AI in the Digital Transformation of SMEs

Context

Artificial intelligence has become an omnipresent selling point in enterprise software. Vendors promise productivity gains, cost reduction, and competitive advantage through predictive models, intelligent automation, and natural language analysis. For many Portuguese SMEs, the pressure to "adopt AI" is real—boards wonder if they are falling behind, middle managers receive commercial proposals weekly, and European institutional communications reinforce the urgency of digital transition.

The problem is not the promise of AI; it is the order of execution. AI in digital transformation only creates value when the company already knows which operational or strategic decisions it wants to improve, has governed data feeding those decisions, and has digitalized processes that allow the solution to scale. Without this foundation, AI projects become isolated pilots, consuming budget and management time, rarely moving beyond proof of concept. Evidence shows that digital maturity precedes the useful adoption of AI—not the other way around.

This article examines why sequence matters. It analyzes the state of digital transformation in Portuguese SMEs, the technical and organizational prerequisites for AI, the mechanisms linking data governance to business value, and the decisions boards and executive teams should make before investing in machine learning capabilities. This is not an implementation guide; it is an analysis of the conditions under which AI in digital transformation moves beyond hype and becomes a strategic lever.

What is lost when this topic is treated superficially: companies invest in tools before clarifying the decision problem, accumulate data silos without governance, and blame technology when ROI fails to materialize. The consequence is not just financial waste—it is the erosion of internal trust in digital change, making future initiatives harder to approve and execute.

The State of the Evidence

Portugal ranks 17th among the 27 EU Member States in the DESI 2025 index, according to the European Commission. The State of the Digital Decade 2025 report highlights strengths in digital public services and 5G coverage (65.2% of households in the 3.4–3.8 GHz spectrum), but identifies digital skills as a structural weakness: only 56% of the Portuguese population has basic digital skills, close to the European average of 55.6%. This indicator is critical because digital skills determine an organization’s ability to absorb new tools—including AI.

The Portuguese business fabric consists of 532,174 non-financial companies, of which 99.9% are SMEs, according to definitive INE data for 2024. These companies generated an aggregate turnover of approximately €319.2 billion in 2023, representing about 58% of the total non-financial sector. The concentration in micro, small, and medium-sized enterprises means that most organizations have limited resources for technological experimentation and depend on solutions that deliver rapid and measurable returns.

Research on AI adoption in SMEs shows consensus on three points. First, companies with digitalized processes and centralized data are more likely to extract value from AI than those still operating with fragmented or analog systems. Second, successful AI use cases start with well-defined decision problems—demand forecasting, inventory optimization, predictive maintenance, customer segmentation—not with generic exploration of "AI opportunities." Third, the lack of internal skills is the most cited obstacle, ahead of technology cost or data availability.

Where there is disagreement: the role of generative AI (large language models) in SMEs. Some authors argue that LLMs democratize AI by eliminating the need for structured historical data, enabling automation of writing, customer service, and internal search with minimal investment. Others warn that generative AI without human validation and without integration into business processes produces inconsistent outputs, compliance risks, and a false sense of digital progress. Evidence on sustainable ROI from LLMs in SME operational contexts remains scarce.

The Portuguese innovation ecosystem shows signs of dynamism. The Startup Portugal 2024 report records 4,719 active startups, a 16% increase over the previous year, with 63% focused on information and communication technologies. Many of these startups offer AI solutions targeted at SMEs—from chatbots to predictive analytics tools. Capital raised reached €2 billion in 2024, a 40% increase over 2023. This ecosystem creates accessible technology offerings, but also commercial noise that makes it difficult for SMEs to distinguish between mature solutions and premature promises.

COTEC Portugal recognized 1,056 companies with the COTEC Innovative Status in 2024, a 33% increase over 2023. Companies with this status invest more than 10% of gross value added in research and development, signaling a technological absorption capacity above average. However, they represent a minority of the business fabric—the majority of SMEs still operate with insufficient digital maturity to integrate AI in a structured way.

The Mechanisms

Decision Before Data, Data Before AI

The first causal mechanism is the logical sequence between decision problem, data governance, and AI application. Companies that start by identifying critical decisions—how much stock to order, which clients to prioritize, when to perform preventive maintenance—can then determine what data they need, with what quality and frequency. AI enters as a tool to improve those decisions, not as an end in itself.

Companies that reverse the sequence—acquiring AI tools before mapping decision processes—face two problems. First, they do not know what questions to ask the models, resulting in dashboards no one consults or forecasts no one uses. Second, they discover too late that the necessary data does not exist, is scattered across incompatible systems, or lacks the quality to train reliable models.

