LLM Landscape 2026: Strategic Guide for Enterprise Decision-Makers

Introduction: Why the LLM Market Demands C-Level Attention Now

The large language model (LLM) market has fundamentally transformed. As of early 2026, over a dozen frontier models compete across a 1,000× price range—from $0.05 to $168 per million tokens. For C-level decision-makers in Germany, Austria, and Switzerland, the question is no longer whether to deploy LLMs, but which models, for which tasks, under what regulatory framework, and at what cost.

Enterprise spending on generative AI reached $37 billion in 2025, representing a 3.2× increase year-over-year. Yet 30% of all GenAI projects are discontinued after proof of concept—primarily due to inadequate risk controls, unclear business value, or regulatory uncertainty. The DACH region faces particularly complex challenges: the EU AI Act's high-risk obligations take effect in August 2026, GDPR enforcement for AI is intensifying, and German, Austrian, and Swiss regulators are each building distinct national frameworks.

This strategic guide provides the intelligence enterprise leaders need to navigate the 2026 LLM landscape with confidence, combining technical depth with regulatory clarity and cost optimization strategies.

The 2026 LLM Market: Three Structural Shifts Reshaping Enterprise Strategy

The frontier LLM market in early 2026 is defined by three fundamental transformations that directly impact enterprise deployment decisions.

Pricing has collapsed by approximately 80% year-over-year. What cost $25 per million output tokens in early 2025 now costs $5 or less. DeepSeek V3.2 delivers competitive performance at $0.28 per million output tokens—roughly 100× cheaper than GPT-5.2 Pro. This dramatic price compression makes previously cost-prohibitive use cases economically viable and shifts the total cost of ownership calculation toward operational considerations rather than pure API costs.

Context windows have standardized at one million tokens. Google Gemini offers 1M token context as standard across all models. Claude provides 200K standard with 1M in beta. Meta's Llama 4 Scout variant supports an industry-record 10M token context window. Extended context windows enable entirely new application architectures—processing entire codebases, analyzing quarterly reports in single prompts, and maintaining conversation state across complex multi-step workflows without expensive retrieval systems.

Reasoning models with explicit chain-of-thought capabilities have become the primary differentiation factor. OpenAI's o3 and o4 series, Claude's extended thinking modes, and DeepSeek's R1 model represent a shift from pattern matching to systematic problem decomposition. GPT-5.2 Pro achieves 93.2% on GPQA Diamond (PhD-level science questions), while DeepSeek R1 earned gold medals at IMO, ICPC World Finals, and IOI 2025. Enterprise applications requiring complex analysis, strategic planning, or technical problem-solving now have access to capabilities that approach domain expert performance.

Comprehensive LLM Comparison 2026: Capabilities, Costs, and Strategic Positioning

Proprietary Market Leaders

Anthropic Claude currently leads human preference rankings. Claude Opus 4.6 (February 2026) achieves the highest Chatbot Arena Elo score (~1503) and dominates agentic coding benchmarks with a 14.5-hour autonomous task completion horizon. The pricing structure positions Claude strategically: Opus 4.6 at $5/$25 per million input/output tokens for frontier reasoning, Sonnet 4.6 at $3/$15 delivering near-Opus quality for standard production workloads, and Haiku 4.5 for high-volume lightweight automation. Anthropic holds 32–40% enterprise market share and dominates code generation with 42–54% market share. Claude's strength lies in nuanced instruction following, multilingual capability across German, French, and Italian, and consistent performance without the quality variance that affects some competitors.

OpenAI is transitioning to the GPT-5 family, with GPT-4o, GPT-4.1, o3, and o4-mini being phased out since February 2026. The current lineup spans from GPT-5 nano ($0.05/$0.40) for simple classification to GPT-5.2 Pro ($21/$168) for maximum reasoning capability. OpenAI maintains 25–27% enterprise market share and offers the broadest model lineup, but rapid deprecation cycles and premium pricing in the top segment create friction for enterprise customers requiring long-term stability. The strategic advantage: deepest ecosystem integration with Microsoft Azure, most mature API infrastructure, and strongest brand recognition among non-technical stakeholders.

