Can Your Swiss AI Project Qualify for Innovation Funding? A Seven-Gate Fit Test

Posted on Oct 23 • Originally published at iapmesuisse.ch

Short answer: an AI project may fit an Innosuisse instrument when the company is testing a genuinely uncertain, innovative concept with a Swiss research partner—not merely buying a standard tool, connecting an API, or funding ordinary rollout. For an early feasibility question, an Innovation Cheque may be the relevant route; for a larger science-based development project, a joint innovation project may fit better. Neither route is automatic, and neither should be budgeted as money already awarded.

This guide is a decision aid, not a legal opinion or a promise of funding. The official conditions and application platform control. The internal Search Console snapshot for 14–21 October 2026 showed the query “aide startup suisse” reaching the source page at position 29.1 with eight impressions, while “ki beratung kmu” reached another IAPME page at position 16 with 43 impressions.

Those are small, query-specific observations from one site and a delayed reporting window—not market size, demand forecasts, or evidence that readers will qualify. Keyword research can reveal the language people use, but it cannot prove grant fit or purchasing power; even commercial SEO guidance treats query discovery and commercial intent as separate analysis steps ( Ahrefs’ keyword-research guide ). They point to a practical question: before paying someone to write an application, how can a Swiss SME distinguish an innovation study from ordinary software procurement?

The existing Swiss innovation-aid overview for SMEs and start-ups is a landscape page. This article adds a reusable fit test: seven gates, a route comparison, a one-page research brief, and a small local script. The objective is to help a founder decide whether to stop, clarify, find a research partner, or investigate an application. It cannot predict an expert decision.

An Innovation Cheque is a voucher for a preliminary study of an innovative idea. Innosuisse says the study can examine feasibility, develop a concept, or analyse innovation and market potential with a Swiss research partner. The agency covers the research partner’s costs up to CHF 15,000. That ceiling is not a cash grant paid to the SME for a software subscription, a general consulting retainer, or a standard deployment. The official description, required application information, and current eligibility conditions are on Innosuisse’s Innovation Cheque page.

The current eligibility screen includes an organisation with fewer than 250 full-time equivalents, a Swiss base and UID number, an already identified Swiss research partner, and no Innovation Cheque awarded to the organisation in the previous two years, measured by decision date. Group-company FTEs count. Passing these formal checks only permits consideration; it does not establish that the concept is novel, worth studying, or likely to be approved.

A larger innovation project with an implementation partner is a different route. It is for a science-based innovation project jointly conducted by an implementation partner and a research partner. Innosuisse’s current conditions include Swiss added value, a willing Swiss research partner, a company contribution amounting to 40–60% of total project costs, and a cash contribution to the research partner of at least 5% of total project costs. The proposal also needs substantial preliminary work on innovation, user need, market potential, and existing solutions. See the agency’s implementation-partner project requirements and national-project FAQ.

These are not two sizes of the same cheque. One pays a bounded preliminary study; the other co-funds a substantive collaborative development project with a material company contribution. An SME that has a normal product purchase should not relabel it as research just because the product uses machine learning.

The distinction also protects against over-reading a global AI trend. Stanford HAI’s AI Index collects evidence about technical progress, economic influence, and societal impact; it is useful context for a state-of-the-art review, but global measures cannot establish that a particular Swiss firm has an innovative question. The applicant must compare the specific problem and proposed method with current science and available solutions, not cite “AI is moving quickly” as novelty.

The U.S. Small Business Administration’s AI guidance for small businesses recommends testing value and considering risks, but it is a general operational resource from another jurisdiction, not a source for Swiss funding eligibility or Swiss law. Its narrow transferable lesson is to test a tool against a business need before adopting it; programme rules must still come from Innosuisse.

Use the gates in order. A “no” at an early gate is useful: it can save application-writing time. “Unknown” means gather evidence before paying for a dossier. Do not turn the test into a score that claims an approval probability.

This is a screening protocol built from official criteria; the green evidence is a recommended preparation standard, not an Innosuisse checklist or formal scoring rubric. The legal basis describes preliminary studies as work that helps determine the effective feasibility of company innovation projects and lists innovation content, benefit, sustainability contribution, and cost-benefit ratio among its assessment criteria. Read the Innosuisse Funding Ordinance on Fedlex alongside the current programme page; when their wording or a secondary guide differs, confirm the current interpretation with Innosuisse.