Data governance includes four dimensions: quality (accuracy, completeness, consistency), accessibility (centralization, APIs, permissions), security (encryption, access control, audit), and regulatory compliance (GDPR, retention, anonymization). Many Portuguese SMEs lack integrated ERP systems or consistent data hygiene practices. Sales data may be in spreadsheets, production data in legacy systems, customer data in CRM not synchronized with invoicing. Without prior integration, any AI model consumes biased or incomplete data, producing irrelevant outputs.

Digital Maturity as Prerequisite, Not as Outcome

The second mechanism is the relationship between digital maturity and AI absorption capacity. Digital maturity is not binary—companies are on a spectrum from analog processes to fully digitalized and automated processes. AI requires a minimum level of maturity: core processes digitalized, historical data accessible, a culture of evidence-based decision-making, and internal skills to interpret model outputs.

Companies at early stages of digital maturity benefit more from basic digitalization—implementing ERP, centralizing data, automating financial reporting, training teams in data literacy—than from AI pilots. The temptation to "skip steps" is understandable, especially when vendors promise that AI solves data problems. In practice, AI amplifies input data quality: garbage in, garbage out.

Digital maturity models (for example, MIT Sloan or Gartner frameworks) identify progressive stages: process digitalization, system integration, descriptive analytics, predictive analytics, and autonomous optimization. Generative AI can support any stage—email drafting, internal FAQs, document synthesis—but predictive or prescriptive AI requires the company to have already reached intermediate stages.

Internal Skills and Vendor Dependency

The third mechanism is the trade-off between internal skills and dependency on external vendors. SMEs rarely have data scientists or machine learning engineers on permanent staff. The options are to hire specialized consulting, acquire pre-trained SaaS solutions, or train internal teams in low-code/no-code tools.

Each option has implications. External consulting accelerates diagnosis and prototyping but creates dependency if knowledge is not transferred. SaaS solutions reduce technical complexity but limit customization and may not integrate with legacy systems. Internal training builds sustainable capacity but requires time and ongoing investment.

Evidence suggests that SMEs successful in AI-driven digital transformation adopt hybrid models: they hire consulting for diagnosis and data architecture design, acquire SaaS tools for standard use cases (predictive CRM, chatbots), and train internal teams for maintenance and incremental evolution. The common mistake is to outsource everything, losing the ability to govern the solution in the medium term.

Data Culture and Organizational Resistance

The fourth mechanism is cultural. AI requires decisions to move away from intuition or hierarchy and incorporate quantitative evidence. This shift meets resistance in organizations where accumulated experience is valued above data analysis, or where middle managers see AI tools as a threat to decision-making autonomy.

John Kotter, in his eight-step change management model published in 1996 by Harvard Business School Press, identifies creating a sense of urgency, forming a guiding coalition, and communicating the vision as critical initial steps. Applied to AI-driven digital transformation, this means the board must clearly articulate why data-driven decisions are strategically necessary, who leads the change, and what concrete benefits the organization expects to achieve.

Without cultural alignment, technically successful AI pilots fail in adoption. Teams continue to use old processes in parallel, ignore recommendations from predictive models, or passively sabotage data collection. Digital transformation is not an IT project—it is an organizational change project that requires executive sponsorship, persistent communication, and redesigned incentives.

ROI and Learning Cycles

The fifth mechanism is the relationship between ROI expectations and learning cycles. Companies expect rapid returns from AI investments, but many use cases require iteration: training models, validating outputs, adjusting processes, retraining with new data. This cycle can take six to twelve months to stabilize.

SMEs that do not define clear KPIs before AI pilots rarely scale solutions. KPIs should be specific (15% reduction in dead stock, 10% increase in conversion rate, 20 hours per month reduction in manual reporting) and continuously monitored. Without metrics, it is impossible to distinguish success from failure or justify incremental investment.

The common trap is to measure activity (number of models trained, volume of data processed) instead of impact (improved decisions, costs avoided, incremental revenue). AI that does not change business decisions does not generate value, regardless of technical sophistication.

The Portuguese Case

Portugal invests 1.75% of GDP in research and development, according to 2024 data from INE, Pordata, and Eurostat, with a national target of reaching 3% by 2030. This positions Portugal fifth among EU countries with the highest R&D increase over the past decade, signaling political commitment to innovation. However, investment is concentrated in universities and research centers—transfer to SMEs remains limited.

The Portugal 2030 program provides €23 billion in European funds for 2021–2027, with Compete 2030 allocating €3.9 billion specifically for innovation and digital transition. By April 2026, over €3.8 billion had been executed and more than €7.9 billion approved. These funds finance digital maturity diagnostics, ERP implementation, digital skills training, and AI pilots. AICEP Portugal Global supports over 17,000 exporting companies, many of which seek digitalization to compete in international markets where clients demand digital supply chain integration.

