← all projects

AI Sales Intelligence

active

An AI account-research pipeline that turns hours of manual research per account into a ranked assessment, built for KUKA's global sales organization.

Jul 2026 – Present · @ Device Insight GmbH (KUKA Group) · closed source

  • Kotlin
  • Spring Boot
  • Anthropic API
  • OpenTelemetry
  • SQLite

Problem

"Which customer actually has potential? Nobody really knows." KUKA's sales organization works across about 15,000 accounts, with a CRM that is incomplete and often stale. Sales knows too little about most of them to decide which to call next, about what, and why, and researching a single account by hand takes hours (the sales side's estimate).

What I built

I designed the solution in pre-sales, worked out the sales logic with sales in workshops, and built it as technical lead.

  • Any source becomes facts. CRM history, web research and company data merge into one fact store per account.
  • An ontology encodes the sales logic. Signals, chances and actions, written with sales and versioned as code, give the models a menu to reason with instead of a blank page. Code makes the final pick.
  • Provenance all the way down. Every claim links to its source, down to the web page quoted, and every AI step is recorded, so users and engineers know exactly what the AI did across the whole system. A claim without a source is refused.
  • Observability first. OpenTelemetry traces every model call with its cost, tokens and refusals, so the answers exist before anyone asks.
  • Built for scale. Designed to assess all 15,000 accounts on a schedule.

What the PoC proved

  • One assessment in about 7 minutes, at about $1 per account, across 78 accounts profiled end to end.
  • A golden set scored by sales reps checks every change.
  • Delivered in about two months and presented to KUKA's board.

There are no production users yet, so the business impact is still ahead.

What's next

AI Sales Intelligence became the first use case of the Global KUKA AI Use Case Team, which I lead. The production deployment on Azure is being built.