Services · AI Services

Put AI where it serves your business — pragmatically, with clear costs.

Microsoft Copilot when it pays off. Mistral or Aleph Alpha when EU hosting matters more. RAG with your documents when you want to make better use of your own knowledge. And governance so the EU AI Act doesn't become a surprise.

Microsoft AI Partner EU LLM experience (Mistral, Aleph Alpha) EU AI Act Advisory Sovereignty-aware

For Managing Directors · ROI before the pilot

No Copilot hype. A clear number: what does AI deliver for your business, in euros?

The AI Readiness Check (prices on request — separate tiers for SMB and mid-market) delivers an honest economic analysis per use case — Copilot license cost vs. productivity gain, RAG in-house build vs. 3-year cloud OpEx. Fixed price, no sales pressure. If AI doesn't pay off in your business, we say so openly — and you know it before the investment instead of after the failed pilot.

Start ROI conversation

For Department Heads · EU AI Act & adoption

You need to defend AI internally — to the works council, data protection, and management.

We deliver a documented outcome report: EU AI Act risk classification per use case, GDPR and works-council argumentation line, prioritized use-case matrix with adoption metrics (active users, prompt quality, measured time savings). A Discovery Spike is the typical first engagement, with a business-case skeleton for procurement and CFO — defensible at the next steering committee.

Request Discovery Spike

For IT Leadership · RAG architecture without marketing

Mistral vs. Aleph Alpha vs. Azure OpenAI — worked through on the whiteboard.

Direct with the architect — no account manager layer. Topics: RAG patterns (hybrid retrieval, re-ranking, citation grounding), vector DB choice (pgvector vs. Qdrant vs. Azure AI Search), MCP server topology, embedding strategy per language, tenant boundaries for sensitive data, EU hosting with Mistral La Plateforme or Aleph Alpha PhariaAI, latency budget, and token cost modeling.

45-min architecture conversation

Why arades GmbH

AI advisory that doesn't sound like a Microsoft sales pitch.

We're a Microsoft Partner — and we still (or precisely because of that) give honest advice on whether Microsoft Copilot is really the right path. Sometimes it's Mistral. Sometimes it's an in-house RAG build. Sometimes the answer is: not yet.

Platform comparison, not platform sales

We know Microsoft Copilot and Copilot Studio from daily practice — and we have Mistral, Aleph Alpha, and locally hosted open-source models in production use. In the Readiness Check we compare honestly, not from a sales interest.

Fixed prices instead of daily rates

Diagnostics, pilot, rollout, governance — all with clear timeframes and a fixed price. No open consultant accounts, no scope-creep arguments. Whatever doesn't fit in the pilot is clearly named and planned for phase 2.

EU AI Act from day one

We build governance not as an afterthought but as a foundation. Risk classification, documentation structures, Microsoft Purview integration — so you're prepared for audits, compliance inquiries, and works-council discussions.

Diagnostic packages

AI Readiness Check — fixed price, honest recommendation.

Before you invest, you know: where do you actually stand? Which use cases pay off? Which risk level falls under the EU AI Act? The AI Readiness Check answers these questions in a structured way — across five dimensions, with clear recommendations.

Recommended entry
Fixed price

AI Readiness Check

Prices on request (SMB & mid-market, net)

Structured maturity assessment in five dimensions: data maturity, technology maturity, use cases, governance/compliance, change readiness. Includes EU AI Act risk assessment of your planned use cases and an honest platform recommendation (Microsoft Copilot vs. EU LLM vs. in-house RAG).

Deliverables: 20-page readiness report, prioritized use-case list, platform recommendation with rationale, EU AI Act classification, 12-month roadmap outline, closing workshop (90 minutes).

Prerequisites: access to 2–3 interview slots (management, IT leadership, business line), data inventory overview, Microsoft 365 tenant read access (if available).

Not included: implementation, pilot build, license procurement — the Check is deliberately neutral, without sales pressure. If an implementation engagement follows, the Check fee is fully credited.

The five dimensions in detail

  • Data maturity — Which structured and unstructured data exists? What's the data quality? Where does the data sit (M365, on-prem, cloud storage)? Which sensitivity labels exist?
  • Technology maturity — Which Microsoft 365 licenses are in place? Is Entra ID tidy? Are Purview, sensitivity labels, DLP configured? Which integration interfaces are available?
  • Use cases — Which concrete use cases deliver measurable value? Which are technically mature? Which fail due to data or process gaps?
  • Governance & compliance — EU AI Act risk classification per use case, GDPR view, works-council topics, approval processes, documentation status.
  • Change readiness — What's your adoption track record so far (Teams, SharePoint, Power Platform)? How is change communicated? Who owns user training?

Why compact: the Check is a structured maturity assessment with a clear interview guide and scoring matrix — not a 360-degree consulting engagement. If it becomes clear that a deeper audit is needed (e.g., complex multi-tenant setups), we say so and propose a deep-assessment package.

Microsoft AI packages

Microsoft 365 Copilot and Copilot Studio — when the platform fits.