Family businesses represent about 75% of the Portuguese business fabric, according to the Family Business Association. These companies face specific challenges in AI-driven digital transformation: generational succession can create friction between founders resistant to change and successors advocating accelerated digitalization; family governance can hinder technology investment decisions when returns are uncertain; and organizational culture may value personal relationships and intuition over formal processes and data.

The PME Líder 2024 program, managed by IAPMEI, recognized 13,394 companies with superior financial performance: aggregate turnover above €61 billion, exports over €10 billion, more than 429,000 jobs, and average financial autonomy of 59.4%. Of these, 71.9% are small companies, 22.3% medium, and 5.8% microenterprises. This segment is the natural target for structured AI adoption—they have the scale to justify investment, financial capacity to absorb risk, and sufficiently formalized processes to digitalize.

Where Portugal diverges from the international pattern: exports represent approximately 50% of GDP, with a public target of reaching 60% by 2030. This export orientation creates competitive pressure for digitalization—international clients demand traceability, EDI integration, and compliance with quality standards that depend on structured data. Exporting companies have a stronger incentive to invest in digital maturity than those focused exclusively on the domestic market.

Implication for SMEs: the Portuguese context offers significant public funding, a growing tech startup ecosystem, and competitive pressure that justifies digital transformation. However, most SMEs still lack the foundation of digitalized processes and governed data necessary for AI to generate value. The strategic priority is not to adopt AI quickly—it is to build digital maturity in phases, starting with critical decisions and the data that supports them.

Management Decisions

Boards and executive teams face three structuring decisions before investing in AI: which operational or strategic decisions they want to improve, what level of digital maturity the company has, and which capability model (internal, external, hybrid) to adopt.

First decision: map critical decisions. The board should ask: which decisions do we make today without sufficient data? Where do intuition or experience fail most often? Which decisions, if improved, would have the greatest impact on margin, risk, or growth? Examples: pricing decisions (what margins to apply to which clients), inventory decisions (how much stock to keep of each SKU), commercial resource allocation (which leads to prioritize), maintenance decisions (when to intervene in equipment before failure).

This analysis does not require external consulting—it can be conducted internally through workshops with operational managers. The output should be a short list (three to five decisions) prioritized by potential impact and data feasibility. Decisions dependent on non-existent or dubious quality data should be excluded until data governance improves.

Second decision: diagnose digital maturity. The company should assess: are core processes digitalized? Is data centralized or scattered in silos? Is there a culture of evidence-based decision-making? Do teams have basic data analysis skills? A rigorous diagnosis can use public frameworks (for example, the MIT Sloan Digital Maturity Model) or hire specialized management consulting in digital transformation for an independent audit.

If the diagnosis reveals low maturity (analog processes, data in spreadsheets, no integrated ERP), the priority is not AI—it is basic digitalization. Investing in AI at this stage leads to waste. If maturity is intermediate (ERP implemented, data centralized but with quality gaps, some descriptive analytics), the company can explore simple AI use cases (demand forecasting with historical sales data, automatic customer segmentation) while improving data governance.

Third decision: choose capability model. SMEs have three main options. First: hire external consulting for diagnosis, data architecture design, and pilot implementation, with knowledge transfer to the internal team. This option accelerates time-to-value but requires active dependency management—the contract should include documentation, training, and clear handover. Second: acquire vertical SaaS solutions (predictive AI CRM, predictive maintenance platforms, dynamic pricing tools) that do not require deep technical skills. This option reduces technical risk but limits customization. Third: train the internal team in low-code/no-code tools (Microsoft Power BI with machine learning, Google AutoML, citizen data science platforms). This option builds sustainable capacity but requires time and executive sponsorship.

The choice depends on three variables: strategic urgency (how quickly the company needs to improve decisions), internal talent availability (is there appetite and capacity to learn?), and budget (capital available for investment vs. operational expense). Hybrid models—consulting for architecture, SaaS for standard cases, internal training for maintenance—tend to balance risk and sustainability.

Trade-offs to consider. Investing in AI before digitalizing core processes results in technological silos: the AI tool operates in isolation, does not integrate with existing systems, and requires manual rework to feed data or extract outputs. Investing only in digitalization without AI use cases can generate frustration—teams wonder when they will see returns from data governance efforts. The balance is to design a phased roadmap: digitalize critical processes first, implement AI pilots in parallel to generate quick wins, and scale solutions only after ROI validation.

Another trade-off: speed vs. learning. Acquiring a ready-made SaaS solution accelerates implementation but reduces organizational learning—the company uses the tool but does not understand the underlying mechanisms. Developing internal capacity takes longer but builds knowledge that allows the solution to evolve over time. The decision depends on the time horizon: if the company faces immediate competitive pressure, SaaS may be the rational choice; if the goal is to build sustainable competitive advantage, investment in internal capacity is justified.

