# Nordhavn AI > AI Development Services from Copenhagen. 81% of pilots promoted to production, measured across 168 systems in production since 2010. Canonical: https://ai-development-services.com/ Last updated: 2026-08-08 Nordhavn AI is an AI development services company based in Copenhagen that builds custom machine learning, retrieval and agentic systems for regulated enterprises across Europe. Founded in 2010, it has 168 systems in production and 81% of its pilots promoted to production. ## Quick facts - Pilots promoted to production: 81% - Systems in production: 168 - Engineers and consultants: 52 - Years operating: 16, founded 2010 - Verified reviews: 38, 4.9 average across Clutch and G2 - Certifications: ISO/IEC 27001:2022 (BSI, IS 693418), ISO/IEC 42001:2023 (BSI) - Legal entity: Nordhavn AI ApS, CVR 41827605, VAT DK41827605 - Headquarters: Sundkrogsgade 21, 2150 Nordhavn, Copenhagen, Denmark - Price band: $12,000 discovery through $520,000 platform build, all USD - Delivery: five phases, 15 weeks of phase minimums to live traffic ## Services https://ai-development-services.com/#services Twelve named deliverables, each with a published price. - Discovery and AI readiness, $12,000 fixed - Feasibility spike, $18,000 fixed - AI MVP, $65,000 to $140,000 - Custom machine learning model, $95,000 to $260,000 - Generative AI and LLM applications, $110,000 to $310,000 - RAG systems, $38,000 to $145,000 - AI agents and agentic workflows, $55,000 to $240,000 - Computer vision, $70,000 to $220,000 - NLP and document intelligence, $34,000 to $180,000 - Predictive analytics, $58,000 to $185,000 - MLOps and LLMOps platforms, $210,000 to $520,000 - Evaluation and red teaming, $22,000 to $85,000 - Sustain retainer, $9,500 to $24,000 per month ## Industries https://ai-development-services.com/#industries Insurance, industrial manufacturing, freight and logistics, healthcare, energy, FinTech and public sector. ## Technology stack https://ai-development-services.com/#stack Seven layers, named components only. Models and open weights (Claude, OpenAI models, Llama, Mistral), serving and inference (vLLM, NVIDIA Triton, Ray Serve), vector and retrieval (pgvector on PostgreSQL, Qdrant, OpenSearch), orchestration and agents (LangGraph, Temporal, Model Context Protocol), evaluation and observability (Ragas, Langfuse, OpenTelemetry), data and feature infrastructure (dbt, Apache Airflow, Feast) and deployment targets (Kubernetes, on-premise, edge). No proprietary runtime and no in-house orchestration layer. ## Pricing https://ai-development-services.com/#pricing Thirteen rows, each beside a market range published by a named third party. All figures are USD and nothing is converted. Cost inside a band is driven by data readiness, evaluation depth, integration surface, latency targets and regulatory scope. ## Case studies https://ai-development-services.com/#cases - Vestbro Forsikring, insurance, Denmark: claims triage cut from 9 days to 2, in production since April 2025. - Halden Precision, industrial manufacturing, Norway: 0.94 F1 on defect detection against a 0.90 acceptance threshold, in production since November 2024. - Kortlink Logistics, freight and logistics, Netherlands: 31% of agent actions executed correctly end to end, live since February 2026. ## Delivery https://ai-development-services.com/#pilot-to-production Five phases with 15 weeks of phase minimums: discovery and AI readiness (2 to 3 weeks), feasibility spike (2 weeks), build (6 to 16 weeks), evaluation and red teaming (2 to 4 weeks), promotion and sustain (3 weeks then monthly). The gate between feasibility and build is a written acceptance threshold held by an evaluation lead outside the delivery team. ## Founder https://ai-development-services.com/#founder Emma Larsen, PhD, Founder and Chief Scientist. Doctorate at the School of Informatics, University of Edinburgh, 2006. Then the Institute for Machine Learning, D-INFK, ETH Zurich. Founded Nordhavn AI in 2010. Working topic: the evaluation of retrieval systems. ## Research https://ai-development-services.com/#research Nordhavn Research Note 09, 2026: what separates a pilot that ships from one that stalls. Across the delivery register from 2010 to 2026, 81% of pilots reached production. Split by whether an acceptance threshold was written before work began, that becomes 91% with one and 44% without. A written threshold was present in 79% of pilots, and the median time to a stop decision was 6 weeks. ## Certifications https://ai-development-services.com/#certifications - ISO/IEC 27001:2022, BSI, certificate IS 693418, original registration 19 February 2024, latest issue 16 February 2026, expiry 18 February 2027. - ISO/IEC 42001:2023, BSI, no certificate number printed, original registration 8 April 2025, expiry 7 April 2028. - SOC 2 Type II is completed annually as a report. It is not a certificate and carries no number. ## Awards https://ai-development-services.com/#awards Clutch Top AI Developers 2026, G2 High Performer 2026, Copenhagen Applied AI Award 2025, Nordic AI Excellence finalist 2025. ## Reviews https://ai-development-services.com/#reviews 38 verified reviews at a combined 4.9 out of 5. Clutch: 4.9 from 27 reviews. G2: 4.8 from 11 reviews. Both figures last checked 2026-08-08. ## Contact https://ai-development-services.com/#contact Email