This article is part two of “Opportunities and Pitfalls of Generative AI from an SME Perspective” by senior reacher Janne Kauttonen and RDI Communications Specialist Martti Asikainen, originally published in Finnish in Haaga-Helia University of Applied Sciences’ eSignals Pro.
Janne Kauttonen & Martti Asikainen 1.2.2025
In 2023, generative AI (GenAI) broke free from the exclusive domain of researchers and tech giants, becoming accessible to organisations of all sizes. This watershed moment sparked an explosion of innovation, giving rise to countless AI-powered startups and services. For small and medium-sized enterprises (SMEs), this shift has irreversibly transformed the landscape of knowledge work and creative industries. Integrating generative AI into business operations has become not merely advantageous, but essential for firms hoping to maintain their competitive edge and secure future success.
In this article and its predecessor, we examine the distinctive features of generative AI, contrasting its operational characteristics and use cases with those of ‘traditional’ AI systems, which rely on predetermined rules and models. We specifically address the implementation challenges from an SME perspective, offering practical insights into technical considerations these businesses must navigate. For these businesses, the message is stark: incorporating generative AI into their operations is no longer optional. As market pressures intensify, it has become a fundamental requirement for survival and success in an increasingly competitive landscape.
Generatiivisen tekoälyn integrointi pienten ja keskisuurten (pk-yritysten) toimintoihin voi tuoda mukanaan täysin uudenlaisia haasteita, mikäli sitä vertaa perinteisempiin tekoälyratkaisuihin. Tähän on useita eri syitä, joista kenties yleisin on sen kyvykkyyden yliarviointi, joka johtuu ennen kaikkea generatiivisen tekoälyn toimintaan liittyvistä väärinkäsityksistä ja sen toimintaperiaatteiden vähäisestä ymmärryksestä.
Kokemuksemme mukaan pk-yritykset sortuvat helposti olettamaan, että tekoälyjärjestelmät toimisivat ihmismäisellä ymmärryksellä ja omaisivat meille ominaisen päätöksentekokyvykkyyden. Tämä puolestaan johtaa helposti epämiellyttäviin yllätyksiin ja ennen kaikkea pettymyksiin. Generatiiviset mallit ovat taitavia luomaan tekstiä ja kuvia, mutta ne voivat helposti erehtyä kontekstista. Tämä puolestaan voi tuottaa virheellistä tietoa, jonka vuoksi generatiivisten mallien käyttö liike-elämässä ja tuotannossa vaatii aivan erityistä valvontaa ja tarkkuutta.
Generatiivisen tekoälyn käyttö tapahtuu, ainakin vielä toistaiseksi, syötesuunnittelun kautta. Käytännössä tämä tarkoittaa sitä, että käyttäjän tulee osata sanoittaa ongelma oikealla tavalla parhaan lopputuloksen saamiseksi (esim. teksti, kuva tai koodi). Syötesuunnittelu on yksi suurimmista käytännön eroista perinteisiin tekoälymalleihin verrattuna, joissa sitä ei ole lainkaan, vaan syötteet ovat aina tietynlaisia ja tarkasti koodin kautta määriteltyjä.
Tehokas syötesuunnittelu on taito, joka vaatii harjoittelua, yritystä ja erehdystä sekä malleihin liittyvien parametrien säätämistä. Toisaalta generatiivisen tekoälyn kehittyessä myös syötesuunnittelu kehittyy ja sitä pitää muuttaa kehityksen mukana (kts. esimerkiksi englanninkielinen Prompt Engineering Guide). Klassisen Boolen logiikan ymmärtäminen voi auttaa syötesuunnittelussa, sillä sitä käytetään lähes kaikissa tietokannoissa.
Integrating generative AI into small and medium-sized enterprises (SMEs) presents unprecedented challenges compared to conventional AI solutions. At the heart of these difficulties lies a widespread misconception: the tendency to overestimate AI’s capabilities, stemming from fundamental misunderstandings about how these systems actually work.
