95% cheaper and sovereign AI: Why we dropped Copilot at Coop Bank — LinkedIn post by Jannik William Frederiksen
Jannik William Frederiksen · article · 2026-06-06 Reader 🖥️
Coop Bank rejected Microsoft Copilot because EU data sovereignty (bank-regulated data possibly egressing the EU under peak load) and cost (200,000 DKK/year for 80 users) failed the bar for a bank; instead they wired Scaleway-hosted open-weight models (Gemma/Mistral) behind a one-day custom chatbot UI, achieving full EU data residency and a >95% annual cost reduction (under 10,000 DKK) — at the price of not being a 1-1 Copilot replacement, but with hyper-flexibility that lets business ideas reach production in moments; the implicit argument is that for regulated EU institutions the default Microsoft path is both more expensive and less sovereign than a hosted-OSS path, and that vendor-default thinking is leaving competitive lead on the table.
practical · expository-with-argumentative-coda — Short Danish-language LinkedIn case-study post (~230 words). The bulk is an architecture-and-procurement walkthrough (problem → constraint → chosen alternative → cost delta → two named benefits). The closing turns argumentative: status-quo licence-buying is sloth, and Frederiksen is happy competitors keep doing it. Reads as a confident practitioner field report, not a research piece.
Structure
| Part | Role |
|---|---|
| Hook: 95% cheaper and superior AI — why we dropped Copilot at Coop Bank | headline / promise |
| Constraint setup: Copilot price (200,000 DKK/year for 80 users) and EU-egress risk under peak load | problem statement |
| Categorical rejection: 'Det går ikke i en bank' ('That doesn't fly in a bank') | regulatory framing |
| Alternative chosen: Open Source models hosted 100% in EU | thesis statement |
| Architecture: open-weight models (Gemma/Mistral) via Scaleway Model-as-a-Service + custom chatbot UI built in ~1 day | implementation detail |
| Benefit 1: Full data sovereignty — API calls remain in a clean European cloud environment | named benefit |
| Benefit 2: Massive savings — only pay for actual token usage; ~200,000 DKK → under 10,000 DKK for 80 users | named benefit + cost claim |
| Concession: Not a 1-1 replacement for Copilot | honest caveat |
| Reframe: Hyper-flexibility — business ideas can go to production 'in moments'; 'I've never seen that before' | reinterpretation of trade-off |
| Generalisation: 'Det betaler sig at udfordre status quo' ('Challenging the status quo pays off') | moral / call to action |
| Competitive coda: Glad competitors stick with the easy/expensive standard solution — widens our lead | rhetorical close |
Key terms
- Datasuverænitet (data sovereignty) Data Sovereignty
API calls and the data they carry never leaving an EU-hosted environment — operationalised as 'rent europæisk cloud-miljø' (a 'clean' European cloud environment), implicitly excluding any US-jurisdiction provider regardless of EU-region availability zones - Open-weight models Open-Weight Models
Models whose weights are openly published (Gemma, Mistral named) and can therefore be served by any infrastructure provider — distinct from API-only proprietary models - Model as a Service (Scaleway) Local AI
Hosted inference endpoint for open-weight models, billed on token consumption, run by Scaleway (French cloud provider) — the EU-sovereign substitute for the Azure/OpenAI stack underneath Copilot - Skræddersyet chatbot ('custom chatbot')
The user-facing UI/wrapper Coop Bank built themselves in under a day, sitting on top of Scaleway's hosted open-weight model endpoints - Følsomme bankdata (sensitive bank data)
Regulated banking-customer or operational data subject to Danish/EU financial supervision (Finanstilsynet, DORA, GDPR) — the specific data class whose extra-EU egress disqualifies Copilot - Status quo (the default solution)
Buying Microsoft Copilot licences as the unreflective enterprise-IT path — Frederiksen treats this as the lazy/sloth-emoji default that most peers will take
Propositions
- explicit-empiricalMicrosoft Copilot would have cost Coop Bank approximately 200,000 DKK per year for 80 users.conf: high