Google Gemini 3.1 Pro (February 2026) delivers the best native multimodal capabilities—processing text, images, audio, video, and PDFs without preprocessing. All Gemini models support 1M token context windows as standard, and the Gemini 2.5 Flash-Lite tier provides usable quality at only $0.075/$0.30 per million tokens. Deep ecosystem integration with Gmail, Google Docs, Android, and Google Cloud Platform makes Gemini particularly attractive for organizations already invested in Google infrastructure. Performance on coding benchmarks lags Claude and GPT-5, but multimodal capabilities and pricing create compelling use cases for document-heavy workflows.

Open-Weight Challengers Disrupting Enterprise Economics

DeepSeek V3.2 (China) has fundamentally reset pricing expectations at $0.14/$0.28 per million tokens while achieving gold medal results at IMO, ICPC World Finals, and IOI 2025. All DeepSeek models release under the permissive MIT license. The critical constraint: Chinese censorship requirements, geopolitical risks, and server instability make DeepSeek unsuitable as a sole provider for European enterprises. However, as a self-hosted model behind a European firewall, these concerns largely disappear. DeepSeek represents the most aggressive price-performance ratio available and forces proprietary providers to justify premium pricing.

Alibaba Qwen has established itself as the most versatile open-weight ecosystem. Qwen 3.5 (February 2026) supports 201 languages under the Apache 2.0 license—the gold standard for enterprise use without commercial restrictions. The lineup ranges from 0.6B parameters (edge devices) to over one trillion (cloud deployment). The Qwen3-Coder variant claims 83× lower cost than Claude Opus for coding tasks. Over 300 million downloads on Hugging Face demonstrate massive community adoption. For DACH enterprises requiring multilingual support, data sovereignty, and unrestricted commercial use, Qwen represents the strongest open-source foundation.

Meta Llama 4 (April 2025) introduced a mixture-of-experts architecture with an industry-record 10M token context window in the Scout variant. Llama 4 Maverick activates only 17B of its 400B total parameters per token, optimizing inference costs. Critical consideration: Meta's Llama Community License excludes EU users from certain provisions and requires a separate license above 700M monthly active users. DACH enterprises must carefully review terms. Llama's advantage: largest open-source community, most extensive fine-tuning resources, and strongest ecosystem of derivative models.

Mistral AI (France) occupies a strategically unique position for European enterprises. Mistral Large 3 (December 2025) is a 675B MoE model under Apache 2.0, and the Devstral 2 coding model achieved 72.2% on SWE-bench Verified—state-of-the-art for open-weight coding. Mistral excels at European languages, offers full self-hosting, and represents genuine European digital sovereignty. Pricing at $2/$6 per million tokens positions Mistral between premium closed-source and budget open-source options. For organizations prioritizing European data residency and regulatory alignment, Mistral provides frontier-competitive performance without US or Chinese dependencies.

European Sovereignty Models: Strategic Options for Regulated Industries

Aleph Alpha (Heidelberg) has pivoted to PhariaAI—an enterprise GenAI operating system emphasizing explainability, on-premise deployment, and guaranteed European data residency. The T-Free tokenizer-free architecture promises up to 70% compute cost reduction. Target market: government, public sector, defense, and critical infrastructure. Performance on standard benchmarks trails frontier models, but the value proposition centers on compliance, auditability, and sovereignty rather than raw capability.

OpenEuroLLM project (€37–52M EU funding, 20+ participants) is building open-source multilingual LLMs for all 24 EU languages. Switzerland launched Apertus (CHF 20M state funding) as its first public multilingual open-source LLM. None of these models compete on raw benchmarks with frontier models, but they address genuine market demand: 88% of German enterprises consider the AI provider's country of origin important. For public sector and highly regulated industries, sovereignty models provide legally defensible alternatives to US and Chinese providers.

Open Source vs. Closed Source: The Enterprise Strategic Calculus

The capability gap between open-weight and proprietary models has narrowed to single-digit percentage points for most practical tasks. Yet closed-source LLMs still constitute ~87% of deployed enterprise workloads, with 41% of organizations planning to expand open-source deployment.

When Open Source Wins: Three Decisive Factors

Data sovereignty is the primary argument. Self-hosted models eliminate cross-border data transfer complexities under GDPR, provide full audit trail control, and remove the risk that the US CLOUD Act could compel American cloud providers to surrender European customer data. For financial services, healthcare, and government sectors, data residency isn't a preference—it's a legal requirement. Self-hosted open-source models provide the only architecture that guarantees data never leaves European jurisdiction.

