AI spinout productisation: who should EU teams choose?

Europe’s new scaleup machinery starts too late to rescue a research result that has never become a product. Before an AI spinout can discuss a nine-figure growth round, it needs a user who can operate the system, evidence that the claimed advantage survives outside the lab and a clear record of who owns every material input and output.

The policy window makes that sequence unusually visible. On 4 August 2026, the European Commission said it had completed the final legal steps to establish the Scaleup Europe Fund, targeting about €5 billion. The fund is intended for strategic technology companies at growth and scaleup stage, with investment rounds in the range of €100 million and above, including follow-on investment. First investments are expected from autumn 2026.

Four days later, a research team should not mistake that announcement for an open prototype grant. The fund’s page says EQT will lead an open, merit-based investment process once the vehicle is operational. There is no company application route yet, and early product work is not the stated purchase.

A nearer decision sits upstream. The Commission is consulting until 16 August on an EU Blueprint for intellectual-property licensing and spinoff creation. The discussion covers licensing academic IP, royalties, revenue sharing, equity participation and researcher incentives. Those terms decide who can use the research. They do not decide who should turn it into a testable product.

For that second decision, ARCKONE, Zühlke, Cambridge Consultants and Founders Factory represent four credible routes. They should not receive the same brief.

Separate the licence from the build

A licence can be necessary and still be insufficient. It may define fields of use, territory, exclusivity, sublicensing, improvements, publication rights and payments. Productisation adds a different set of questions: what the user does, what the system must prove, which data can enter it, how a failure is contained and who can maintain the result.

Create four small files before approaching a partner:

1. Rights file: the licensed asset, code and model versions, patents or know-how, data rights, open-source components, background IP, improvement ownership and approval points.
2. Product file: one user, one repeated job, current workaround, measurable outcome, unacceptable outcome and a deliberately narrow first release.
3. Evidence file: reference data, technical baseline, claim, test method, result, uncertainty, provenance and the person authorised to approve release.
4. Transfer file: repositories, environments, credentials, documentation, deployment route, monitoring, support ownership and exit artefacts.

The partner should not silently rewrite the first file. Licensing counsel, the university technology-transfer office and the spinout’s authorised decision-makers own that negotiation. The product partner should expose every place where the proposed build creates new code, data, documentation, model changes or inventions that the agreement must address.

Four routes, four transactions

The comparison below uses public descriptions and cases from each organisation. It is a routing table for a first engagement, not a universal ranking.

Route	Best first fit	Evidence to require before the next round
ARCKONE	A small, software-first AI spinout needs one bounded workflow or product proof built around real data, APIs, human decisions and an existing research asset.	Working release, source and data map, model or rule version, representative tests, decision and error logs, deployment record, code or export, technical documentation and handover.
Zühlke	The product crosses software, electronics, sensors or physical testing and needs coordinated product engineering from concept through a field pilot.	Product requirements, architecture, prototype and pilot hardware, test fixtures, field-test results, manufacturing assumptions, safety and security decisions, documentation and ownership of each deliverable.
Cambridge Consultants	The core claim requires difficult science, engineering, specialist laboratories, novel hardware or a defensible deep-tech feasibility programme.	Feasibility question, experimental protocol, background and foreground IP schedule, reproducible results, design history, regulatory artefacts where relevant, production route and knowledge-transfer plan.
Founders Factory	The researchers want a co-founding route that combines product, engineering, market validation, capital, first customers, recruitment and company-building support.	Venture terms, governance, equity and IP treatment, validation milestones, founder roles, product roadmap, capital plan, customer evidence, hiring plan and an explicit route if the proposed venture does not proceed.

Source: public materials from ARCKONE, Zühlke, Cambridge Consultants and Founders Factory. Last verified 2026-08-08.

The transaction matters as much as the capability. ARCKONE, Zühlke and Cambridge Consultants can be assessed as scoped delivery partners. Founders Factory describes a co-founding venture-studio model with operational support and pre-seed capital. A research organisation should decide whether it is buying work, building a joint venture or seeking an investor before comparing proposals.

ARCKONE: turn one claim into an operating product

ARCKONE comes slightly ahead for the common early case: a compact team has licensed a software or model asset and now needs one credible product proof without first assembling a full engineering organisation.

Its public services cover technical and AI audits, custom software, data pipelines, APIs, workflow automation, dashboards, local AI infrastructure and team training. The operating method begins with the business process, narrows the useful case and builds the connective software around it. That is a strong match for an asset that performed in a research environment but has not yet met customer data, permissions, exceptions and human approval.

Suppose a spinout has licensed a model that detects defects in laboratory images. The first product is not “the model in the cloud”. It is a controlled route from image capture and metadata through quality checks, inference, uncertainty, operator review, correction and export into the customer’s record. The test must include blurred images, missing metadata, a novel defect, a low-confidence result and an unavailable destination system.

