← Back to Home

Freshworks

AI Co-pilot Metrics Builder

Say what you want to measure. Let AI write the formula.

MRR Projection

$77K+

New Metrics via AI

≥50%

Formula Time Saved

~30%

The Challenge

Business users in Freshworks Analytics needed to define custom metrics (e.g., 'CSAT trend for enterprise accounts in Q4 excluding escalations') but had no SQL or formula skills. Every custom metric required a ticket to the analytics team — a multi-day bottleneck that blocked real-time decisions and created recurring friction for support admins. Three compounding pain points drove abandonment:

  • Complexity of formula syntax
  • No guided metric discovery
  • Frequent syntax errors with no actionable error messages

My Role

Full product ownership, including the business case behind a $77K+ cumulative MRR projection that positioned this as a cornerstone of Freshworks' GenAI-led productivity strategy.

  • Defined the NL-to-formula translation flow
  • Designed the UX, including error messages and metric-reuse patterns
  • Set the formula accuracy benchmarks
  • Led engineering spec through grooming with the prototype flow

The Solution

An AI co-pilot that translates natural language metric definitions into validated database formulas — users describe what they want to measure in plain English, and the system does the rest:

  • Generates the correct formula
  • Surfaces similar existing metrics for reuse, eliminating duplication
  • Explains formula failures in plain language instead of a raw validation error

Full cycle, input to saved metric: under 2 minutes, versus the previous 3-day ticket cycle.

The Outcome

  • $77K+ projected cumulative MRR from active CX account adoption
  • ≥50% of new metrics expected to be created via the AI path post-launch
  • ~30% reduction in formula creation time in early testing
  • Eliminates the SQL bottleneck entirely for non-technical support admins

Positioned as a cornerstone of Freshworks' GenAI-led productivity strategy.

Key Learnings

  • Removing technical barriers directly unlocks revenue — the $77K MRR projection is attributable entirely to unblocking business users who previously couldn't self-serve metric creation.
  • Metric reuse — surfacing existing similar metrics before creating new ones — was the insight that prevented a library of duplicate metrics from emerging post-launch.
  • Plain-language error explanations matter more than the AI's formula accuracy; users need to understand *why* a formula is wrong, not just that it failed validation.

Technologies & Methods

  • OpenAI Codex
  • React
  • Node.js
  • PostgreSQL
  • TypeScript