Bridging the AI Skills Gap: How the GAIK Toolkit Helps Companies Build Practical, Workflow-Integrated AI Solutions
Text by Umair Ali Khan, 30.11.2025 | Photo by Adobe Stock Photos
Despite tremendous advancements in AI, most organizations still struggle to turn it into real, repeatable value. Many companies experiment with tools like ChatGPT or Microsoft Copilot, run a few pilots, then hit a wall when they try to connect AI to their actual processes and data.
A growing body of reports points to the same root problem: a serious skills gap. We have highlighted the skill-gap dilemma in AI adoption across several studies (e.g., Khan et al., 2025a), based on our direct observations and hands-on AI consultancy work within the Finnish AI Region. This is also evident in our analysis of the AI consultancy data, which shows a strong demand for PoC services, an indication that many companies lack the technical expertise required to assess the feasibility of their own business ideas (Khan, 2025b).
AI budgets and ambitions are rising, but the people who can design, implement, and maintain robust AI solutions are in short supply. In some estimates, the global AI talent gap is close to 50 percent of demand, even as AI investment passes 550 billion USD (IBM, 2024). McKinsey’s recent report also notes that almost half of technology leaders see AI skill gaps as a major barrier to adoption (McKinsey Global Institute, 2025).
At the same time, generic tools and off-the-shelf platforms are not built for the messy, domain-specific reality of business. The result is a growing divide between what is technically possible and what most organizations can actually deploy. A recent MIT study even suggests that around 95 percent of generative AI projects fail to deliver meaningful transformation, often because solutions are not tailored to real workflows and constraints (MIT, 2025).
Our ongoing GAIK project (Generative AI-Enhanced Knowledge Management in Business) is specially designed to work directly in this gap. It focuses on giving companies, especially SMEs, a way to build their own practical, knowledge-driven AI use cases without having to reinvent the entire AI stack from scratch. Based on our AI need analysis of over a hundred companies at FAIR in 2024-2025, and identification of the most recurring knowledge processes and solution needs, we are developing a modular GenAI toolkit that guides organisations from selecting a use case to deploying a functioning, value-creating AI solution.
To understand why such a toolkit is necessary, it is important to look at the practical challenges companies face when adopting AI.
Why AI adoption is hard in practice
Most organizations now have access to powerful AI models; however, that does not mean they are ready to use them effectively. Several cross-cutting challenges keep coming up:
- Skills and expertise. Companies lack people who understand both the business context and the technical landscape: data infrastructure, model selection, security, evaluation, and integration. In many sectors, skills shortages are now one of the top-reported barriers to AI adoption (McKinsey Global Institute, 2025)
- SME reality. Only a small fraction of European enterprises have adopted even one AI technology. Many SMEs are interested but lack internal expertise, confidence, and resources to go beyond simple experiments (EIT Manufacturing, 2025).
- From tools to solutions. Buying an AI tool is easy. Turning it into a workflow that reliably supports sales, operations, or customer service is hard. That requires process redesign, change management, and long-term maintenance.
One of our recent studies (Khan et al., 2025c) describes how many companies currently sit in an “experimentation” phase. They try generic end-user tools and one-off proofs of concept, but often lack the skills, patterns, and frameworks to move toward integrated, custom solutions.
Why generic AI tools are not enough
Most companies start their AI journey with popular end-user tools such as ChatGPT, Microsoft Copilot, Gemini, etc. Our study (Khan et al., 2025c) shows that while these tools are good for simple and recurring tasks, they fall short on several points that are critical for serious business use, such as hallucinations, lack of transparency, limited control over how the models work, and difficulty integrating them into existing systems or workflows.
For example, ChatGPT cannot directly plug into an ERP system to fetch live inventory data and then automatically create a sales order. We argue that generic tools are essentially “standalone assistants”. They sit beside the business, not inside it. They cannot easily connect to CRMs, process automation, or domain-specific rules, which is where real business value usually lies.
Code-based AI frameworks: powerful, but out of reach for many
To build custom AI solutions, developers can use code-based frameworks such as LangChain, LlamaIndex, LangGraph, AutoGen, or CrewAI. However, these frameworks come with their own set of barriers (Khan et al., 2025c):
- They require strong programming skills plus knowledge of cloud infrastructure, deployment, monitoring, and security.
