve adoption across sales and finance users.
business questions/week now answered directly by the AI assistant - including revenue gaps, top customers, margin drops, and inventory status.
About the Client
A U.S.-based D2C nutraceutical company offering a diverse portfolio of premium vitamins, supplements, and personal care products. With a strong focus on clean ingredients and holistic wellness, the company reaches health-conscious consumers through its owned digital channels, retail partnerships, and in-house manufacturing supported by science-led product development.
Key Challenges
BI bottlenecks slowed down decision-making—business users had to wait for analysts to respond to routine questions.
Low dashboard adoption—users found dashboards overwhelming or unintuitive for their day-to-day questions.
Analysts were overloaded with repetitive, low-complexity data requests that delayed deeper work.
Cross-functional misalignment on KPIs due to inconsistent reporting across teams and regions.
Our Approach
Identify Repetitive, High-Value Questions
- Conducted workshops with sales, finance, and inventory users to map their most frequent data queries.
- Prioritized use cases like:
“Which campaigns underperformed?”
“Show top 10 SKUs by margin drop.”
“Compare forecast vs actual revenue this quarter.”
Connect Secure Data Sources
- Integrated Marketplaces, Marketing data sources, Shopify and Order Management Platform into GCP BigQuery using DataChannel.
- Defined semantic layers (KPIs, time filters, data sources).
- Created guardrails to limit access to sensitive fields and apply output validation.
Build the Conversational Analyst
- Leveraged Azure OpenAI (GPT) via LangChain to interpret natural language and generate SQL/dash calls.
- Fine-tuned prompts to align with business vocabulary, synonyms (e.g., “sales rep” = “seller”), and filter logic.
- Enabled context retention for follow-ups and clarification questions.
Pilot, Test, and Expand
- Launched a pilot with 15 users across sales ops, finance, and regional leadership.
- Collected query patterns, top intents, and feedback for tuning and training.
- Expanded access across 5 business units and embedded the assistant in Slack + internal portal.
Implemented Solution
Conversational Interface for Internal Teams
- Simple web and Slack interface where users ask business questions in plain English.
- Instant answers delivered with contextual charts, KPIs, or downloadable reports.
- Example queries:
“Which product lines are underperforming this month?”
“Show me margin trend for Region East, Q1.”
Live Query + Narrative Response
- Auto-generates SQL and runs it securely on GCP BigQuery.
- Returns not just data, but business context:
“Revenue dropped 5% YoY due to decline in Product Line B in Region South.”
Context-Aware Follow-Ups
- Users can ask follow-ups like:
“What about Region West?”
“Drill into customer-level view.” - Assistant retains memory within the session for fluid conversations.
Compliance & Feedback Loop
- All user prompts, queries, and responses are logged and auditable.
- Users can flag incorrect responses or request deeper insights.
- Admin dashboard shows usage trends, most asked questions, and improvement areas.
Technologies Used