- explicit-empirical/interpretiveMicrosoft Copilot carries a risk that sensitive bank data is sent outside the EU during peak load.conf: medium
- explicit-prescriptiveExtra-EU data egress is disqualifying for a bank ('Det går ikke i en bank').conf: high
- explicit-empiricalCoop Bank instead hosts open-weight models (Gemma, Mistral) 100% in the EU via Scaleway Model as a Service.conf: high
- explicit-empiricalCoop Bank built its own chatbot UI on top in under a day.conf: high
- explicit-interpretiveThe chosen architecture provides full data sovereignty because API calls remain in an EU-only cloud environment.conf: medium
- explicit-empiricalAnnual cost dropped from ~200,000 DKK to under 10,000 DKK for the same 80 users — a >95% reduction.conf: medium
- explicit-empiricalThe new stack is not a 1-1 replacement for Copilot.conf: high
- explicit-empirical/interpretiveThe new stack is hyper-flexible and allows new business ideas to reach production in moments.conf: medium
- explicit-prescriptiveChallenging the status quo pays off (general moral).conf: high
- explicit-empirical/interpretiveMany competitors will continue choosing the standard (Copilot) solution.conf: medium
- explicit-interpretiveCompetitors sticking with Copilot widens Coop Bank's competitive lead.conf: low
- implicitToken-based pricing (pay for actual consumption) is structurally cheaper than per-seat enterprise licensing for this usage shape.conf: medium
- implicitcorroboratedOpen-weight models (Gemma/Mistral) at the current capability frontier are good enough for Coop Bank's internal-knowledge-worker AI use cases.conf: mediumsupports ←Alexander Haneng's Post · p1 — Both posts assert that Gemma-family open-weight models are 'good enough' for real use — Frederiksen for enterprise knowledge-work at a bank, Haneng for a personal ChatGPT substitute — though neither provides benchmark evidence.
- implicitThe Scaleway/open-weight stack is more flexible than Copilot for adding new business use cases.conf: medium
Arguments
For an EU-regulated bank, the Scaleway-hosted open-weight-models path is strictly dominant over Microsoft Copilot. strength: high
Explicit premises:
- Copilot costs ~200,000 DKK/year for 80 users at Coop Bank.
- Copilot carries a risk of data egress outside the EU under peak load.
- EU-egress is disqualifying for sensitive bank data.
- The Scaleway/open-weight stack costs under 10,000 DKK/year for the same users and keeps all calls in the EU.
Implicit premises:
- Scaleway is operationally and contractually trustworthy as an EU-sovereign provider.
- Open-weight models (Gemma/Mistral) deliver sufficient quality for the use cases Copilot was being considered for.
- The internal engineering cost of building and maintaining the custom chatbot UI is small enough to ignore relative to the licence savings.
- The cost comparison is apples-to-apples — i.e. usage volume on the new stack is comparable to the per-seat Copilot model.
Inferences:
- Banks evaluating Copilot should run a dual-track procurement that includes an EU-sovereign open-weight alternative before signing.
- Per-seat enterprise AI licensing may be structurally mispriced for low-to-moderate-intensity internal use cases.
- The 'standard solution' in regulated EU contexts may be the worst-value option on both axes simultaneously (cost and sovereignty).
Challenging the vendor-default procurement path pays off (generalised from the Coop Bank case). strength: low
Explicit premises:
- Coop Bank achieved a >95% cost reduction and improved sovereignty by rejecting the default.
- Many competitors will choose the default Copilot path.
- Therefore, those who challenge it can build a competitive lead.
Implicit premises:
- The competitive arena rewards back-office IT cost efficiency or AI delivery velocity (otherwise the 'lead' is illusory).
- Coop Bank's experience is replicable at other Danish/EU banks of comparable size.
- The trade-offs Frederiksen accepted (no 1-1 Copilot parity, in-house engineering responsibility) are net positive in the long run, not just at deployment time.
Inferences:
- Vendor-default thinking in regulated EU IT procurement deserves systematic scrutiny.
- Engineering-led organisations have a procurement advantage over IT-led organisations in the AI era.
