Build vs. Buy

Mid-sized companies caught in the strategy gap

Three pressure factors are pushing German mid-sized companies into the AI decision in 2026: talent shortage, rising process complexity, and EU AI Act regulation [13]. This year marks the transition from experimentation to operational deployment [13]. Yet the data reveals how unprepared many companies are for this shift.

The FZI study (517 decision-makers from companies with 20 to 500 employees) paints a sobering picture: 40% already use AI, but only 21% have an AI strategy [8]. 64% of AI-adopting companies operate without a formal strategy [8]. For LLMs, 73% of companies permit their use, with 48% allowing unrestricted access to freely available models [8]. Larger mid-sized companies with 250 to 500 employees lean more toward in-house development, with 49% AI adoption and 35% running proprietary models [8]. Sales and marketing lead as the primary use case at 15% intensive adoption [8].

The Cognizant study (600 AI decision-makers, March 2026) confirms this picture at the enterprise level. Compliance (33%), ROI proof (31%), talent shortage (27%), and data quality (27%) rank as the most common barriers [7]. "Generic, off-the-shelf AI solutions" are cited as the primary reason for rejecting AI vendors [7]. Even the highest expected full automation sits at just 20% in sales, while customer service operates AI-dominantly at 76% but reaches only 9% full automation [7]. 84% of companies maintain formal AI budgets, 52% already invest over USD 10 million annually, and 91% expect budget growth within the next two years [7]. The money is flowing. It just flows without strategic guardrails more often than not.

The EU AI Act raises the stakes. High-risk obligations take broad effect starting August 2026 [13]. Violations carry penalties of up to EUR 35 million or 7% of global revenue [6]. Mid-sized companies get some relief: simplified documentation and access to regulatory sandboxes [13]. But companies already operating without a strategy will hardly use regulation as the structuring opportunity experts recommend [13]. Regulation can serve as an argument for both sides: Buy for compliance assurance through experienced vendors, Build for full control over data processing and documentation [13]. Without a strategic framework, however, it remains above all one thing: an additional cost factor.

What actually works

The most consistent finding across all 16 sources analyzed: The answer is neither Build nor Buy. It follows a clear logic along three touchstones.

First: Differentiation potential. "Buy standard capabilities, build only where data and process are your differentiation lever" [13]. IBM, Bain, and Product School confirm this heuristic independently [2, 5, 14]. Does the capability define competitive advantage? Then build. Is the capability interchangeable? Then buy. Not sure? Start with Buy, learn from the data, escalate to Build when needed. Typical Buy scenarios for mid-sized companies: HR, document management, ticketing, CRM assistance. Typical Build scenarios: proprietary quality inspection, process optimization on your own product [13]. The appliedAI whitepaper, developed with participation from BMW, Rohde & Schwarz, and BayWa, puts it plainly: "The field of AI is evolving so fast that hardly any company can or should tackle all topics in-house" [15]. Clustox also recommends phased validation before major investments: First validate the use case with a Minimum Viable Agent, then scale iteratively [12]. Anyone who starts with a USD 500,000 investment upfront will most likely build the wrong thing.

Second: The orchestration layer. Bain identifies agent orchestration as the architectural centerpiece of the years ahead [5]. Product School adds: Control over the orchestration layer is often more critical than ownership of individual agents [14]. 96% of companies plan to expand their AI agent usage within the next 12 months [12]. Emerging standards like MCP (Model Context Protocol) and A2A (Agent-to-Agent) aim to standardize communication between agents but have not yet achieved market dominance [2, 5]. Bain puts it sharply: "The first semantic layer to establish an industry-wide standard will reshape the AI ecosystem" [5]. The orchestration layer is becoming the new lock-in risk: Whoever controls the semantic layer today controls the ecosystem tomorrow [5].

Third: Organization before technology. Starting with a concrete 100-day plan yields better odds of success than any company that begins with the technology choice [13]. The proven sequence: diagnosis in the first 30 days, use case selection and pilot in days 31 to 60, implementation with ROI measurement by day 100 [13]. Quick wins exist for mid-sized companies too: invoice auditing, proposal assembly, support draft responses, error code explanations [13]. The hybrid standard in practice: a foundation model as the base, proprietary company data connected via RAG, standard processes covered through SaaS [13]. As Oliver Breucker puts it: "The right pace is not fast -- it is sustainably operable" [13].

The uncomfortable truth

The SaaS market is under real pressure. Revenue growth dropped from 21% to 12%, hitting negative 2% in Q1 2025 [11]. The iShares Expanded Tech-Software Sector ETF lost over 20% of its value [4]. AI-native applications are growing significantly faster: ChatGPT doubled its user base to 400 million in six months, Midjourney shows 17.4% user growth, and Canva gained 10% in active users [10]. But SaaS is not disappearing. Systems of record remain as the foundation [4, 5]. Bain identifies specific scenarios where SaaS stays resilient: regulated industries with deep domain expertise, areas with deep data integration, and workflows where AI augments the existing tool rather than replacing it [5]. Vulnerable, on the other hand, are vendors of generic, easily replicable solutions without proprietary data advantages [10]. The future is not "no SaaS" -- it is SaaS with radically changed pricing models: away from seat-based, toward outcome-based pricing [5]. The ambitious valuation multiples of the SaaS industry, built on permanently high growth rates, need recalibration [10].