Self-hosting becomes cost-effective above approximately two million tokens per day. Below this threshold, API pricing is cheaper when accounting for GPU infrastructure ($15,000–$50,000+ monthly), personnel costs (typically 5–10 FTE), and operational overhead. Above this threshold, the economics reverse dramatically. One fintech case study reduced monthly AI spending from $47,000 to $8,000 (83% reduction) through hybrid self-hosting. At enterprise scale—tens of millions of tokens daily—self-hosting delivers order-of-magnitude cost advantages.

Customization and fine-tuning requirements favor open weights. Proprietary APIs offer limited customization—primarily through prompt engineering and retrieval-augmented generation. Open-weight models enable domain-specific fine-tuning, custom tokenizers for specialized vocabularies, and architectural modifications for specific performance profiles. Industries with specialized terminology (legal, medical, technical) or unique compliance requirements benefit substantially from fine-tuning capabilities unavailable with closed-source models.

When Closed Source Remains Superior: Three Scenarios

Frontier reasoning quality is paramount. Claude Opus 4.6 and GPT-5.2 Pro continue to lead on the most difficult benchmarks. When the task requires PhD-level analysis, complex strategic reasoning, or novel problem-solving, the 5–15% performance advantage of frontier closed-source models justifies premium pricing. Customer-facing applications where quality directly impacts brand perception should prioritize the highest-capability models regardless of cost.

Time-to-market is critical. Proprietary APIs enable production deployment in days rather than months. No infrastructure provisioning, no model selection and benchmarking, no fine-tuning pipeline development. For startups, pilots, and rapid innovation cycles, closed-source APIs remove operational complexity and accelerate value realization. The opportunity cost of delayed deployment often exceeds the total API costs.

Lack of internal ML infrastructure capability. Self-hosting requires specialized expertise: ML engineers, infrastructure specialists, security teams, and ongoing operational support. Organizations without existing ML capabilities face 6–12 month buildout timelines and substantial hiring costs. For companies where AI is important but not core competency, managed API services provide professional-grade capability without building internal expertise.

The Optimal Strategy: Hybrid Architecture

The most sophisticated DACH enterprises—already 37% of organizations—deploy hybrid strategies: sensitive, high-volume workloads on self-hosted open models; customer-facing interactions and complex reasoning tasks on proprietary APIs. This architecture delivers 40–60% cost savings versus single-model approaches while optimizing for performance, compliance, and risk management across different use case profiles.

Three-Tier LLM Routing Architecture: Maximizing Performance Per Dollar

No single LLM is optimal for all tasks. The most cost-effective enterprise architecture routes requests to different models based on complexity, achieving 40–60% cost reduction versus single-model approaches.

Tier 1 – Frontier Reasoning (15–20% of requests)

Models: Claude Opus 4.6 or GPT-5.2 Pro

Cost: $5–$168 per million output tokens

Use cases: Complex analysis requiring multi-step reasoning, production code generation, legal/compliance review, strategic decision support, novel problem-solving

Routing logic: Requests explicitly flagged as high-complexity, tasks requiring domain expert-level reasoning, customer-facing scenarios where quality is paramount

Frontier models justify their premium pricing for tasks where incremental quality improvements deliver disproportionate business value. A 5% improvement in legal contract analysis accuracy prevents costly disputes. A 10% improvement in strategic analysis quality influences million-dollar decisions. Tier 1 deployment should be selective but unrestricted by cost when business impact warrants premium capability.

Tier 2 – Mid-Tier Production (40–50% of requests)

Models: Claude Sonnet 4.6, GPT-4o, or Gemini 3.1 Pro

Cost: $1–$15 per million tokens

Use cases: Customer-facing interactions, content creation, marketing automation, data analysis, document processing, general business workflows

Routing logic: Default tier for most production workloads, requests requiring strong performance but not frontier reasoning

Tier 2 represents the sweet spot for enterprise deployment—delivering 90–95% of frontier model quality at 20–40% of the cost. Claude Sonnet 4.6 at $3/$15 provides near-Opus quality for standard production workloads. Most customer service, content generation, and analytical tasks perform excellently at this tier. Marketing teams report 30–45% productivity gains deploying Tier 2 models for campaign content, social media, and email automation.