Narratives
- Coop Bank's Copilot evaluation → rejection → in-house Scaleway-hosted alternative — done quickly enough that the UI was built in under a day.
→ Concrete case proving the abstract sovereignty/cost argument; the speed of the build also functions as a credibility marker (this isn't a multi-year procurement saga, it's a small team executing). - The 95% cost compression as headline number — '200,000 DKK → under 10,000 DKK'.
→ Memorable, shareable figure that turns the post into a quotable LinkedIn artefact; reframes AI procurement from 'enterprise sticker price' to 'consumption economics'. - The 'sloth competitor' coda — 'glad competitors are sticking with the easy and expensive option'.
→ Self-positioning move; converts the case study into a competitive boast and signals tribal identification with the engineering-challenger side of the Danish banking market.
Questions addressed
- ✅ Should regulated EU banks buy Microsoft Copilot?
Frederiksen's answer is no — both cost and sovereignty fail. He demonstrates an alternative that beats both axes for Coop Bank's profile. - ✅ Is a hosted-open-weight stack operationally feasible for a bank?
Yes — Scaleway MaaS plus a one-day custom UI was sufficient to ship something internal users use. - ✅ How big is the cost gap between per-seat AI licensing and consumption-billed open-weight inference?
For 80 users at Coop Bank: >95% (200K DKK vs. <10K DKK annual). Not generalised to other usage profiles. - ⭕ What capabilities are lost by moving off Copilot?
Frederiksen concedes 'not a 1-1 replacement' but doesn't enumerate what's missing. The IDE integration, M365 grounding, and developer-coding assistance that come bundled in Copilot are silently dropped in the comparison. - ⭕ What is the full TCO of running an in-house chatbot stack including engineering maintenance, model upgrades, eval, prompt engineering, and incident response?
Entirely absent. The cost figure is the Scaleway invoice; the loaded engineering cost is not priced in. - ⭕ What use cases is the Coop Bank chatbot actually serving?
Not specified. 'New ideas from the business' is vague — whether this includes regulated customer-facing flows, internal Q&A, document search, or coding assistance is unclear. Capability gap with Copilot depends entirely on this. - ⭕ Is Scaleway's sovereignty claim legally water-tight for Danish-regulated bank data?
Frederiksen takes the residency claim at face value; the deeper jurisdictional question (Scaleway is French-owned, no US-jurisdiction parent) is not engaged in the post but is presumably central to the procurement decision. - ⭕ Does this generalise to larger banks or to capability-frontier use cases?
Implicitly generalised by the 'challenge status quo' framing, but the case is one mid-size cooperative bank with 80 users on a presumably-modest set of use cases. Frontier model capability (Claude, GPT-class) is not on the table in his comparison.
Gaps
- incompleteThe post never names which use cases the chatbot is serving. 'Hyper-flexible — business ideas to production in moments' is evocative but unactionable. Whether this stack is replacing Copilot Chat for general knowledge work, M365 Copilot for document grounding, or GitHub Copilot for coding makes a huge difference to whether the 'not a 1-1 replacement' concession is a small footnote or a load-bearing caveat.
Read next: Look for follow-up posts or case studies from Coop Bank on actual use cases. Compare to Norwegian-bank or DNB AI-adoption pieces. Anthropic's enterprise PDF in this corpus has a better breakdown of what enterprise users actually use AI for. - incompleteTCO blind spot. The 10,000 DKK figure is Scaleway's invoice. It excludes the engineering team's time to build and maintain the custom UI, evaluate models, manage prompt regressions on model updates, handle incidents, and own the security review. Microsoft's 200,000 DKK at least bundles those responsibilities into a vendor SLA. The fair comparison would include loaded engineering cost — likely 1-3 FTE-equivalent depending on scope, which could be a multiple of the licence savings.
Read next: DORA/state-of-AI reports on hidden costs of self-hosted AI; HashiCorp/Cloudflare PMOs on build-vs-buy economics; any post-mortem of a 'we built our own ChatGPT' enterprise project at 6/12/24 month checkpoints. - incompleteCapability gap is conceded but not priced. Microsoft Copilot's value proposition is heavily about M365 grounding (your SharePoint, email, Teams content) and IDE integration, neither of which is replicable in one day of custom UI work on top of Scaleway. The cost-vs-cost comparison may be comparing two different products.