The real danger does not lie in choosing the wrong technology. It lies in treating the decision as a technology question. 68% of failed AI initiatives fail because of people and processes [1]. Not algorithms, not infrastructure, not the choice between Build and Buy. Gartner predicts that 70% of enterprise AI workloads will run on hybrid architectures by 2026 [1]. The path forward does not run through the perfect technology decision. It runs through three questions: Where does the actual competitive advantage lie? Which combination of buying, customizing, and building fits your organizational maturity? And does the company have the strategy, the people, and the processes to actually execute that combination?

References

[1] Zartis Team (2025). "The Build vs. Buy Dilemma in AI: A Strategic Framework for 2025". *Zartis*. https://www.zartis.com/the-build-vs-buy-dilemma-in-ai-a-strategic-framework-for-2025

[2] Brans, Pat (2025). "Your Next Big AI Decision Isn't Build vs. Buy -- It's How to Combine the Two". *CIO.com*. https://www.cio.com/article/4097339/your-next-big-ai-decision-isnt-build-vs-buy-its-how-to-combine-the-two.html

[3] Margus (2025). "AI Coding vs. SaaS -- The Beginning of the End for Large Software Companies?". *SaaS-Welt.de*. https://www.saas-welt.de/artikel/ki-coding-vs-saas-zukunft

[4] Eriksson, Viktor (2026). "AI Could Replace Every Other Business Application". *Computerwoche*. https://www.computerwoche.de/article/4134493/ki-konnte-jede-zweite-business-software-ersetzen.html

[5] Crawford, David; McLaughlin, Chris; Doddapaneni, Purna; Fiore, Greg (2025). "Will Agentic AI Disrupt SaaS?". *Bain & Company Technology Report 2025*. https://www.bain.com/insights/will-agentic-ai-disrupt-saas-technology-report-2025/

[6] Rafalski, Kacper (2025). "Build vs Buy AI: Which Choice Saves You Money in 2025?". *Netguru*. https://www.netguru.com/blog/build-vs-buy-ai

[7] Cognizant (2026). "Cognizant Research Shows Plug-and-Play AI is a Myth". *Cognizant Newsroom*. https://news.cognizant.com/2026-03-10-Cognizant-Research-Shows-Plug-and-Play-AI-is-a-Myth

[8] Haes, Johanna (2025). "AI Adoption in German Mid-Sized Companies -- Dynamic, but Often Without Strategy". *FZI Forschungszentrum Informatik / Hochschule Karlsruhe*. https://www.fzi.de/2025/10/06/ki-einsatz-im-deutschen-mittelstand/

[9] Sarazin, Denise (2026). "Build vs Buy Software -- How AI-enabled Software Development and Vibe Coding are Changing the Game". *AppDirect*. https://www.appdirect.com/blog/build-vs-buy-software-how-ai-enabled-software-development-and-vibe-coding-are-changing-the-game

[10] Glasek, Nicolas (2025). "How AI Is Attacking the Business Models of Software Giants". *DAS INVESTMENT*. https://www.dasinvestment.com/endet-das-goldene-zeitalter-der-software-giganten/

[11] Wolfenstein, Konrad (2026). "Managed AI and the End of SaaS -- Why Companies Are Now Building Their Own Software Again". *Xpert.Digital*. https://xpert.digital/en/the-end-of-saas/

[12] Farooq, Maryam; Aslam, Omer (2026). "Build vs. Buy: Should You Use AI SaaS Tools or Build Custom AI Agents? (2026 Guide)". *Clustox*. https://www.clustox.com/blog/build-vs-buy-ai-tools/

[13] Breucker, Oliver (2026). "AI Strategy for Mid-Sized Companies 2026". *Roover*. https://roover.de/ki-strategie-fuer-den-mittelstand-2026/

[14] Gonzalez de Villaumbrosia, Carlos (2026). "Build vs Buy: Making Smarter Software Decisions in 2026". *Product School*. https://productschool.com/blog/leadership/build-vs-buy

[15] Liebl, Andreas; Hartmann, Philipp; Schamberger, Maria (2021, updated 2026). "Enterprise Guide for Make-or-Buy Decisions". *appliedAI Initiative / UnternehmerTUM*. https://www.appliedai.de/en/insights/make-or-buy-decisions/

[16] Bosankic, Leopold (2024, updated 2026). "Make or Buy Decision: When to Build Software In-House and When to Buy?". *Researchly*. https://www.researchly.at/post/make-or-buy-entscheidung-software