Microsoft 365 Copilot is the right choice when your users already work deeply in Microsoft 365. Copilot Studio fits when you need concrete agents — HR FAQ, IT helpdesk, sales assistance. We help with an honest evaluation and a clean rollout.

Adoption · 3 phases

Microsoft 365 Copilot Adoption

Price on request (net)

Structured 3-phase program for a clean Microsoft 365 Copilot rollout — from technical readiness through pilot to rollout with success measurement.

  • Phase 1 — Readiness: Microsoft Purview, Entra ID, sensitivity labels, DLP check, data inventory, license check. So Copilot doesn't answer over sensitive data it shouldn't see.
  • Phase 2 — Pilot: 20–50 selected users from two to three business areas, clear use cases, training, weekly feedback rounds, metric collection.
  • Phase 3 — Rollout & success measurement: phased expansion, 90-day success measurement with hard metrics (adoption rate, use-case hit rate, time-saved estimate, ROI derivation).

Honest assessment included: if after Phase 1 it's clear Microsoft 365 Copilot doesn't fit your use case — because data maturity is missing, the per-user value is too low, or an EU LLM solution makes more sense — we say so. You pay for Phase 1 and we propose an alternative.

Agent development

Copilot Studio Agent Development

Standard agent / Custom agent — prices on request

Concrete agents for clearly scoped use cases, built in Microsoft Copilot Studio — with connections to SharePoint, Teams, Dataverse, external APIs.

Standard agent:

  • HR FAQ agent: answers employee questions about vacation, travel expenses, payroll — based on your HR policy documents.
  • IT helpdesk agent: first-level answers to standard tickets (password, VPN, license requests) with clean escalation to ServiceNow or Jira.
  • Sales assistant: product information, price lists, competitive comparisons from your sales enablement material.

Custom agent: more complex logic, multiple data sources, external API calls, workflow integration, approval paths. Classic examples: sales configurator, internal compliance advisor, multi-step approval bot.

Platform-independent AI packages

When EU hosting matters more than Microsoft integration.

For IT leaders with sovereignty concerns, regulated industries, critical infrastructure, or data with trade-secret character: AI on EU LLMs, locally hosted, or open source. We build that with Mistral, Aleph Alpha, and locally hosted open-source models.

Pilot · 8–12 weeks

AI Integration with EU LLMs

Pilot — price on request (net)

Integration of an EU-hosted LLM into your application landscape — typically Mistral (France), Aleph Alpha (Germany), or a locally operated open-source model (Llama, Qwen, Mistral OSS).

Typical use cases:

  • Retrieval-Augmented Generation (RAG) on your own documents
  • Internal search with semantic retrieval
  • Classification and tagging of unstructured data
  • Translation and summarization pipelines
  • Text generation for internal tools (reports, email drafts)

Pilot deliverables: architecture draft, model selection with rationale, integration pilot in a clearly scoped use case, user testing, scaling roadmap.

RAG · fixed-price pilot

RAG Implementation: Document Chat / Knowledge Assistant

Fixed-price pilot — price on request (net)

Retrieval-Augmented Generation on your own documents — so the model answers from your knowledge, not general training material. With source citations and measurable answer quality.

Engineering building blocks:

  • Vector database: embeddings index for your documents (PDF, Word, HTML, SharePoint, Confluence)
  • Embeddings: selection and tuning of the right embedding model (multilingual, domain-specific)
  • Prompt engineering: structured prompts with source grounding and answer constraints
  • Quality metrics: answer relevance, source hit rate, hallucination rate, user ratings
  • Hallucination mitigation: constraints, fallback logic ("not found"), human-in-the-loop for critical topics

Use cases:

  • Internal knowledge management (employee self-service on wikis, handbooks, policies)
  • Customer support knowledge base (agent assist or direct customer bot)
  • Compliance search (contract research, standards lookups, audit preparation)

Governance & Compliance

The EU AI Act is no longer a "later" topic.

The EU AI Act has been in force since August 2024. High-risk obligations apply fully from August 2026 — anyone building AI today should set up the governance structures now, not retrofit later.

Fixed price + optional recurring

AI Governance & EU AI Act Compliance

Initial setup / quarterly reviews — prices on request

Governance structure for your AI use: risk classification of your use cases under the EU AI Act, documented approval processes, responsibilities, Microsoft Purview integration.

Initial setup (one-time, price on request):

  • Inventory of your AI use cases
  • EU AI Act risk classification (minimal / limited / high / prohibited)
  • Documentation structures (system cards, data logs, transparency obligations)
  • Approval processes: who decides on new AI use cases, with which escalation?
  • Microsoft Purview integration: sensitivity labels, DLP, audit logs
  • Training concept for users and decision-makers

Quarterly reviews (per quarter, price on request):

  • Update of the use-case inventory
  • Reassessment after case law and regulation updates
  • Sample audit of documentation quality
  • User feedback and incident reviews
  • Recommendations for the next quarter

Note: we don't provide legal advice. For legal assessment we recommend involving your legal department or a law firm specializing in IT law — we deliver the technical and procedural foundation on which such an assessment can build.

