Back to Tellefsen

    01 — Case

    Urhandleren
    Copenhagen

    In production · 2026

    AI-powered operations for a Nordic luxury watch marketplace.

    Three production systems on one Operating Layer. Forty hours of weekly manual work, removed. Inventory intelligence, pricing analysis, and financial workflows now run autonomously inside the business every day.

    40 hrs

    Weekly manual work removed

    3

    Production AI systems running daily

    93%

    Reduction in listing time per item

    < 8 wks

    Scope to production deployment

    "Tellefsen delivered in weeks what I expected to take months. The systems run every day without intervention — it's like having three extra team members who never sleep."

    Founder · Urhandleren

    Luxury watches arranged on linen surface — Urhandleren marketplace

    02 — Challenge

    Context matters more than code.

    Urhandleren is a fast-growing Nordic luxury watch marketplace dealing in hundreds of premium timepieces. The founder was spending 40+ hours per week on operations that required deep domain knowledge: cataloguing condition grades, researching specifications across dozens of brands, analysing secondary-market pricing, managing invoices, and tracking margins on every transaction.

    The problem wasn't a lack of tools. The problem was that every task required context — knowledge of the watch market, the business model, the pricing dynamics, and the financial rules specific to the marketplace. A new listing didn't just need data entry. It needed someone who understood the difference between a reference 16610 and a 126610LN, what "box and papers" does to resale value, and where to position a piece against live market comparables. That process consumed roughly 45 minutes per watch.

    Financial workflows were handled manually in spreadsheets. As transaction volume grew, so did the bottlenecks — and the risk of error. The business was trapped: scaling meant hiring, but the knowledge required to do the work well was concentrated in one person.

    The real challenge was not automation. It was encoding context into systems that could operate with the same judgement as the founder.

    03 — Approach

    Scope, build, deploy in under eight weeks.

    /01 Scope

    Scope

    Two-week diagnostic mapping every manual workflow across inventory, pricing, and finance. Instead of asking "what can we automate?", we asked "where does your knowledge live — and how do we capture it?" The diagnostic identified three high-impact domains where context-aware AI could replace manual effort entirely, with measurable ROI targets for each.

    /02 Build

    Build

    Rapid development of three integrated AI systems — each one trained on the marketplace's specific business context, not off-the-shelf templates. The architecture centred on Airtable as the operational backbone, with AI layers that understand watch specifications, market dynamics, and financial rules. Every system was designed to run autonomously inside the existing business stack with zero additional headcount.

    /03 Run

    Run

    Deployed to production with monitoring, error handling, and continuous improvement loops. The systems don't just execute — they learn from corrections and adapt to new inventory patterns. The founder reviews outputs, not inputs. All three systems operate daily without manual intervention.

    04 — Systems

    Three production AI systems, running every day.

    /01 System One

    Inventory Intelligence

    Automated product cataloguing that understands watch specifications, condition grading, and market positioning for hundreds of luxury timepieces. The system ingests a reference number and produces complete, market-ready listings — specifications, condition assessment, imagery requirements, and competitive positioning. What used to take 45 minutes per watch now takes under 3 minutes.

    The difference from generic automation: this system carries context. It knows that a Rolex Submariner 116610LN in mint condition with original box and papers commands a specific premium. It understands brand hierarchies, complication categories, and how provenance affects value. That knowledge is embedded in the system — not looked up each time.

    93%

    Time reduction per listing

    /02 System Two

    Pricing Engine

    Real-time market analysis across global secondary luxury watch markets. AI-driven pricing recommendations based on brand, model, reference, condition, and market velocity — continuously calibrated against live transaction data.

    The system doesn't just pull a number. It understands that a vintage Omega Speedmaster Professional has different pricing dynamics than a modern Audemars Piguet Royal Oak. It factors in regional demand, seasonal patterns, and the specific margin targets the marketplace operates with. The founder sets the strategy. The system executes the analysis.

    ±2%

    Pricing accuracy vs. live market

    /03 System Three

    Financial Automation

    End-to-end economic workflows — invoicing, margin calculation, tax compliance, and reporting. Integrated directly with the existing accounting platform (e-conomic), eliminating manual double-entry and spreadsheet reconciliation.

    Every transaction flows from the operational system to the financial system automatically. Margins are calculated in real time. VAT is applied correctly. The founder sees a clean financial picture at any moment — not a spreadsheet that needs updating.

    100%

    Transactions reconciled automatically

    05 — Results

    Measurable impact from day one.

    40 hrs

    Weekly manual work removed

    3

    Production AI systems running daily

    93%

    Reduction in listing time per item

    < 8 wks

    Scope to production deployment

    "Tellefsen delivered in weeks what I expected to take months. The systems run every day without intervention — it's like having three extra team members who never sleep."

    Founder · Urhandleren

    06 — Lesson

    Context is the infrastructure. AI is the engine.

    Most AI projects fail because they treat automation as a technology problem. Plug in an API, write some prompts, ship it. The result is fragile systems that break the moment they encounter edge cases — which in a real business, is constantly.

    The Urhandleren engagement proved the thesis that drives every Tellefsen delivery: context is the concrete machine for profitable AI. The three systems work because they carry the founder's domain knowledge, business rules, and operational judgement inside them. They don't just process data. They understand the business.

    That's the difference between a demo and a production system. And it's why these systems still run — autonomously, every day — months after deployment.

    07 — Next

    Ready to see what AI can do for your operations?

    Book a conversationBack to all work