The decision path is intentionally asymmetric: it has several early exits because “not a fit” is a useful result.

flowchart TD A[Name Swiss applicant and UID] --> B{Formal limits and prior-cheque rule verified?} B -->|No or unknown| C[Stop and verify facts] B -->|Yes| D{Only standard purchase or rollout?} D -->|Yes| E[Procurement or ordinary business budget] D -->|No| F{Unresolved innovation or feasibility question?} F -->|No or unclear| G[Clarify need and compare existing solutions] F -->|Yes| H{Bounded preliminary study with Swiss research partner?} H -->|Yes| I[Discuss Innovation Cheque fit] H -->|No; substantive collaborative development| J[Assess joint innovation project and cost share] I --> K[Confirm official rules; approval is not guaranteed] J --> K Gate 1: name the applicant, not just the founder Write down the legal entity that would submit the application, its Swiss registered base, UID, ownership or group relationship, and the source of the FTE count. A brand name, founder’s residence, or Swiss customer does not by itself answer who the implementation partner is. If a parent company, subsidiary, or sister company is involved, map the group before using an employee threshold.

The official Innovation Cheque page says the group’s total FTEs are decisive, so a small local subsidiary should not be assessed in isolation.

Check the prior-cheque condition against the organisation’s decision records, not only against the founder’s memory. The rule refers to whether the organisation has been granted a voucher in the previous two years, measured from the decision date. If a company has restructured, merged, changed legal form, or received support through a related entity, do not guess whether the rule follows the current entity. Ask the agency or a qualified adviser how it treats that fact pattern. This guide cannot interpret corporate succession for a specific applicant.

Make a one-page applicant record before writing the technical idea. Include the company name and UID, Swiss value-creation location, group map, FTE source and date, prior-cheque decision dates, and one person authorised to submit. These facts are boring but useful: they expose formal issues before the team spends weeks describing a model architecture. If any item is unknown, mark it “unverified” rather than silently treating it as eligible.

“We want to use generative AI for customer replies” is a use case, not yet an innovation question. If an existing product can draft responses and staff can configure templates, the remaining work may be procurement, data governance, integration, and change management. Those can be valuable business projects, but the existence of an AI component does not make them research.

A more credible preliminary-study question might be: “Can a method that combines low-resolution inspection images with our maintenance history identify a defect class early enough to change repair scheduling, while keeping false negatives below a pre-agreed threshold?” The firm still has to prove that the question is genuinely novel relative to available solutions and worth studying. It may turn out that an off-the-shelf vision model solves it. That is a valid reason not to apply.

Another example: “Can a retrieval system answer questions from our internal policies?” is often a standard integration task. “Can a constrained retrieval-and-verification method reliably detect conflicting versions of safety procedures across three languages, on a representative corpus and with an auditable source trace?” is more testable. Even then, novelty is not established by wording. The team must compare the proposed method with current research and commercial offerings, state what is unknown, and explain why the result could create value.

The practical rule is to write the uncertainty before naming the vendor. A vendor’s proposal can help estimate scope, but it is not independent proof of novelty, market need, or scientific feasibility. A research partner should be able to challenge the premise and propose a method that could produce a negative result.

Choose the Innovation Cheque when the immediate decision is whether a bounded preliminary study can establish feasibility or innovation and market potential. Think “answer this question before we commit to a larger project.” The company must already identify a Swiss research partner, and the study must fit the current programme’s scope and time limit. Innosuisse says the decision usually takes four to six weeks and the preliminary study must be completed within six months after approval. Do not schedule work to begin before approval and the partner agreement are in place.

Investigate an innovation project with an implementation partner when the concept has advanced beyond a preliminary question and the company and research institution need to develop a product, service, or process together. This is not a shortcut for a company to outsource a normal build. The agency expects a science-based project and asks for preliminary clarification of the innovation content and value creation. The company also needs to plan for its own project costs and the research-partner cash contribution. Project decisions usually take six to eight weeks after initial submission, and the application should be submitted at least six weeks before the relevant Innovation Council meeting, subject to the current schedule.

Research projects without an implementation partner are aimed at researchers who have a promising idea but have not yet found a market implementation partner. That is not the same as an SME applying alone for a business rollout. Start-up instruments and other programmes have their own entry conditions; do not infer them from the Innovation Cheque page. The Innosuisse national funding overview is the safer starting point for choosing a route.