Tier 3 – Lightweight Automation (30–40% of requests)

Models: Claude Haiku 4.5, GPT-5 nano, Gemini 2.5 Flash-Lite, or self-hosted Mistral/Qwen

Cost: $0.05–$2 per million tokens

Use cases: Classification, simple summaries, data extraction, high-volume preprocessing, sentiment analysis, entity recognition

Routing logic: Requests with simple, well-defined tasks; high-volume batch processing; internal workflows where minor quality variance is acceptable

Tier 3 handles the long tail of simple, repetitive tasks that consume significant token volume but don't require sophisticated reasoning. Gemini 2.5 Flash-Lite at $0.075/$0.30 delivers usable quality for classification and extraction tasks. Self-hosted Qwen 3.5-14B on European infrastructure provides GDPR-compliant, cost-effective processing for high-volume internal workflows. Proper Tier 3 deployment can reduce overall AI spending by 40–60% while maintaining quality for complex tasks.

Task-Specific LLM Recommendations: Matching Models to Business Outcomes

Customer Service & Chatbots

Recommended: Claude Sonnet 4.6 for nuanced multilingual responses in German, French, and Italian; Gemini 3.1 Pro for organizations with Google Workspace integration

Architecture: RAG with company knowledge base, Tier 2 model for responses, Tier 1 escalation for complex issues

Results: A European bank achieved 20% CSAT improvement in seven weeks deploying Claude Sonnet with custom knowledge integration

Customer service represents one of the highest-ROI LLM applications. The combination of reduced response time, 24/7 availability, and consistent quality drives measurable satisfaction improvements. Critical success factors: comprehensive knowledge base, escalation paths to human agents, and multilingual capability for DACH markets.

Content Creation & Marketing Automation

Recommended: GPT-4o for high-volume campaign content; Claude Sonnet 4.6 for long-form brand voice content; Gemini Pro for real-time data integration

Architecture: Agentic workflows automating end-to-end campaign creation, distribution, and optimization

Results: Marketing teams report 30–45% productivity gains; 81% of marketing technology leaders are piloting AI agents

Marketing automation represents the fastest-growing LLM application category. Autonomous agents can plan campaigns, generate content, distribute across channels, and optimize based on performance—end-to-end workflows previously requiring multiple team members and days of coordination. Blck Alpaca specializes in exactly these agentic marketing workflows, combining multiple LLMs with custom automation to deliver enterprise-grade marketing operations.

Code Generation & Software Development

Recommended: Claude Opus 4.6 or Sonnet 4.6 (42–54% market share); Devstral 2 (Mistral, open-weight, 72.2% on SWE-bench Verified) for self-hosted coding assistants

Architecture: IDE integration, repository-level context, automated testing and review

Results: Development teams report 25–40% productivity improvements; reduced time-to-production for new features

Claude dominates code generation for good reason: superior instruction following, strong reasoning about code architecture, and excellent debugging capabilities. For organizations requiring self-hosted solutions, Mistral's Devstral 2 provides state-of-the-art open-weight performance. The 14.5-hour autonomous task completion horizon demonstrated by Claude Opus 4.6 enables truly agentic development workflows.

Document Processing & RAG Applications

Recommended: Any frontier model combined with vector database; self-hosted Qwen 3.5-122B (Apache 2.0) on European datacenter for GDPR-sensitive document analysis

Architecture: Document ingestion, embedding generation, semantic search, LLM synthesis

Results: RAG is the dominant enterprise integration pattern for 30–60% of use cases

Retrieval-augmented generation solves the fundamental LLM limitation: lack of current, proprietary, or domain-specific knowledge. By combining semantic search over company documents with LLM synthesis, RAG architectures provide accurate, sourced, and current responses. For DACH enterprises processing sensitive documents—legal contracts, financial records, HR files—self-hosted open-source models on European infrastructure provide GDPR-compliant document intelligence.

EU AI Act Compliance: The August 2026 Deadline and What It Means for LLM Deployment

The EU AI Act's high-risk system obligations take effect in August 2026, creating compliance requirements that directly impact LLM deployment strategies for DACH enterprises.

High-Risk System Classification

LLMs deployed in certain contexts are classified as high-risk systems requiring: conformity assessments before deployment, ongoing monitoring and logging, human oversight mechanisms, and transparency obligations. High-risk contexts include: employment decisions (hiring, promotion, termination), credit scoring and lending decisions, law enforcement applications, and critical infrastructure management.

