AI for Creating Metrics

Portátil mostrando el panel de métricas de Graphext.

The Problem

Business clients often struggled to generate meaningful metrics because they didn’t always know what to look for in their data. This lack of direction made it difficult to define and create the metrics that would unlock valuable insights, leading to frustration and limiting the value they could extract from the platform.

Diagrama del recorrido del cliente al intentar crear métricas.

My role

As Product Designer, I worked with development and business teams to create an AI-powered, no-code interaction that guides users — especially those unsure where to start — by suggesting relevant metrics based on their data.

Design process

  1. Briefing analysis

    User research revealed that the biggest challenge was users’ uncertainty about which metrics to create or how to start their analysis.

  2. Ideation

    Based on the feedback, I brainstormed several flows that would allow an AI assistant to guide users from a point of uncertainty to concrete, valuable insights.

  3. Prototype & Validation

    I made metric creation easily accessible by placing the entry point prominently on the dataset’s Home page.

    In designing the AI-assisted flow, I ensured users could still create manual metrics with full functionality.

Solution

The final design introduced an AI-powered assistant integrated into the metric creation flow. Users could describe the metric they were interested in (e.g., “What are the average sales by region?”), and the AI would:

  1. Open metrics panel,
  2. Parse a request,
  3. Suggest an appropriate metric formula + display a visual preview of the result,

*Users can refine or adjust their request using natural language.

Access to metrics

  • Sidebar access for better visibility
Panel lateral de métricas dentro de Graphext.
Vista del conjunto de datos con el acceso a métricas destacado.

Create a new metric with NL

Define calculation with prompt:

Which metric best shows how well my predictive models match real customer behavior?

Asistente de IA proponiendo una fórmula de métrica a partir de una pregunta.

Validate calculation

Pantalla de validación de la métrica generada, con su fórmula y vista previa.

Key design decisions included:

  • Conversational UIto create a more approachable and less intimidating experience.
  • Inline explanationsto clearly show how the AI generated the metric.
  • The option for advanced users to manually adjust the formula if desired.

Impact

User feedback showed increased confidence in data exploration, thanks to AI guidance when users were unsure where to start. Fewer support tickets about metric creation indicated the feature effectively solved a major pain point, helping Graphext empower non-technical users to extract valuable insights independently.