Questions the board should ask before approving AI investment: (1) What specific decision does this tool improve? (2) Do we have the necessary data, with sufficient quality? (3) Who in the organization will use the outputs, and are they trained to interpret them? (4) How do we measure success—which KPI changes, by how much, and in what timeframe? (5) If the pilot fails, what do we learn and how do we adjust?

Limits and Unknowns

The available evidence on AI ROI in Portuguese SMEs is scarce. International studies show mixed results: success stories in demand forecasting and predictive maintenance coexist with high pilot abandonment rates. There is no consensus on what percentage of AI projects in SMEs reach operational scale—estimates range from 20% to 40%, depending on sector and prior digital maturity.

The argument of this article—that digital maturity precedes the useful adoption of AI—applies less to generative AI use cases that do not depend on structured historical data. A chatbot trained on internal documentation can generate value even in companies with low digital maturity, as long as the documentation exists and is accessible. However, the sustainability of this value remains uncertain: LLMs require continuous updating, human validation, and management of compliance risks (hallucinations, bias, privacy).

Contexts where the argument does not apply: tech startups born digital, professional services firms with already digitalized processes, and highly regulated sectors (banking, insurance, healthcare) where data governance is legally mandatory. In these cases, the barrier to AI adoption is not digital maturity—it is talent availability, regulatory clarity, or cultural resistance.

Critical unknown: how AI costs will evolve over the next three to five years. If AI tools become accessible commodities (via low-cost SaaS or open-source), the entry barrier decreases and the argument to "build maturity first" loses strength. If, on the other hand, useful AI continues to require deep customization and integration, digital maturity remains an unavoidable prerequisite.

Implications for Decision-Makers

For CEOs, CFOs, and boards of Portuguese SMEs, the central implication is clear: this is not a sprint—it is a marathon of organizational capability building. Companies that invest in AI before building a foundation of digitalized processes and governed data waste capital and management time. Companies that indefinitely postpone digitalization lose competitiveness to rivals already making evidence-based decisions.

The next step is not to hire an AI vendor. It is to conduct a rigorous internal diagnosis: map critical decisions, assess digital maturity, identify data gaps, and design a phased roadmap that aligns process digitalization with measurable AI use cases. This diagnosis can be conducted internally or with the support of management consulting specialized in digital transformation and data analytics.

Three diagnostic questions for the board:

• Which three operational or strategic decisions, if improved, would have the greatest impact on margin or growth in the next 12 months?
• Do we have the necessary data to support these decisions, with sufficient quality and accessibility, or do we first need to invest in data governance?
• Do we have internal skills to interpret AI model outputs, or are we fully dependent on external vendors for operation and maintenance?

If the answer to the second or third question is negative, the strategic priority is not AI—it is to build the digital foundation that enables AI to generate sustainable value. Public funding through programs such as Compete 2030 can co-finance digital maturity diagnostics, ERP implementation, and digital skills training. Eligibility for incentives should be validated before designing the investment roadmap.

For companies that already have intermediate digital maturity—core processes digitalized, data centralized, analytics culture—the next step is to select an AI use case with three characteristics: measurable impact (clear KPI), available data (sufficient quality history), and executive sponsorship (a decision-maker committed to using the outputs). Successful pilots generate organizational learning and internal credibility, facilitating later scaling.

Digital transformation does not end with AI adoption. It is a continuous process of aligning strategy, processes, data, technology, and organizational culture. Companies that treat AI as an isolated project fail. Companies that integrate AI into a coherent vision of digital transformation, with clear governance and structured learning cycles, build sustainable competitive advantage.

Sources

• European Commission (2025), State of the Digital Decade 2025, available at https://digital-strategy.ec.europa.eu/
• INE (2024), Empresas em Portugal — definitive data, available at https://www.ine.pt/
• Startup Portugal (2024), Startup Entrepreneurial Ecosystem Report 2024, available at https://startupportugal.com/startup-entrepreneurial-ecosystem-report-2024/
• Kotter, J. P. (1996), Leading Change, Harvard Business School Press
• COTEC Portugal (2024), COTEC Innovative Status — 2024 Edition, available at https://cotecportugal.pt/
• IAPMEI (2024), PME Líder 2024, available at https://www.iapmei.pt/PRODUTOS-E-SERVICOS/Qualificacao-Certificacao/PME-Lider.aspx
• Family Business Association (2024), data on family businesses in Portugal, available at https://www.empresasfamiliares.pt/
• Agency for Development and Cohesion (2026), Portugal 2030 — Execution Report, available at https://portugal2030.pt/
• INE / Pordata / Eurostat (2024), data on R&D investment in Portugal, available at https://www.pordata.pt/