- They are fast-moving and fragmented. Frameworks evolve rapidly, best practices change, and documentation can be incomplete.
- They often assume that evaluation, testing, and data engineering are already mature in the organization, which is rarely true in SMEs.
In other words, code-first frameworks are great if you already have an AI development team. They are much harder to use if you are a medium-sized manufacturer with one overloaded IT generalist and no in-house data scientist.
Low-code and no-code AI platforms: fast starts, hard limits
To lower the barrier, many vendors offer low-code or no-code AI platforms. These tools often provide drag-and-drop workflows, prebuilt connectors to common SaaS systems, and ready-made templates for chatbots, automations, and RAG apps. One such example of low/no-code frameworks is n8n. This looks attractive, especially for SMEs that lack technical skills. However, there are important drawbacks:
- Studies show that only a subset of use cases can be fully tailored in low-code AI platforms, and users have fewer algorithm and architecture choices than with traditional development (DigitalDefynd, 2025).
- Applications depend on a specific platform’s way of handling data, deployments, and pricing. Moving away later can be costly and risky (Pandium, 2024).
- Hidden constraints often appear when workflows grow in complexity, data volume, or user load, especially when large models or heavy integrations are involved.
- Creating AI workflows with a no-coding platform often requires extensive coding for specific use cases, which mitigates the benefits of the no-coding platform and results in an overly complex workflow.
Therefore, companies face a trade-off:
- Code-based frameworks are flexible but demand high technical skills and effort.
- Low-code tools lower the barrier but can block the path to truly tailored, long-term solutions.
The result is a “missing middle” where many organizations get stuck. They can prototype, but they cannot systematically create robust, business-specific AI solutions across their knowledge processes.
The GAIK toolkit for knowledge-driven, company-specific AI
The GAIK project (Generative AI-Enhanced Knowledge Management in Business) is designed to operate in the “missing middle” between generic AI tools and fully bespoke solutions. It brings together academia-industry collaboration to create a business-oriented GenAI toolkit for knowledge management in SMEs. Instead of starting from specific technologies or vendors, GAIK is built around core knowledge processes that reflect how organizations actually create, store, and use information.
GAIK focuses on the following knowledge processes:
- Knowledge capture: Extraction of needed information from business documents/notes, voice recordings, videos, and emails.
- Knowledge access: Intelligent access to organizational knowledge (document repositories, databases, wikis, CRMs)
- Knowledge generation: Automatic generation of business reports and documents
All building blocks, templates, and methods in the toolkit are organized around these processes rather than around any particular model, vendor, or framework. The core of the concept is a simple but powerful distinction between a solution space and a problem space.
The solution space contains generic building blocks, reference workflows, and reusable patterns. The problem space represents each company’s specific use cases, such as “incident reporting in our construction sites” or “sales proposal generation for our product catalog”.
The purpose of the toolkit is to bridge these two spaces so that generic, research-driven artefacts can be configured and guided into concrete, company-specific solutions instead of starting from zero every time.
- generation of business reports and documents
To support different stakeholders inside a company, GAIK toolkit is structured into three complementary layers.
i) The business layer provides tools such as a GenAI product canvas, use-case templates, and a catalog of knowledge services (for example, “knowledge capture through voice” or “document/report generation”), helping managers and domain experts describe their needs in clear business language rather than technical jargon.
ii) The technical layer offers requirements and architecture templates, reusable components that can be used via no-code interfaces or code, and connectors plus security patterns that allow IT and AI specialists to design systems and integrate them with existing infrastructure in a structured way.
iii) The GenAI implementation layer includes AI readiness and maturity assessments for companies, planning tools for implementation, monitoring and governance, and process models for moving from idea to production and keeping solutions up to date, so that the resulting systems are not just prototypes but part of a managed and evolving lifecycle.
How GAIK toolkit helps companies create their own specific use cases
Key concepts in the GAIK Toolkit include the distinction between generic and company-specific use cases, modular building blocks, workflow assets, and a guided configuration process. A generic use case might be something like “incident reporting” or “knowledge access via RAG,” while a company-specific use case turns this into a concrete solution, such as an “incident reporting assistant for a particular construction company” or a “sales proposal generator for a specific manufacturing company.”