Read next: Compare to recent extractions: HBR managers-AI-productivity piece, Anthropic enterprise PDF on what users actually do with Copilot. Microsoft's own published case studies on Copilot ROI claims. - incompleteSovereignty definition is implicit. 'Rent europæisk cloud-miljø' could mean (a) EU data residency only, (b) EU-owned operating entity (Scaleway parent: Iliad SA, French), or (c) full immunity from US CLOUD Act/extraterritorial discovery. These are very different bars and matter differently to a Danish bank under DORA and Finanstilsynet supervision. Frederiksen doesn't disambiguate.
Read next: Schrems II commentary; EU Cloud Code of Conduct; Danish Finanstilsynet guidance on cloud sourcing for credit institutions; the EUCS sovereignty levels (High/Substantial) debate. - incompleteCompetitive-lead claim doesn't connect. The post leaps from 'our IT stack is cheaper and more sovereign' to 'this widens our competitive lead'. Retail/cooperative banking competition is driven by deposit rates, distribution, brand trust, regulatory standing — not by internal AI chatbot choice. The closing boast is a rhetorical flourish, not an argued claim.
Read next: Look for case studies where a specific AI-IT decision moved a measurable bank KPI (NIM, cost-to-income ratio, customer acquisition cost). Beck/Shoup Jevons-paradox piece for why cheaper IT often gets eaten by induced demand rather than dropping to the bottom line. - uninformedThe post predates / sidesteps the harness-engineering thread in this corpus. Frederiksen is solving the procurement layer (which model, hosted where, billed how). Pocock / Anthropic / Cloudflare Mythos / smallcode are solving the orchestration layer above it (how to actually get reliable work out of any model, sovereign or not). The two layers are orthogonal — but a complete enterprise AI strategy needs both, and Frederiksen's post implies that getting procurement right is sufficient. It isn't.
Read next: Pocock workshop extraction; Anthropic harness paper; Bernhard post on value-stream targeting; Cloudflare Mythos for the orchestration economics. - uninformedDoesn't engage with the Norwegian-public-sector / Nordic AI-sovereignty parallel discussion the user has been tracking (Entur, NAV, Heis ep 11 on AI in the public sector). The Danish-bank case is structurally similar to public-sector deployment decisions that this corpus already covers — same EU-egress constraint, same Microsoft-default temptation. The cross-Nordic comparison would strengthen the case.
Read next: Heis ep 11 AI-special; the Entur and NAV AI adoption pieces in the user's library; any Finanstilsynet/Datatilsynet (Norwegian) guidance on AI sourcing. - incompleteThe 'under a day' UI build is the most contentious empirical claim in the post and the least examined. In most banks, the path from 'we built a chatbot UI' to 'production-grade chatbot serving internal users under bank IT governance' includes security review, authentication, audit logging, prompt-injection mitigations, content filtering, and operational monitoring — none of which fit in a day. Either the chatbot is internal-pilot scope and the production claim is loose, or substantial prior infrastructure made the one-day claim possible (auth, logging, deployment pipeline already in place).
Read next: OWASP LLM Top 10; Anthropic's enterprise security posture documentation; case studies of bank chatbot post-mortems on production-grade rollouts.
Related documents
- AI Costs Are Surging and the Cheap Model Fix Might Not Last
- GitHub - musistudio/claude-code-router: Use Claude Code as the foundation for coding infrastructure, allowing you to decide how to interact with the model while enjoying updates from Anthropic.
- How the Escalating AI Wars Benefit You
- Local AI
- Generative AI in the Real World: Local Voice AI with Pete Warden
- AI models can disappear overnight. Is your engineering team built to survive it?
- AI Sovereignty: Bargaining with Big Tech and the Promise of Full Stack Open Source AI
- Alexander Haneng's Post
- Baseten CEO Tuhin Srivastava on the AI Inference Crunch, Custom Models
- Building AI Agents For The Enterprise