The IA360 Plan launches an AI Voucher for 25,000 SMEs and freelancers. It requires a prior diagnosis, a specific use case and productivity measurement: exactly what we do.

The President of the Government presented the IA360 Plan yesterday, the national roadmap to bring AI to 55% of companies by 2030 (today only 21.1% of those with ten or more workers use it, according to the INE). Its flagship measure is an AI Voucher endowed with €600 million, with a pilot phase in the first half of 2027 and a general call before the end of that year.

The detail that changes the rules: the voucher will not pay for AI tool subscriptions. It will only finance services and solutions provided by European technology companies that contribute development, integration, data processing or process redesign. In other words, real implementation is subsidised, not off-the-shelf software.

The conditions already announced describe a serious project: prior diagnosis of digital maturity, identification of a specific use case, milestone-linked payments, staff training and subsequent measurement of the impact on productivity. The stated goal is to transform 25,000 companies.

For SMEs this means the money will be available, but only for those who arrive with a plan: knowing which process to automate, with what data and how the result will be measured. Companies that arrive with that diagnosis done will have a clear advantage when the call opens.

The voucher requires diagnosis, use case and impact measurement before paying. Our self-diagnostic and the 45-minute session get you ready to apply when the call opens.

Source: elEconomista — Bono IA de 600 millones

Read on SynapsIA

SMEs with a workplace in Madrid can cover 80% of R&D&I expenses, up to €60,000, by hiring a Madrid-based provider such as SynapsIA.

The Community of Madrid keeps the Innovation Voucher call open, a direct aid line for SMEs that hire external R&D&I services. It covers 80% of eligible expenses (excluding VAT) with a cap of €60,000 per beneficiary, accumulable across several actions.

The optimal project base is around €75,000: above that amount the excess is not subsidised, although it can still be part of the project. The aid is channelled through two lines: contracting with public centres, universities or Madrid-based IMDEA institutes (Line a), or with technology innovation support consultancies (Line b).

The external provider is the decisive filter. It must have its tax domicile and workplace in the Community of Madrid. Panel Sistemas Informáticos, S.L. —the company behind SynapsIA— meets this requirement with its address at C/ Nanclares de Oca 5, 28022 Madrid, so an AI implementation project can be submitted under this line.

The strictest rule is the start date: already-started work is not subsidised. A project that starts after the grant resolution, lasting between 1 and 12 months and beginning within 30 days of the resolution, is the ideal candidate. The call works on a first-come-first-served basis until funds are exhausted, so it is best to apply before signing the execution contract.

Documentation includes the representative's ID and power of attorney, VAT number, clearance certificates from AEAT and Social Security, Mercantile Registry certificate or deed of incorporation, and IAE registration. You must also check the accumulated de-minimis limit before submitting. The regulatory bases (06/02/2025), the call (06/10/2024) and the amendment (30/09/2025) are published in the BOCM and should be reviewed before applying.

If you are an SME in Madrid and your AI project has not yet started, the Innovation Voucher can cover 80% of the cost: apply before signing with the provider.

Source: Cheque Innovación — Comunidad de Madrid

Read on SynapsIA

Since 2 August 2026 the bulk of Regulation (EU) 2024/1689 applies, including Article 50 transparency. It has not been postponed.

General application of the AI Act arrived on schedule. The Article 50 transparency obligations — telling people when they interact with an AI system and labelling generated or manipulated content — are now enforceable.

Regulation (EU) 2026/1744, the Digital Omnibus on AI, postponed Annex III high-risk systems to December 2027 and those embedded in regulated products under Annex I to August 2028, but it did not change the transparency timeline, nor the prohibited practices and AI literacy rules in force since 2025.

In practice, any organisation running public-facing conversational assistants, text or image generation, or automated triage of requests already needs a systems inventory, a usage policy and evidence.

If you run a public chatbot or generate content with AI, transparency already applies: review notices, labelling and your systems register.

Source: Reglamento (UE) 2024/1689

Read on SynapsIA

Article 4 requires a sufficient level of AI competence among users. That is proven with records, not good intentions.

The AI literacy obligation has been in force since February 2025 and applies to providers and professional deployers alike. It does not require a diploma: it requires the people operating AI systems to understand their limits, risks and what data may be entered.

The usual mistake is a one-hour talk with no record. In a review, what is requested is the list of trained people, the content delivered, the date and the accepted usage policy.

A realistic plan combines role-based training (management, back office, customer service, technical), a usage policy written in plain language and an annual review.

Training without records is no training. Keep attendees, content, date and policy acceptance.

Read on SynapsIA

Invoice reconciliation, collections chasing and repetitive email replies automate well. Unsupervised decision-making does not.

The pattern that works in production is narrow: an agent reads a document, extracts fields, checks them against the ERP and leaves a proposal for a person to approve. The real saving is in preparation time, not in removing review.

What keeps failing is the autonomous agent with broad system access and no stopping criteria. Without traces, action limits and an error metric, the project gets pulled after three months.

Our rule: no agent goes to production without a measured baseline, an explicit action limit and a full log of what it did.

Automate the preparation of the work, not the final decision. And measure before and after.

Read on SynapsIA

More tenders and corporate procurement processes ask for a certifiable AI management system as a solvency criterion.

ISO/IEC 42001 defines an AI management system: policy, roles, risk assessment, supplier control and continuous improvement. It is to AI what ISO 27001 was to information security.

The interesting shift is not certification itself, but large buyers requiring its controls from suppliers. Anyone unable to document how they govern the AI they use will be excluded from certain tenders.

The good news: much of the documentation overlaps with what the AI Act already requires. Do it once, use it for both.

Prepare the documentation once and reuse it for both the AI Act and ISO 42001.

Read on SynapsIA

Free accounts with customer data pasted into the prompt. It is the most common incident we find in diagnostics.

Almost every diagnostic surfaces the same thing: competent people using free AI tools to move faster, with data that should not leave the organisation. It is not bad faith, it is the absence of an official alternative.

Banning it by memo does not work: usage just hides. What works is providing an acceptable corporate tool, stating clearly what data may be entered and auditing usage proportionately.

The cost of not doing it is paid twice: data protection and contractual confidentiality with clients.

Offer an official alternative before banning. Without one, a ban only removes visibility.

Read on SynapsIA

Without a baseline there is no provable improvement. Three metrics are enough to start: time per case, error rate and volume handled.

The first step of any serious rollout is measuring how work happens today: how long a case takes end to end, how many come back due to errors and how many are handled per week and per person.

With that baseline, the effect of automation stops being an impression. And it enables something more important: stopping in time whatever does not improve.

Publishing generic industry percentages adds nothing. The only numbers that convince a management committee are their own, measured before and after.

Measure for two weeks before touching anything. It is the cheapest part of the project and it holds up everything else.

Read on SynapsIA