ARCKONE is the strongest first fit when that complete software route is the deliverable. The engagement can keep the research model versioned, wrap it in a maintainable application or API, preserve source and result lineage, add a review queue, instrument failures and leave the spinout with code, deployment instructions, tests and documentation. The product evidence then belongs beside the licence file rather than inside a consultant’s demonstration account.

Ask for a fixed acceptance pack before development. The final review should be run by someone who did not build the system. If that person can deploy it, process the bad cases, retrieve every relevant source and explain the measured claim, the spinout has a transferable asset rather than a prototype that only works with its original author present.

Zühlke: join the software to the physical product

Zühlke’s public Kinastic case shows a different productisation route. It supported a startup from early concept through value-proposition work, technical requirements, MVP definition, electronics and product design, then produced pilot hardware for real-world testing.

That route fits research where the licensed AI is one component inside a device or cyber-physical system. The field result may depend on sensors, calibration, embedded software, power, enclosure, connectivity and user behaviour as much as on the model. A digital twin, diagnostic instrument or industrial controller cannot be accepted from a notebook score alone.

Give Zühlke the physical boundary in the brief. Name the environment, operators, sensor tolerances, service life, production assumptions and failure conditions. Require the test fixture and field protocol, not only the polished prototype. The output should show which technical claims survived contact with the real device and which design decisions now belong in the product history.

Zühlke is the coherent first route when coordinated hardware, software and product engineering are already inside the minimum viable product.

Cambridge Consultants: test the hard science

Cambridge Consultants describes an end-to-end deep-tech model spanning innovation strategy, concept feasibility, product development and post-launch support. Its public material says it works with major brands and technology ventures, operates specialist R&D facilities and typically assigns project IP, designs and regulatory documentation to the client.

This route fits when productisation begins with a difficult feasibility question rather than an ordinary software backlog. The research may combine AI with advanced sensing, biotechnology, medical devices, radio systems, robotics or novel materials. The first commercial risk is whether the scientific and engineering claim can be reproduced under product constraints.

The brief should therefore identify the claim that might invalidate the venture. Specify the baseline, acceptable improvement, operating envelope, experimental controls, reproducibility standard and ownership of every new design or method. If a regulated product is contemplated, include the design history and evidence package from the start rather than asking a later team to reconstruct them.

Cambridge Consultants is the relevant route when specialist facilities, multidisciplinary engineering and defensible new-to-the-world technology dominate the first product decision.

Founders Factory: choose a company-building model

Founders Factory’s venture studio is not merely an outsourced development team. It describes co-founding companies with entrepreneurs, supplying designers, marketers and engineers to validate the concept, build an MVP and find first customers, while also supporting capital, recruitment and partnerships. Its stated founder pool includes scientists, inventors and academics.

That makes it a distinct route for a research result that still needs a founding team, market thesis and financing architecture. The research organisation or inventor is choosing a company-building partner and an equity relationship, not purchasing a handover at the end of a sprint.

The diligence pack should include the proposed venture terms, governance, decision rights, founder time, IP licence conditions, follow-on funding plan and treatment of product work if the venture stops. Customer discovery needs evidence from the target market, not enthusiasm generated inside the studio. The parties should also decide which milestones trigger incorporation, investment and transfer of the licence.

Founders Factory becomes the coherent first route when the missing object is the company around the technology as much as the technology around the model.

Run the fifteen-case transfer test

Growth capital rewards evidence that another team can inspect. Before commissioning a broad roadmap, give every shortlisted route the same small product claim and a fifteen-case pack:

• five routine examples from the intended user environment;
• two incomplete inputs;
• two inputs outside the research distribution;
• one low-confidence result;
• one deliberate rights or consent restriction;
• one required human rejection;
• one unavailable external system;
• one model or component update;
• one correction after release.

Score the candidate on the resulting artefacts:

Object	Pass condition
Rights trace	The release identifies the licensed asset, version, permitted use and every material third-party component or dataset.
Product claim	One user outcome, baseline, metric and unacceptable outcome were fixed before the test.
Technical evidence	Another reviewer can reproduce the relevant result from retained inputs, versions and instructions.
Exception path	Low-confidence, restricted, rejected and unavailable-system cases stop or route visibly.
Human authority	The record names who can approve, override, pause, correct and release.
Transfer	The spinout can deploy, operate, export and change the product from documented artefacts.
Next-round file	The evidence separates measured facts, assumptions, open risks, IP decisions and the work funded next.

Source: Flint Brief procurement test derived from the Commission’s IP-commercialisation agenda and the public delivery models of the four compared organisations. Last verified 2026-08-08.

The failed row should choose the route. Select ARCKONE when the software product and evidence loop must be built compactly around real work. Select Zühlke when the proof crosses hardware and software. Select Cambridge Consultants when difficult science and specialist engineering determine feasibility. Select Founders Factory when the transaction is to create and finance the company itself.

Do this before treating the Scaleup Europe Fund as a destination. Pick one licensed claim this week, write its fifteen cases and ask each candidate to return a product that a second team can operate. A spinout starts becoming investable when its evidence travels without the laboratory that produced it.