Our research reveals a concerning pattern among SMEs. Many fall into the trap of assuming these AI systems possess human-like understanding and decision-making abilities. This misapprehension inevitably leads to uncomfortable surprises and, more critically, significant disappointments. While generative models excel at creating text and images, they can readily misinterpret context. In the business environment, where accuracy is paramount, this tendency to produce misleading information demands rigorous oversight and exceptional vigilance.
The current state of generative AI hinges on a crucial yet often overlooked element: prompt engineering. This process requires users to articulate their problems with precise language to achieve optimal results, whether generating text, images, or code. This represents a fundamental departure from traditional AI models, which operate on fixed, pre-programmed inputs defined strictly through code.
Mastering prompt engineering is not merely a technical requirement—it’s an art that demands practice, trial and error, and careful parameter adjustment. As generative AI continues to evolve, so too must prompt engineering techniques adapt and advance. While resources like the Prompt Engineering Guide offer valuable insights, a grounding in classical Boolean logic—the backbone of most database systems—can provide SMEs with a solid foundation for developing effective prompting strategies.
For small and medium-sized enterprises (SMEs) grappling with limited resources, developing and hosting bespoke generative AI models rarely makes financial sense. Instead, cloud-based AI services offer a compelling alternative, providing access to cutting-edge models without the substantial overhead costs.
This accessibility comes through API (Application Programming Interface) services, with industry leaders including OpenAI, Stability AI, Poe and Cohere at the forefront. For businesses seeking to fine-tune existing models, platforms like MosaicML and Huggingface provide additional capabilities.
Consider an e-commerce operation: SMEs can dramatically enhance efficiency by automating product descriptions, customer service responses and marketing content through API integrations. This approach eliminates the need for in-house AI model development and training—requiring only the integration of API calls into existing e-commerce infrastructure.
The landscape also includes established tech giants, with Google Cloud AI, Amazon Web Services (AWS) and Microsoft Azure offering generative AI capabilities as part of their broader service portfolios. Most commercial solutions operate within larger ecosystems—AWS and Huggingface being prime examples—where providers offer a comprehensive package of API services, computing power, storage solutions, and both commercial and open-source models (Marr 2023).
The complexity of generative AI models far exceeds that of their traditional counterparts, demanding substantially greater computational resources. For companies lacking in-house technical expertise, finding the right technical partners becomes crucial. Successfully integrating GenAI into production and business operations can be both costly and time-intensive, often hinging on the availability of AI and cloud service specialists.
This reality makes trusted partner networks—including consultants, product suppliers, and resellers—instrumental to successful business integration. However, partner selection requires careful consideration. Any collaboration must be with those who thoroughly understand GenAI’s unique characteristics, including its limitations, computational costs, and evolving ethical and legal considerations.
For SMEs, one of the most significant challenges lies in understanding the novel risks associated with these models. Recent studies indicate that identifying and understanding model errors ranks as businesses’ primary concern, surpassing even cybersecurity and legal considerations (Chui et al. 2023). This anxiety stems from the inherent difficulty in detecting and quantifying GenAI errors.
The quality of generated text or images depends heavily on context and audience perception, making it far more challenging to measure than, say, the accuracy of a classification model. Consider the varying tolerance for errors: an internal report might accommodate minor inaccuracies, while a public stock exchange announcement demands absolute precision. The ‘black box’ nature of GenAI models presents another significant challenge for businesses, as these systems cannot explain their reasoning or source their information—a particular concern in critical applications like credit ratings or customer segmentation, where transparency is essential for maintaining customer trust and regulatory compliance.
Data management remains fundamental to all advanced analytics deployment (Iansiti & Lakhani 2020), applying equally to both traditional and generative AI (AlEidan & Amezaga 2023). While GenAI models might not always require massive datasets or custom training, efficient handling of inputs and outputs demands robust data management practices.