If the only desired outcome is to buy a subscription, connect a standard CRM, or train staff, stop and compare procurement, internal funding, cantonal support, and commercial financing instead. No grant route should be selected just because it has a large headline amount.

Use a sentence with four parts: “We do not know whether [method or concept] can achieve [measurable task] on [defined data or users] under [operational constraint], compared with [current method or baseline].” A good question can fail. “Prove that our idea works” is not a research plan; it asks the partner to confirm a conclusion already chosen.

Specify deliverables the company can use to decide: a documented baseline, a reproducible experiment, a limitations report, a prototype or technical note where appropriate, and a recommendation to stop, redesign, or proceed. The deliverable should not require a positive result to be useful. For example, a feasibility report that shows the data do not support the desired prediction can prevent a larger investment. That is a business benefit even if it is not a product launch.

Set a stop condition before testing. Examples include insufficient data quality, no defensible novelty over existing products, unacceptable false-negative behavior, or a cost-to-benefit ratio that does not justify follow-up. This is an internal governance recommendation, not a stated Innosuisse pass/fail metric. It makes the research partner’s independence more credible and protects the company from turning a preliminary study into an open-ended sales exercise.

The research partner is not a decorative university logo. The partner should have the methods and expertise to investigate the uncertainty and should be willing to conduct the study within the programme’s rules. Innosuisse publishes information about eligible research institutions and the application process. The company should contact more than one suitable group if the expertise is specialised, explain the question neutrally, and ask what evidence would make the study inconclusive.

Before submission, agree the scope, roles, work plan, personnel and material-cost assumptions, project contact, expected deliverables, access to company data, confidentiality arrangements, publication expectations, and how results can be used. Innosuisse states that for an Innovation Cheque the implementation partner formalises the legal relationship with the research partner; the agency does not provide a contract template. For larger collaborative projects, intellectual-property and usage rights can become especially important. Innosuisse recommends addressing these issues early; the FAQ explains that parties must arrange their legal relationship rather than rely on the agency to draft their private contract.

Do not wait until the application is approved to discover that the research group cannot use the required data or that the company expects exclusive rights the institution cannot grant. Ask for the institution’s technology-transfer or contracts office early. If you cannot agree on data access, publication, or use of the result, the project may be commercially unworkable even if the technical question is interesting.

The company remains responsible for its own decisions. A research partner’s participation does not guarantee approval, product-market fit, legal compliance, or future investment. Keep the relationship independent enough that the study can recommend stopping. A partner who is also selling the implementation may have a commercial interest; disclose and manage that conflict rather than treating the relationship as neutral.

Innosuisse asks for preliminary clarifications such as literature research, needs assessment or proof of potential demand, and analysis of existing solutions and providers. This is more than a list of enthusiastic interview quotes. Identify who experiences the problem, how they handle it today, what it costs or delays, what alternatives they already use, and what would make them change.

For AI specifically, compare the proposed method against current products and non-AI process changes. Search relevant research and product documentation. If a simple form, rule, queue change, or better data source solves the problem, include it as an alternative. A proposal that ignores a cheaper baseline may appear to be looking for a technology to justify rather than testing a need.

The Swiss federal SME portal reported in July 2026 that an EY survey of 604 employees found a mixture of strategic deployment, pilots, and companies without concrete initiatives. The survey is useful context for the breadth of experimentation, but the sample and question wording are not a proxy for your industry’s demand or your application’s novelty. Swiss SME Portal, “AI is becoming widespread in Swiss companies”. Treat it as context, not a business case.

For an Innovation Cheque, do not write “CHF 15,000 grant income” in the company forecast. The official instrument finances the research partner’s preliminary-study costs up to that ceiling. The SME still needs to plan internal time, data preparation, legal review, management attention, and whatever implementation might follow. Some of these may be outside the funded study; confirm eligible cost treatment against the current rules and partner budget rather than guessing.

For a joint innovation project, translate the official cost-sharing conditions into cash before signing. As a simple arithmetic illustration, on a hypothetical total project budget of CHF 200,000, 40–60% corresponds to CHF 80,000–120,000 of company-incurred project costs, and the minimum cash contribution to the research partner at 5% would be CHF 10,000. This is only arithmetic using the published percentages, not a forecast, award estimate, eligible-cost decision, or funding commitment. The FAQ and application budget determine the actual treatment of costs, overhead, VAT, and contributions.