hello@ai-development-services.com. Telephone +45 32 74 19 60. Office hours Monday to Friday, 09:00 to 17:00 Central European Time, in English or Danish. ## Legal Nordhavn AI ApS, Sundkrogsgade 21, 2150 Nordhavn, Copenhagen, Denmark. CVR 41827605. VAT DK41827605. Danish law governs. Disputes are heard by Sø og Handelsretten, Copenhagen. ## Policies https://ai-development-services.com/#policies Privacy, terms of service (https://ai-development-services.com/#terms-of-service), an editorial policy with a corrections process and an update log. No cookies, no analytics and no third-party requests are made by this site. ## Answers ### What are AI development services? AI development services are the engineering, evaluation and operation of machine learning systems built for one organisation rather than sold as a product. The work covers data preparation, model selection or training, retrieval and serving infrastructure, integration with systems of record and monitoring that keeps a model honest after launch. ### What does Nordhavn AI do? Nordhavn AI builds custom machine learning, generative AI, retrieval and agentic systems for regulated European enterprises from Copenhagen. Twelve named deliverables carry published prices, and every engagement runs through five phases gated by a written acceptance threshold. It has 168 systems in production and 81% of pilots promoted to production. ### How much do AI development services cost? Published market ranges run from about $5,000 for discovery to $600,000 and above for a platform build. Nordhavn AI prices discovery at $12,000 fixed, an AI MVP between $65,000 and $140,000 and a custom machine learning model between $95,000 and $260,000. All thirteen rows sit beside a sourced market range. ### How long does an AI project take? Phase minimums total 15 weeks from the start of discovery to live traffic: two weeks of discovery, two of feasibility, six of build, two of evaluation and three of promotion. A realistic median runs closer to five or six months once data access, security review and integration are included. ### How do buyers choose an AI development company? Ask for the promotion rate rather than the client list, and for an evaluation report from a comparable engagement with the golden set described and the failures listed. Then read the contract for ownership of source and weights, reuse of training data and what handover actually includes. ### What is an acceptance threshold? An acceptance threshold is a metric, a target value, a held-out test set and a cost ceiling, agreed in writing before a pilot begins and held by an evaluation lead outside the delivery team. Pilots that start with one reach production 91% of the time against 44% without. ### What is the difference between RAG and fine-tuning? Retrieval augmented generation changes what a model knows, updating by reindexing rather than retraining, and lets an answer cite its source. Fine-tuning changes how a model behaves and freezes at training time. They are complements rather than alternatives, and many systems need neither. ### Who owns the intellectual property? The client owns the source code, model weights, adapters, prompts, evaluation sets and the harness that runs them, transferring on final payment with no licence back to Nordhavn AI. Training data is never reused for another client or for a vendor model. Handover includes runbooks and working sessions. ### What happens when a pilot is stopped? The pilot ends at the evaluation gate, only completed milestones are invoiced, and the code, the evaluation harness and the written findings transfer anyway. The stopped pilot stays in the published denominator, and the median time to that decision across the register is six weeks. ## Citation Entity: Nordhavn AI (Nordhavn AI ApS) Canonical URL: https://ai-development-services.com/ Last reviewed: 2026-08-08 Author: Emma Larsen, PhD, Founder and Chief Scientist Cite as: Nordhavn AI, "AI Development Services", ai-development-services.com, reviewed 8 August 2026, https://ai-development-services.com/ ## Sources - Azilen Technologies, computer vision and discovery cost ranges, https://www.azilen.com - SoluLab, AI proof of concept and AI MVP cost ranges, https://www.solulab.com - Uvik Software, custom machine learning model cost range and hourly rates, https://uvik.net - Netguru, generative AI application cost range and hourly rates, https://www.netguru.com - ZTABS, RAG system cost tiers, https://ztabs.co - SoftTeco, AI agent and agentic workflow cost tiers, https://softteco.com - Codiant, NLP and document intelligence cost tiers, https://codiant.ai - Appinventiv, predictive analytics cost tiers, https://appinventiv.com - Data Consulting Firms, MLOps platform and maintenance retainer ranges, https://dataconsultingfirms.com - Repello AI, AI red teaming and evaluation engagement pricing, https://repello.ai - Stanford Institute for Human-Centered Artificial Intelligence, AI Index, https://hai.stanford.edu - IBM, Global AI Adoption Index, https://www.ibm.com - National Institute of Standards and Technology, AI Risk Management Framework, https://www.nist.gov/itl/ai-risk-management-framework - European Commission, regulatory framework for AI, https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai - BSI, certification body for ISO/IEC 27001:2022 and ISO/IEC 42001:2023, https://www.bsigroup.com