Recurring · Ongoing Support

AI Application Care — operations start at go-live.

Unlike traditional software, AI systems age faster. Models get replaced, prompts need re-tuning, hallucination rates change, costs go up or down. AI Application Care keeps your solution stable in production.

Monthly flat rate

AI Application Care

Price on request / month (net)

Ongoing care for your AI application — from model monitoring through prompt optimization to cost control. So what worked in the pilot still works in year two.

  • Model monitoring: latency, availability, answer quality, error rates
  • Prompt optimization: adjustments for changing user behavior or new use cases
  • Hallucination tracking: sample reviews, feedback analysis, quality metrics
  • Cost monitoring: token consumption, cost trends, budget alerts, optimization suggestions
  • Update care: response to new model releases (e.g., Mistral update, GPT switch) — test, validation, rollover
  • Quarterly review: user feedback, roadmap suggestions, EU AI Act compliance status

Pricing logic: the flat rate starts for a single AI application with a manageable user base. More complex setups (multiple models, high token volumes, regulated use cases with extended audit obligations) are calculated by effort. Prices on request.

Frequently asked questions

What we clarify before every AI engagement.

Microsoft Copilot vs. Mistral — which LLM is right for us?

The honest answer: it depends on which problem you want to solve — and which risk you're willing to carry. Microsoft 365 Copilot is the right choice when your users already work deeply in Microsoft 365 (Outlook, Word, Excel, Teams, SharePoint) and the value comes from productivity gains in everyday office work — the entry barrier is low, integration is native. Mistral or Aleph Alpha are the right choice when data sovereignty, EU hosting, or open-source control are mandatory (critical infrastructure, regulated industries, data with trade-secret character). In the AI Readiness Check we answer exactly this question — with an honest recommendation, even when it's: not yet.

What does AI really cost us per year?

Realistically there are three cost blocks: licenses (Microsoft 365 Copilot per user and month — for EU LLMs usage-based per million tokens), implementation (pilot plus rollout depending on scope), and ongoing operations (AI Application Care as a monthly flat rate). We calculate the total cost for your use case transparently before any decision. Prices on request.

How does RAG work with your own documents?

Retrieval-Augmented Generation (RAG) connects a language model with your own documents: your PDFs, wikis, SharePoint content, Confluence pages, or database entries are indexed in a vector database. When a user asks a question, the system retrieves the thematically matching document excerpts and passes them to the LLM as context — the answer is then based on your knowledge, not on general training material. This drastically reduces hallucinations and enables source citations. Typical use cases: internal knowledge management, customer support knowledge base, compliance search, contract research. We build RAG pilots in 6–10 weeks — fixed price, price on request.

Do we need to implement the EU AI Act already?

Yes — but in stages. The EU AI Act has been in force since August 2024, the obligations apply in steps: prohibited practices since February 2025, general-purpose AI obligations from August 2025, high-risk systems fully from August 2026. If you're already deploying AI today or are concretely planning to, you should now classify the risk of your use cases and build the documentation structures — retrofitting later is consistently much more expensive. For this we offer a fixed-price package: AI Governance & EU AI Act Compliance (one-time initial setup plus optional quarterly reviews). Prices on request.

How do you handle hallucinations?

Hallucinations — model-invented, plausible-sounding, but wrong answers — are not a solved problem but an engineering discipline. We use four levers: first, RAG with clear source grounding so the model cites your documents explicitly. Second, prompt engineering with constraints ("Answer only if the information is in the context — otherwise say: not found"). Third, quality metrics in operations: hallucination-rate tracking, sample reviews, user feedback loops. Fourth, governance: clearly defined use cases where LLM answers must be reviewed by humans — for example compliance texts, legal opinions, medical recommendations. AI Application Care monitors these metrics continuously.

Related services

AI rarely stands alone — these topics connect.

To take with you · two materials

Factsheet and whitepaper.

Two depths for different reading needs. The factsheet is a quick reference (3–5 min) and immediately downloadable. The whitepaper is market education with methodology and comparison data (15–30 min) — you get it by email after a short request.

Factsheet · 2 pages

AI & Copilot Factsheet

3–5 min read · direct download · no form

Concise overview: scope, key figures, pricing model, process — ideal to forward to CFO, procurement, or the business line.

Download factsheet (PDF)

Whitepaper · 12 pages

AI & Copilot — deep dive

15–30 min read · by email on request

Methodology, comparison data, recommendation framework — material for internal argumentation with stakeholders.

AI Readiness Check

Where are you today — and which AI fits you?

Fixed price, honest recommendation. We clarify across five dimensions where you really stand, which use cases pay off, and which platform fits — Microsoft Copilot, EU LLM, or in-house RAG. If an implementation engagement follows, the Check fee is fully credited.

To take with you

AI Services factsheet.

Two-page quick reference with package structure, deliverables, and three reasons for arades — immediately downloadable, no form. Ideal to forward to CFO, procurement, or IT lead.

Factsheet · 2 pages · PDF

AI Services factsheet

3–5 min read · direct download · no form

Download factsheet (PDF, 6 KB)