Add a second scenario in which no award arrives. If the company cannot carry its own share, cannot wait for the decision, or cannot fund implementation after the preliminary study, the proposed route may be premature. A positive feasibility finding does not buy software, integration, cybersecurity work, staff time, or a commercial launch automatically.

Calendar the actual programme milestones from the official pages: application submission, evaluation, decision, agreement, earliest start, study duration, reports, and any next-step application. The website currently describes four to six weeks for an Innovation Cheque decision and a six-month study. For a standard implementation-partner project, it describes a six-to-eight-week decision and a meeting schedule. These are stated usual periods, not service-level guarantees. Recheck dates immediately before submitting; published meeting schedules and rules can change.

Imagine a Swiss equipment-repair SME receiving maintenance requests in German, French, and Italian. The owner proposes to buy an AI assistant that classifies requests and drafts replies. The initial question is not yet grant-ready: a commercial product may already do it, and the project has not identified an unknown that calls for research.

The team maps its current process, samples a defined set of redacted historical requests, and notes which fields are missing, how often requests are routed to the wrong technician, and how much human review takes place. It discovers a narrower unresolved problem: messages often combine multiple languages, model-generated classifications may hide uncertainty, and the available commercial tools do not expose evidence for the recommended queue. This is a hypothesis to investigate, not proof that the company has invented a new method.

The SME then asks a Swiss research group whether a bounded preliminary study could compare a transparent multilingual classification approach with the current process and existing products. The study would define the sample, labels, languages, critical routing errors, data minimisation, and a stop rule. The deliverable is a comparison and feasibility report, not a promise to build a production system. If the research group finds that an off-the-shelf tool is adequate, the company can stop and buy or configure that tool from its ordinary budget. If the study identifies a genuinely novel technical path and material value, the company can examine a larger innovation project with the partner.

The example does not claim that this particular project qualifies. It shows how the seven gates change the conversation: from “which AI product should we buy?” to “what decision remains unknown, what evidence would answer it, and is a research instrument appropriate?” If the team cannot state the unknown, the smallest useful next step is process discovery—not an application.

Before contacting a research group, prepare this brief. Keep it to one page at first; add evidence only where it changes the decision.

The brief is an original working template assembled from the programme’s stated application fields and evaluation criteria. It is not an official form and does not replace the online application. The goal is to expose missing evidence early enough to improve the idea or abandon it cheaply.

This Python example is an original, deterministic checklist helper. It asks for facts and returns a next action. It never estimates award probability, makes an eligibility determination, or sends information over a network. It deliberately treats an unknown answer as a reason to verify, not as a “yes.” Run it locally with Python 3.10 or later; no third-party package is required.

#!/usr/bin/env python3 """ Screen an idea for a discussion with Innosuisse; not an eligibility decision. """ from dataclasses import dataclass def ask ( prompt ): while True: value = input ( prompt + " [y/n/u]: " ). strip (). lower () if value in { " y ", " n ", " u " }: return value print ( " Enter y (yes), n (no), or u (unknown). " ) @dataclass class Answers: swiss_entity: str under_250_group_fte: str uid: str partner_identified: str cheque_in_last_2_years: str genuine_uncertainty: str ordinary_purchase_only: str can_fund_company_share: str def screen ( a ): formal = [ a. swiss_entity, a. under_250_group_fte, a. uid, a. partner_identified ] if " n " in formal or a. cheque_in_last_2_years == " y ": return " STOP: verify the current formal conditions before drafting.

" if " u " in formal or a. cheque_in_last_2_years == " u ": return " HOLD: verify applicant, group FTE, UID, partner, and prior decision. " if a. ordinary_purchase_only == " y " or a. genuine_uncertainty == " n ": return " LIKELY PROCUREMENT: compare products/process changes; do not assume a grant fits. " if a. genuine_uncertainty == " u ": return " CLARIFY: write the falsifiable research question and compare existing solutions.

" if a. can_fund_company_share == " n ": return " HOLD: model the company contribution and no-award scenario first. " return " DISCUSS: ask the research partner/Innosuisse whether a preliminary study fits. " if __name__ == " __main__ ": labels = [ " Swiss applicant ", " Under 250 FTE in the group ", " Swiss UID ", " Swiss research partner already identified ", " Cheque awarded in the past two years ", " Genuine unresolved question ", " Only an ordinary product purchase/rollout ", " Can fund the company share ", ] values = [ ask ( label ) for label in labels ] print ( screen ( Answers ( * values ))) Enter fullscreen mode Exit fullscreen mode The output “DISCUSS” means only that the answers justify a conversation.