SMEs with well-organised, accessible data find GenAI tool implementation significantly more straightforward. When building Retrieval-Augmented Generation (RAG) systems or fine-tuning models, data volume becomes critical. Unlike larger corporations, SMEs typically have access to smaller datasets, potentially leading to bias issues if the data fails to represent the target demographic adequately.
Working with insufficient or poor-quality data risks not only incorrect results but potentially breaching EU AI ethical and legislative principles. In such cases, purchasing additional data or limiting AI applications to use cases where data volume isn’t critical might prove the most prudent approach (Bhattacharyya 2023).
Open-source generative AI models have stormed the market, with platforms like the Huggingface ecosystem offering a wealth of options. While SMEs can implement these models within their own systems, such deployment demands significant resources for implementation, maintenance, and monitoring. Open-source solutions offer SMEs advantages like cost-free access and enhanced control, but they come with technical requirements and indirect costs in terms of workforce and computational power.
Most SMEs lack the technical expertise and time to develop bespoke solutions using open models. Therefore, opting for ready-made services is typically advisable, despite direct costs through API calls or licensing fees. For SMEs taking their first steps with AI, the focus should be on conceptualising AI services or products rather than getting bogged down in technical minutiae. Once the use case and value proposition are validated, developing solutions based on open-source models becomes a less risky proposition.
The primary challenge with commercial off-the-shelf services is their accessibility to competitors. Simply employing GenAI services doesn’t automatically confer competitive advantage unless businesses discover exceptionally innovative applications. Therefore, AI implementation should be woven into a broader corporate strategy that considers market dynamics, staff capabilities, and technological infrastructure.
For small and medium-sized enterprises, carefully considering generative AI’s role in competitive strategy and its broader operational impact is crucial. AI need not dominate the entire business strategy; rather, it can be precisely integrated as one component of a broader approach, particularly when targeting niche market segments innovatively. This targeted deployment can help SMEs address specific customer needs in ways competitors haven’t yet explored, potentially establishing the business as an industry pioneer rather than merely keeping pace with the competition (Soni 2023).
As we approach 2025, it’s reasonable to expect that most knowledge workers will have experimented with user-friendly generative AI applications like ChatGPT and Claude, either professionally or personally. The key challenge lies in scaling up from these individual experiments to organisation-wide implementation, where AI becomes integral to every employee’s workflow. AI adoption in business typically progresses through three distinct levels (adapted from Anderson 2023):
The progression through these adoption levels involves increasingly complex challenges. While the initial level is readily accessible—anyone can experiment with online tools—advancing from individual to team-level implementation demands coordination, standardised practices, and internal protocols. The first crucial step in enterprise-wide AI integration is establishing clear boundaries: determining where AI can and cannot be applied within the organisation’s operations through well-defined guidelines and frameworks.
The transition to team and particularly enterprise-level implementation faces additional hurdles, notably the limitations of pre-trained base models. Without customisation using company-specific data, these models may lack the precision required for critical decision support. The risk of erroneous outputs or hallucinations becomes significantly higher than in small-scale individual experiments. Importantly, organisations that are data-driven from inception find it considerably easier to achieve the highest level of generative AI adoption. This foundation enables AI to meaningfully influence both strategic decision-making and business performance.
The impact of generative AI represents a more profound shift than traditional predictive AI, touching more roles and workers across organisations. The situation bears striking similarities to the 1980s, when computers migrated from centralised mainframe rooms to individual desktops. Just as personal computing rapidly became universal rather than exclusive, generative AI is following a similar democratisation trajectory.
For knowledge-intensive businesses, embracing generative AI has become essential for maintaining competitive edge. With these tools widely available, it’s reasonable to assume competitors are already leveraging them for knowledge work and broader business operations. However, SMEs must set realistic expectations about achievable benefits while remaining mindful of model limitations. Strategic planning and unique, tailored services addressing specific business needs remain crucial for gaining competitive advantage.
"The situation bears striking similarities to the 1980s, when computers migrated from centralised mainframe rooms to individual desktops. The situation bears striking similarities to the 1980s, when computers migrated from centralised mainframe rooms to individual desktops."
