ACCIÓ AI grant 2026: how SMBs can use it to deploy AI agents

There is a new grant on the table for SMBs looking to adopt artificial intelligence in their processes. It is called the ACCIÓ AI Voucher, part of the 2026 ACCIÓ Innovation Vouchers programme, and it covers up to 8,000 EUR non-repayable. The relevant point is not the amount: it is what the voucher pays for and what it leaves out.

This article explains how the voucher works, who can apply, what it covers and what it does not, and how an SMB should prepare so that the 8,000 EUR translate into a productive AI agent rather than a report that ends up in a drawer.

What the ACCIÓ AI voucher actually is

ACCIÓ (Agència per a la Competitivitat de l’Empresa, under the Government of Catalonia) runs an annual package of innovation vouchers. The 2026 call covers four lines: European R&D programmes, green and climate, industrial property, and AI adoption. The AI line covers up to 8,000 EUR.

The amount is paid at 100%: the company does not advance any of the subsidized cost. The payee is not the beneficiary company but the accredited assessor recognised by ACCIÓ as technically qualified to perform the work.

That already signals what the voucher pays for: advisory services, not construction.

What is covered and what is NOT

This is the point where most commercial communications about the voucher are ambiguous. Let us make it explicit.

Covered by the voucher (the 8,000 EUR):

• Identification and prioritization of viable AI use cases for the business
• Analysis and preparation of the data the agent will require (quality, format, accessibility)
• Analysis of the technical infrastructure and implementation costs
• Final report with prioritized use cases, data map, and technical and economic estimate

Not covered by the voucher (contracted separately):

• Technical build of the agent (prompt design, orchestration, code)
• Integrations with CRM, ERP, email, WhatsApp, web, or any existing tool
• Inference cost of the AI models (Claude, GPT, Gemini, open-weights)
• Internal team training once the agent is live
• Maintenance, observability, ongoing evaluation, and incident response
• Kill-switch and human fallback (baseline pieces of any serious implementation)

The voucher, therefore, funds structured diagnosis. Very useful for deciding which agent to build and in what order. It does not replace the implementation phase.

Access requirements

The AI line carries specific requirements that filter a significant share of the SMB fabric:

• Priority sectors of the new European industrial strategy. Specific sectors are detailed in the regulatory bases. Not all sectors are eligible.
• Minimum 10 employees. Microbusinesses under 10 staff are out of this line, though other vouchers in the package may apply.
• Minimum 2 years of existence since company incorporation.
• Headquartered in Catalonia.

If the company does not meet any of these requirements, the conversation shifts from “is it worth applying” to “what other support or path fits.”

Timing is the critical variable

ACCIÓ allocates vouchers in chronological order of application. Each year, funds run out in days, sometimes hours. The difference between securing the voucher and missing it is rarely about the project; it is about preparation.

What preparation means in practice:

1. Define the use case before the call opens. The company that already knows which process to automate arrives with its homework done.
2. Pick the accredited assessor early. The voucher is filed jointly with a specific assessor; searching for one once the call is open is too late.
3. Corporate documentation ready. Tax identity, social security status, annual accounts: prepared before day zero.
4. Internal brief written. What the team does today in the candidate process, what volume it handles, what tools are involved, what the success metric looks like.

The preparation is not bureaucratic, it is strategic: it forces the company to think clearly about the problem it wants to solve.

Separating diagnosis from implementation usually yields better results

Some accredited consultancies run the voucher and also handle implementation. Others do only the covered part and refer the build to partners. And some companies commission the diagnosis with one assessor and the implementation with a separate team.

None of the three models is inherently wrong. But in our experience, separating the two roles offers three advantages:

1. Independence of judgment. The assessor has no incentive to inflate scope to bill more implementation afterwards.
2. Technical specialisation. Building custom AI agents with real integrations and a kill-switch is a craft. A diagnosis consultancy is rarely the same shop as a technical implementation team.
3. Freedom of decision. If the diagnosis concludes the use case is not viable or another problem should come first, the company is not tied to building anything to fill the consultant’s scope.

How serpixel fits in

serpixel builds custom AI agents for SMBs. It is not an ACCIÓ-accredited assessor, so it does not collect the voucher directly. The typical role is implementation partner:

• The company applies for the voucher with an accredited assessor.
• The assessor delivers the report with prioritized use cases, data map, and implementation estimate.
• With that report in hand, serpixel quotes and builds the agent: prompt design, orchestration, integrations, kill-switch, human fallback, ongoing evaluation.

The agent’s goal is always the same: handle the mechanical layer of a bounded process so the human team focuses on the work that really adds value (judgment, client relationships, ambiguous decisions, craft). It does not replace anyone. It frees time.

Every implementation includes two non-negotiable pieces:

• Kill-switch: a mechanism for the client to disable the agent instantly, without depending on the provider.
• Human fallback: a clear path for the process to continue if the agent is disabled or fails.

Without these two pieces, an agent is not a production project. It is a gamble.

The homework to do now

Regardless of the voucher, any SMB that sees value in adopting AI benefits from preparing this document before talking to anyone (assessor, provider, consultant):

1. Candidate process: which repetitive operation is the first candidate? Pick one, not three.
2. Volume: how many times a month does it happen? Provide a real number.
3. Current state: who runs it, how many hours it takes, what tools it uses.
4. Rules: what percentage of the process follows clear rules and what requires human judgment.
5. Success metric: how will we know the agent works? It must be measurable.
6. Data involved: where it lives and who has access.
7. Impact of freed human time: what the team will do with the recovered hours.

With this document, both the conversation with an accredited assessor to apply for the voucher and the later implementation quote become much faster and much more precise.

What to do now

If your company meets the requirements (10+ employees, 2+ years, priority sector), the reasonable calendar looks like this:

• Now: draft the homework document (points 1-7 above).
• Before the call opens: pick the accredited assessor and align with them on diagnosis scope.
• In parallel: talk to the implementation partner to understand, before the report exists, what building the agent means in your context.
• When the call opens: file the application on day one.

If you want to have this implementation conversation alongside the voucher (or directly outside the voucher if your company does not qualify), let’s talk in a 30-minute discovery session. You arrive with the homework document half-done and leave the session with a clear view of what the implementation looks like: scope, risks, success metric, kill-switch. No commitment.