Selected case study / Product & experimentation

A simpler view
of what works.

A local-first analytics and optimization platform that turns A/B test data into clear next actions, with AI assistance running entirely inside the client environment.

ClientPrivate & Confidential
RoleProduct strategy & digital leadership
StackNode.js · Ollama · Local AI
● ● ●optimization / overviewlocal instance
kg / labOverviewExperimentsAudiencesInsightsSettings
Experiment performanceConversion overview
Last 30 days⌄
Conversion rate+18.4%
May 01May 15May 30
Visitors84,291+12.8%
Winning variantVariant B92% confidence
AI recommendationReady3 actions found

01 / The challenge

Analytics had
become noise.

The team had access to plenty of data, but not enough clarity. Standard analytics tools surfaced reports; they did not help the team decide what to test next, why a result mattered, or how to act on it.

The opportunity was to create a focused alternative: a simpler operating view for experimentation, with an AI layer that could explain patterns and suggest practical optimization paths without sending sensitive data to a third-party model.

02 / The approach

From data
to decisions.

Every screen was designed around the question that follows a metric: “What should we do now?”

01

Compress the view

Remove dashboard sprawl and make experiment health, confidence, and movement legible at a glance.

02

Make tests useful

Bring audience context, variant performance, and qualitative observations into one decision surface.

03

Keep intelligence local

Use a custom Ollama model running locally to interpret results without exporting proprietary data.

03 / The architecture

Private by
design.

Node.js provides the durable application layer. Ollama makes the intelligence layer portable, inspectable, and local.

01Event dataVisits · variants
conversions · context
02Node.js appIngestion · APIs
experiments · auth
03Ollama modelLocal inference
pattern analysis
04Action layerRecommendations
test backlog · learnings
LOCAL-FIRST PRINCIPLE

Client data remains inside the client-controlled environment. The model supports judgment; it does not replace it.

04 / The outcome

A sharper loop
for growth.

The product creates a shared language between marketing, product, and engineering: one place to understand what happened, what it means, and what deserves attention next.

Discuss the work
01Focused operating view
02AI-assisted test analysis
0Third-party model calls