Talk to
Your Data
Product managers at CloudBees were drowning in fragmented data. I partnered with Data Science to give them a natural language interface to everything, so the answer to any product question was a sentence away.
Impact
75%
Less time on data synthesis
PMs reclaimed hours previously spent manually pulling and reconciling data across tools
8
Data sources unified
Support tickets, telemetry, CRM, feedback, and more consolidated into one semantic layer
92%
Agent response efficacy
Validated accuracy rate across query types, sufficient for trusted decision-making
4hrs
Saved per PM per week
Average time reclaimed through natural language queries vs. manual reporting workflows
The Problem
Product managers on Unify were making decisions with incomplete information. Not because the data did not exist, but because it was spread across too many places. Support tickets lived in one tool. Product telemetry in another. Customer feedback somewhere else. CRM data in a completely separate system.
Getting a complete picture of what customers were experiencing required pulling exports, building spreadsheets, and spending hours reconciling data that should have been answering questions, not creating more work. By the time the analysis was done, it was already stale.
The result was product decisions that relied more on instinct and recency bias than on the actual signal sitting in the data.
The Approach
The real problem was not the data. It was the interface. The data existed. It just lived in a shape that required engineering skills to access and hours to interpret. My goal was to change that interface entirely.
I partnered with the Data Science team to build semantic views in Snowflake, mapping all the relevant data sources into a unified layer organized around product and business concepts rather than raw table schemas. That foundation made the next step possible.
We exposed that layer through Snowflake Intelligence, giving product managers a natural language query interface directly over their data. Instead of filing a data request or building a dashboard, a PM could ask: what are the top customer requests this past quarter from customers with over $1M ARR, and get a structured, sourced answer in seconds.
The system was validated for efficacy before rollout. Responses had to be accurate and consistent enough to trust, not just impressive demos. We got there.
What Was Built
01
Semantic Views in Snowflake
Worked with the Data Science team to design semantic views that mapped disparate data sources into a unified, queryable layer. The schema was built around how PMs actually think about their products, not how data engineers think about tables.
02
Natural Language Interface
Leveraged Snowflake Intelligence to expose the semantic layer through a conversational interface. PMs could ask business questions in plain English and receive structured, sourced answers without writing a single query.
03
Cross-Source Synthesis
A single query could now pull from support tickets, product telemetry, customer feedback, and CRM data simultaneously. Answers were synthesized across all sources, removing the need to manually stitch together exports from five different tools.
04
Segmented Customer Insights
PMs could filter by business-relevant dimensions like ARR tier, customer segment, or product line. Questions like which features are top-requested by customers over $1M ARR went from a multi-day research effort to a seconds-long query.
The Outcome
Product managers on Unify went from spending hours on data assembly to getting answers in seconds. The quality of insight improved because the data was now synthesized across all sources, not cherry-picked from whichever tool was easiest to access.
More than a time savings, it changed the nature of the questions PMs were asking. When the friction of getting data disappears, the questions get sharper. That shift from reactive reporting to proactive querying is where the real product value showed up.