The classification depends not on the model itself but on its application. The same LLM used for marketing content (minimal risk) versus hiring decisions (high risk) triggers different compliance obligations. DACH enterprises must conduct use-case-specific risk assessments for every LLM deployment.

Compliance Architecture Requirements

Data governance: High-risk systems require training data that is "relevant, representative, free of errors and complete." For proprietary models, providers must demonstrate compliance. For fine-tuned or self-hosted models, the deploying organization bears responsibility. This requirement favors established providers with documented data governance over smaller or newer models with limited transparency.

Technical documentation: Enterprises must maintain detailed documentation of model capabilities, limitations, performance metrics, and risk mitigation measures. This documentation must be available to regulators upon request. Open-source models provide transparency advantages—full architectural details, training processes, and evaluation metrics are typically public. Closed-source models require reliance on provider documentation.

Human oversight: High-risk systems must enable human oversight, including the ability to interrupt system operation, understand system outputs, and override system decisions. LLM architectures must incorporate human-in-the-loop mechanisms for high-risk applications. Fully autonomous agentic workflows may require architectural modifications to meet oversight requirements.

Strategic Implications for Model Selection

EU AI Act compliance creates several strategic considerations: European providers gain competitive advantage—Mistral AI, Aleph Alpha, and OpenEuroLLM projects benefit from regulatory alignment and reduced cross-border complexity. Self-hosted models provide compliance flexibility—full control over data, logging, and oversight mechanisms simplifies compliance demonstrations. Proprietary API providers must contractually commit to compliance support—enterprises should require AI Act-specific provisions in vendor contracts, including indemnification for non-compliance resulting from provider actions.

The August 2026 deadline is imminent. DACH enterprises deploying LLMs in high-risk contexts should initiate compliance assessments immediately, prioritizing use cases by risk level and business impact.

Where LLMs Must Not Be Deployed: Understanding Failure Modes and Risk Boundaries

Global business losses from AI hallucinations reached $67 billion in 2024. Understanding where LLMs fail is strategically as important as understanding where they excel.

Hallucination Rates Remain Significant

Even the best models hallucinate 0.7–0.8% of the time on simple summarization tasks. For domain-specific queries, rates explode: 69–88% for specific legal questions, 15.6% for medical queries, and 18.7% for legal questions generally. A critical paradox: MIT researchers found models hallucinate more confidently when wrong—they express higher certainty in incorrect responses than correct ones, making error detection more difficult.

Prohibited and High-Risk Deployment Scenarios

Autonomous medical diagnosis or treatment recommendations: Hallucination rates and lack of liability framework make unsupervised medical LLM deployment legally and ethically untenable. LLMs can assist medical professionals but must not make autonomous clinical decisions.

Financial advice without human review: Investment recommendations, tax planning, and financial product selection require regulatory compliance and fiduciary responsibility that LLMs cannot assume. LLMs can draft analyses but require licensed professional review.

Legal document generation without attorney review: While LLMs excel at legal drafting, they cannot replace attorney judgment. Contracts, regulatory filings, and legal opinions generated by LLMs require qualified legal review before execution.

Safety-critical systems without redundant verification: Industrial control, transportation systems, and physical infrastructure management require reliability guarantees that current LLMs cannot provide. LLMs may provide decision support but must not autonomously control safety-critical systems.

Mitigation Strategies for Acceptable Use

When LLMs are deployed in sensitive contexts, implement: Human-in-the-loop verification for all consequential outputs, multi-model consensus requiring agreement between multiple LLMs before accepting outputs, confidence thresholds rejecting responses below specified certainty levels, retrieval-augmented generation grounding responses in verified source documents, and comprehensive logging enabling full audit trails for compliance and error analysis.

Implementation Roadmap: From Strategy to Production

Phase 1: Assessment & Architecture (Weeks 1-4)

Conduct comprehensive use case inventory across the organization, identifying all potential LLM applications. Classify each use case by EU AI Act risk level (minimal, limited, high, unacceptable). Perform cost-benefit analysis for each use case, estimating token volumes, required model tiers, and expected business impact. Design three-tier routing architecture matching organizational use case portfolio. Establish data governance framework ensuring GDPR compliance and AI Act readiness.