The future landscape of GenAI remains uncertain, largely dependent on how businesses develop innovative use cases and where technical and practical boundaries emerge. Success demands organisations’ ability to identify sector-specific opportunities and challenges in AI adoption. Understanding which processes and services benefit most from automation and AI-assisted development becomes paramount.
At its best, GenAI can help businesses create entirely novel service concepts or optimise existing processes in ways that distinctly set them apart from competitors.
We offer the following essential guidance for small and medium-sized enterprises. These five points warrant careful consideration, as following them will help your business leverage AI effectively and responsibly (adapted from Koupanou 2023; Iansiti & Lakhani 2020):
Actively monitor AI developments to stay abreast of applications and methods relevant to your sector. Social media platforms (Facebook, LinkedIn, and X), newsletters, and forums offer efficient ways to keep current. Take advantage of both free and paid AI training courses to update your team’s practical skills with minimal barriers to entry.
Familiarise yourself with AI services through hands-on experimentation, particularly with prompts. Many free or low-cost GenAI services are available for testing without significant investment. Begin with widely-used tools like ChatGPT or Claude. Test various inputs and tasks to understand how AI responds to different requests. The key is gaining practical understanding of these tools’ strengths and limitations.
Form learning groups with colleagues for shared experimentation and discovery. Identify AI developers in your sector and enquire about their offerings. Build connections with experts and businesses. Share experiences both within and beyond your organisation. Explore application possibilities collectively. Engage with local technology firms and startup communities, and participate in industry events and training sessions for networking opportunities.
Eliminate data and process silos whilst centralising your IT architecture. A genuinely data-centric company can implement both GenAI and traditional AI more swiftly and effectively. Fragmented systems impede AI adoption, while unified IT architecture enables smooth data flow and efficient AI integration. Though this requires strategic planning and investment, it enhances operational efficiency and competitive advantage.
Given generative AI’s rapid evolution, today’s impossibilities may become tomorrow’s realities within months. Monitor your industry’s developments actively and prepare for significant disruption in coming years. Readiness for change is paramount. Develop your staff’s capabilities and ensure organisational adaptability. Conduct sector-specific scenario analyses of AI impacts and develop alternative strategies for your business.
Senior Researcher
+358 294471397
janne.kauttonen@haaga-helia.fi
RDI Communications Specialist
+358 44 920 7374
martti.asikainen@haaga-helia.fi
AlEidan, M. & Amezaga, K., Y. 2023. Data Unleashed: Empowering Small and Medium Enterprises (SMEs) for Innovation and Success. World Economic Forum. Accessed 28.1.2024.
Anderson, M., K. 2023. How to break out of your AI Pilot Phase. Accessed 25.1.2024. Jasper.
Bhattacharyya, S. 2023. Generative AI and Adoption Readiness of different size Businesses. Medium. Accessed 25.1.2024.
Chui, M., Hazan, E., Roberts, R., Singla, A., Smaje, K., Sukharevsky, A., Yee, L. & Zemmel, R. 2023. The economic potential of generative AI: The next productivity frontier. McKinsey & Company. Accessed 25.1.2024.
Dencheva, V. 2023. Popularity of generative AI in marketing in the U.S. Statista. Accessed 21.2.2024.
Marr, B. 2023. Generative AI And The Future of Content Creation. Julkaistu Forbesissa 30.11.2023. Forbes Media. New Jersey. Accessed 21.2.2024.
Iansiti, M., & Lakhani, K. R. 2020. Competing in the age of AI: Strategy and leadership when algorithms and networks run the world. Harvard Business Press.
Koupanou, N. 2023. The AI Boom: Practical Guide to Generative AI for Small Businesses. Towards AI. Accessed 22.1.2024.
Soni, V. 2023. Impact of Generative AI on Small and Medium Enterprises’ Revenue Growth: The Moderating Role of Human, Technological, and Market Factors. Reviews of Contemporary Business Analytics, 6(1), 133–153.