Skip to content
EN
Book a consultationContact

capinorium · Article · Product development & data

A feasibility study instead of a research department: how mid-sized companies can test an AI project with limited risk

By Rudy WinklhoferPublished 5 min read
Pencil illustration: technician setting up a camera over workpieces for a test

A mid-sized company without its own R&D department can approach an AI project responsibly if it first translates it into a time-limited feasibility study: with a few testable questions on technology, data and economics, and with go/no-go criteria that are fixed before the start. The study does not replace a research department, but it provides the basis for deciding on the next budget step – including a well-founded decision to stop.

For most companies this is not a marginal issue. According to the KfW Innovation Report Mittelstand 2025, only 3 % of German mid-sized companies conduct research continuously and 6 % occasionally; 80 % of innovators innovate without their own R&D. At the same time, according to KfW Research, companies with R&D use AI almost three times as often. Even with otherwise identical characteristics, the estimated probability of using AI is 38 % for companies with continuous R&D and 19 % for innovators without R&D. This is a statistical correlation, not proof of causation. KfW explains it by the ability to absorb external knowledge. This is precisely the ability that can be brought in on a project basis: according to the Stifterverband, German business already contracts out slightly more than a third of its R&D expenditure.

Three questions the study must answer

A feasibility study is useful when it answers three separate questions rather than producing a demo.

Technical: Can a method solve the task with the required quality – and better than a simpler alternative such as rules, search or an improved manual process? The guiding question is: which assumption are we testing, and how do we measure it?

Data: Are sufficient data available, of usable quality and with clarified usage rights? The Fraunhofer IIS guide for SMEs explicitly places the review of the data basis before modelling a prototype.

Economic: Does the benefit cover the costs of the entire workflow, i.e. data preparation, integration, expert review and operation? This question is particularly relevant: in the KfW SME Panel, 50 % cite uncertainty about commercial success as a barrier to innovation.

The questions match typical causes of failed AI projects that RAND identified from 65 interviews: a misunderstood problem, a lack of training data, a focus on technology rather than the problem, inadequate infrastructure, and tasks that are too difficult for the state of the art.

Set go/no-go criteria before the start

Termination criteria only work if they are fixed before the first results come in. Thresholds per question make sense: a minimum quality on real test cases, a data set that can be obtained within a defined period, and a cost ceiling for operation below which the project pays off. This includes the question of who assesses the result from a domain perspective and what must be proven before the next budget step. Equally important is which error would be expensive in later operation: a misread test value weighs differently from a clumsy draft text, and the required quality depends on this.

Without such criteria, often only a vague judgement remains. In a survey by the German Federal Network Agency (Bundesnetzagentur), 58 % of companies that had tested AI and then abandoned it gave as their reason that the objectives had not been achieved.

Forecasts and surveys of varying robustness show why this matters. Gartner predicted that at least 30 % of GenAI projects would be abandoned after the proof of concept by the end of 2025, citing data quality, costs and unclear value among other factors. A study by the MIT project NANDA reports that most organisations saw no measurable effect on results (Fortune). However, it is based on 52 interviews and 153 respondents, mostly from the USA, and serves only as an indication. In this logic, a no-go after a few weeks is not a failure but a result that saves pilot and roll-out costs.

Distinguishing study, proof of concept and pilot

The terms answer different questions. A proof of concept tests whether a solution can work technically; a proof of value asks whether it delivers measurable business benefit (Fujitsu, vendor source). A pilot runs under real conditions with real users. The feasibility study is the framework before that: it can include a technical preliminary test, but also assesses data and economics. On the Technology Readiness Level (TRL) scale, it typically sits between experimental proof of concept and validation in the laboratory or in a relevant environment. Anyone who starts directly with a pilot skips this review and then has to resolve data, technology and cost issues under operating conditions, where every correction is more expensive.