Phase 2: Pilot Deployment (Weeks 5-12)

Select 2-3 high-value, low-risk use cases for initial deployment. Implement technical infrastructure: API integrations for closed-source models, self-hosting infrastructure for open-source models if economically justified, vector databases for RAG applications, and monitoring and logging systems. Deploy pilot applications with limited user groups. Collect performance metrics, user feedback, and cost data. Refine routing logic and model selection based on pilot results.

Phase 3: Scaled Rollout (Weeks 13-26)

Expand successful pilot applications to broader user populations. Implement additional use cases prioritized by business impact and risk profile. Establish center of excellence for LLM governance, bringing together legal, compliance, IT, and business stakeholders. Develop internal training programs ensuring responsible AI use across the organization. Implement comprehensive monitoring dashboards tracking cost, performance, compliance, and business outcomes.

Phase 4: Optimization & Innovation (Ongoing)

Continuously optimize routing logic based on performance and cost data. Evaluate new models as they release, updating architecture to leverage capability improvements and price reductions. Expand to more sophisticated applications: agentic workflows, multi-model ensembles, and custom fine-tuned models. Maintain regulatory compliance as frameworks evolve, adapting architecture to meet new requirements.

Conclusion: Strategic Imperatives for DACH Enterprises

The 2026 LLM landscape presents DACH enterprises with unprecedented opportunity and complexity. Five strategic imperatives emerge from this analysis:

Adopt hybrid architecture strategies. No single model or provider optimizes for all use cases. The most sophisticated enterprises deploy three-tier routing architectures, combining frontier closed-source models for complex reasoning, mid-tier models for standard production workloads, and lightweight or self-hosted models for high-volume automation. This approach delivers 40–60% cost savings while maintaining quality where it matters.

Prioritize EU AI Act compliance now. The August 2026 deadline for high-risk system obligations is imminent. Enterprises must conduct use-case-specific risk assessments, implement required governance frameworks, and ensure technical architectures support compliance requirements. European providers and self-hosted models offer compliance advantages worth considering in procurement decisions.

Evaluate open-source models seriously. The capability gap has narrowed to single-digit percentage points for most tasks. For organizations processing sensitive data, requiring multilingual support, or operating at scale, open-source models under permissive licenses (Apache 2.0) provide data sovereignty, cost efficiency, and customization capabilities unavailable with closed-source APIs. Qwen 3.5 and Mistral Large 3 deserve evaluation alongside Claude and GPT-5.

Implement robust risk management. Hallucination rates remain significant, particularly for domain-specific queries. High-stakes applications require human-in-the-loop verification, multi-model consensus, confidence thresholds, and comprehensive audit trails. Understanding where LLMs must not be deployed autonomously is as important as identifying high-value applications.

Partner with specialized AI agencies. The complexity of LLM selection, architecture design, regulatory compliance, and ongoing optimization exceeds most organizations' internal capabilities. Specialized agencies like Blck Alpaca combine technical expertise in LLM deployment with deep understanding of DACH regulatory requirements and industry-specific use cases, accelerating time-to-value while managing risk.

The enterprises that will lead their industries in 2026 and beyond are those that move beyond experimentation to systematic, compliant, cost-optimized LLM deployment across their operations. The technology is ready. The regulatory framework is clear. The competitive advantage awaits those who execute strategically.

Take Action: Transform Your Enterprise with Strategic LLM Deployment

The LLM landscape in 2026 offers DACH enterprises transformative capabilities—but only with the right strategy, architecture, and execution. Blck Alpaca specializes in enterprise AI marketing automation, combining deep technical expertise in LLM deployment with comprehensive understanding of EU AI Act compliance and DACH market requirements.

We design and implement three-tier LLM architectures optimized for your specific use case portfolio, cost constraints, and regulatory obligations. Our agentic marketing workflows automate end-to-end campaign creation, distribution, and optimization—delivering the 30–45% productivity gains leading enterprises are already achieving.

Ready to move from strategy to implementation? Contact Blck Alpaca to discuss your enterprise LLM strategy and discover how we can accelerate your AI transformation while managing cost, compliance, and risk.

Visit blckalpaca.at to explore our enterprise AI marketing automation solutions and schedule a strategic consultation with our team.

---

Originally published by Blck Alpaca - Data-Driven Marketing Agency from Vienna, Austria.