The script does not check the official register, group structure, novelty, eligible research institution, cost treatment, or whether a reviewer will approve the application. A false “yes” can still mislead; verify the inputs against current documents.

Innosuisse’s 2023 summary of an impact study says 77% of innovation ideas were developed further after an Innovation Cheque, and 26% later led to a standard innovation project. The underlying survey covered organisations awarded cheques between 2019 and mid-2022: 537 of 1,540 organisations responded. That is about a 35% response rate by arithmetic. It is encouraging evidence that many recipient respondents continued their ideas, but it is not an approval rate, a causal estimate of economic return, or a forecast for an applicant today. The population already consists of funded recipients, and the respondents may differ from non-respondents. Innosuisse’s impact-monitoring summary provides the figures and population description.

The same summary reports that 8% said they would have pursued the idea in unchanged form without the cheque, while many said they would not have pursued it or would have reduced it. Those answers are retrospective self-reports. They can inform a discussion about the role of early-stage support, but they cannot prove that a specific proposal is socially beneficial or commercially viable. The right use is modest: a preliminary study can be valuable even if the result is a well-supported stop decision, and a large follow-on rate among past recipients should not be marketed as a promise.

There is also a measurement conflict in broad AI adoption statistics. A 2025 AXA survey reported by the Swiss SME Portal, an EY employee survey in 2026, and KOF’s earlier firm-level innovation study differ in respondent populations, dates, wording, and definitions of “use,” “integration,” or “strategic deployment.” They should not be averaged into a single adoption rate. For a funding decision, local evidence from the SME’s own process is more relevant than a national headline percentage.

The article’s original asset is the seven-gate fit test, the one-page research brief, and the local screening script. Method: map the current official programme pages and legal criteria as accessed on 23 October 2026; compare them with the selected Swiss SME aid overview and the Search Console query/page observations; then turn the distinctions into a checklist and deterministic helper. No new survey, experiment, interviews, or measured business outcomes were conducted.

The 200,000 CHF budget example is arithmetic on an explicitly hypothetical budget using official contribution percentages, not observed data. Exclusions: no canton-specific support, tax instrument, European programme, individual eligibility advice, or legal opinion is assessed. A relevant Innosuisse explainer video was mentioned on the programme page, but a searchable transcript or independently accessible evidence was not available in the research results; no claim here depends on video content.

Search results for forum/video queries did not surface a relevant, verifiable first-person application discussion, so the concrete reader question is drawn from the site’s Search Console queries rather than treated as representative public sentiment.

Calling an ordinary purchase “innovation.” A standard SaaS licence, API connection, model prompt, or staff training may be useful but does not become research because the supplier calls it AI. Compare existing solutions and state the unknown.

Treating CHF 15,000 as cash to the applicant. The cheque finances a preliminary study with a Swiss research partner. Do not book it as available company revenue or assume it pays implementation costs.

Selecting a partner after writing a fixed answer. If the company has already decided what the research must conclude, the study loses credibility. Let the partner test the premise, including the possibility that the idea should stop.

Ignoring group FTEs or prior support. The formal screen can fail even when the operating company is small. Verify the group scope and decision dates rather than using an informal headcount.

Assuming a positive study means a grant or launch. The preliminary study, a larger project application, a product build, and market adoption are separate decisions with separate budgets and risks.

Skipping data, IP, or privacy review. A funded study does not itself authorise use of personal or confidential data, settle ownership, or remove sector-specific obligations. Get appropriate professional advice for the actual data and contract.

Using survey statistics as an ROI model. Recipient surveys and national adoption studies describe their samples; they do not establish causal returns for one SME. Build the business case from the company’s baseline, planned test, and costs.

This article is an evidence-based screening aid, not a substitute for the current Innosuisse application materials, the relevant legal texts, a research institution’s contracts office, or advice tailored to your organisation. Funding rules and application calendars can change; re-check the official pages immediately before acting.

Decision rule: if the organisation and partner conditions are verified and a bounded feasibility question remains that could change a real investment decision, ask Innosuisse and the research partner whether an Innovation Cheque fits. If the idea is already ready for collaborative science-based development and the company can carry the stated contribution, investigate a joint innovation project. If neither description fits, the responsible answer is not “apply anyway”; it is to clarify the work or choose a different financing route.