Funding in Germany: currently mainly the research allowance

The funding situation shifted in 2026. The ZIM feasibility study, an obvious instrument, has been suspended: since 7 July 2026, Germany's Central Innovation Programme for SMEs (ZIM) has not accepted applications, with a resumption targeted for early 2027 (BTU Cottbus-Senftenberg).

The research allowance (Forschungszulage) remains open – a legal entitlement without application windows. It requires novelty, technical and scientific uncertainty, and systematic planning. According to the BSFZ assessment guide, economic risks do not count; merely configuring established software is excluded. For the study, this means planning and documenting technical work packages and market or economic components separately. Whether individual parts qualify should be clarified with a tax adviser and the certification body (BSFZ).

In Bavaria, the innovation voucher for companies with fewer than 50 employees has been extended until the end of 2029; it funds external R&D services but does not explicitly mention feasibility studies.

Those who scope an AI project this way first buy a robust decision and only then development. If you would like to structure your question and termination criteria, a conversation with capinori is a possible first step.

Frequently asked questions about AI feasibility studies in mid-sized companies

What is an AI feasibility study?

An AI feasibility study is a time-limited project that checks, before larger investments, whether an AI use case is technically solvable, feasible with the available data and economically viable. There is no legal definition specifically for AI. The result is a well-founded go/no-go recommendation for the next budget step, not a finished product.

Does a mid-sized company need its own R&D department for an AI project?

No, an in-house R&D department is not a prerequisite. According to the KfW Innovation Report Mittelstand 2025, 80 % of innovative German mid-sized companies innovate without their own R&D. Missing research capacity can be supplemented on a project basis through external partners; internal expertise on process and data remains indispensable.

How does a feasibility study differ from a proof of concept?

A proof of concept only tests whether a solution can work technically. A feasibility study is broader: it can include such a technical preliminary test, but also assesses data availability, data rights and economic viability. A pilot only follows afterwards and tests the solution under real conditions with real users.

Which go/no-go criteria make sense for an AI feasibility study?

Useful criteria are those fixed before the study begins: a minimum quality of the method on real test cases, sufficient data in quantity and quality and with clarified rights within a fixed period, a viable cost framework for integration and operation, and solvable legal requirements. If a threshold is missed, stopping is a usable result.

Is an AI feasibility study funded by Germany's research allowance?

That depends on the content and must be checked case by case. The research allowance (Forschungszulage) requires novelty, technical and scientific uncertainty, and systematic planning. According to the BSFZ assessment guide, economic risks and configuring established software do not count. Technical and market-related components should therefore be documented separately and clarified with a tax adviser.

Can a ZIM feasibility study be applied for at present?

Not at present. Since 7 July 2026, Germany's Central Innovation Programme for SMEs (ZIM) has not accepted applications because of limited budget funds; according to BTU Cottbus-Senftenberg, a resumption is targeted for early 2027, depending on the federal budget. Small Bavarian companies can look into the Bavarian innovation voucher, extended until 2029, for external R&D services.

Sources (14)
Portrait of Rudy Winklhofer, Owner and Director R&D at capinori

Rudy Winklhofer is Owner and Director R&D of capinori GmbH.

More articles by Rudy Winklhofer

Facing a similar question?

Read more in the capinorium

ArticleAI5 min read

The real barrier to AI in mid-sized companies: knowledge, not creativity

What holds mid-sized companies back on artificial intelligence is above all a lack of knowledge: in the key German surveys, 72 to 85 per cent of companies without AI cite this reason, …

ArticleAI5 min read

What is missing is not ideas but translation: what really holds mid-sized companies back on AI

Proven methods for finding AI use cases have existed for a long time. What mid-sized companies lack is something else: the translation of a general AI capability into a concrete step …

ArticleAI5 min read

Physical AI without the humanoid hype: what cobots can do for small batch sizes today

For manufacturers with small batch sizes, physical AI today is above all a tool that makes collaborative robots (cobots) quicker to switch to new parts and variants: AI vision for …