In other words, the generic level describes the type of problem, and the company-specific level defines how that problem appears in a particular organization, with its own data, language, and processes.
To support this transition, GAIK uses small reusable building blocks such as parsers, transcribers, information extractors, vector databases, report writers, and security modules. These elements can be combined into larger components that match a company’s existing architecture and deployment choices. Therefore, organizations do not have to rebuild everything from scratch for each new use case.
On top of this, the toolkit introduces workflow blueprints and workflow templates. A workflow blueprint is a diagram expressed in business language that shows the main steps of a process, for example, “collect observations → structure information → generate report.” A workflow template is the executable form of that blueprint, implemented in a workflow tool or orchestration framework so it can actually run in production.
A solution configuration wizard, based on agentic AI, ties everything together. It offers a guided process that starts from business requirements and gradually leads to a runnable solution by helping users specify solution requirements, suggesting suitable blueprints, selecting appropriate workflow templates and components, and preparing deployment packages. This turns abstract ideas into concrete, deployable AI solutions in a structured and repeatable way.
In practical terms, a company might use the toolkit like this:
- Assess AI readiness and choose a promising knowledge-related use case.
- Use the business layer to describe objectives, data, and value.
- Let the configuration wizard analyze requirements, propose or create a blueprint and corresponding workflow templates, and select building blocks through code or no-code configuration.
- Deploy a first pilot, then refine it using the evaluation methods provided in the toolkit.
Over time, the company can build a portfolio of GenAI solutions that reuse the same components and patterns. For example, the same speech-to-text and information extraction building blocks might support both incident reporting and sales-call summarization, just arranged differently.
This approach does not remove the need for technical skills entirely, but it reduces the depth and breadth of skills required for each new project. It also gives non-technical stakeholders a clear way to participate in design and governance, which is critical for adoption.
Opportunities and key takeaways
AI is no longer limited by algorithms or cloud capacity. It is limited by skills, integration, and the ability to turn generic technology into specific, maintainable solutions. Skills gaps are now one of the most frequently cited barriers to AI adoption, especially for SMEs that cannot afford large AI teams.
Generic tools like ChatGPT and Copilot are useful starting points, but cannot, on their own, handle complex, domain-specific workflows or strict privacy and integration requirements. Code-based frameworks give maximum control but demand expert teams. Low-code tools lower the barrier but come with constraints on customization, scalability, and long-term flexibility.
GAIK Toolkit offer a different path. By focusing on knowledge processes, reusable building blocks, and guided configuration, the GAIK toolkit helps companies move from scattered experiments to structured, repeatable AI solutions in their own context.
Key takeaways
- The main bottleneck is not access to AI models, but skills and integration capacity.
- Generic tools are helpful, but custom, workflow-integrated AI is where sustainable value is created.
- Code-based and low-code frameworks both have strengths and weaknesses. Organizations need a structured way to combine them.
- GAIK aims to provide that structure, helping companies configure their own knowledge-management use cases instead of starting from a blank page every time.
For business professionals, the message is simple: do not stop at “trying AI tools”. Start thinking in terms of knowledge processes, reusable patterns, and a long-term portfolio of AI solutions. Toolkits like GAIK are being built exactly to support that shift.
The GAIK toolkit is currently under development, and the first version will be publicly released in the first quarter of 2026. The interested companies and the early adopters are advised to follow
References
Khan, U. A., Kauttonen, J., & Kudryavtsev, D. (2025a). AI adoption in Finnish SMEs: Key findings from AI consultancy at a European Digital Innovation Hub. In 2025 IEEE 23rd World Symposium on Applied Machine Intelligence and Informatics (SAMI) (pp. 465–470). IEEE. https://doi.org/10.1109/sami63904.2025.10883271
Khan, U. A. (2025, 2025b). The 2025 state of AI readiness in FAIR customer companies: The case of Finland. Finnish AI Region (FAIR). https://www.fairedih.fi/en/2025/08/25/the-2025-state-of-ai-readiness-in-fair-customer-companies-the-case-of-finland/
IBM. (2024). AI skills gap. IBM Think. https://www.ibm.com/think/insights/ai-skills-gap. Access date: 24 November 2025. IBM
DigitalDefynd. (2025). 20 pros and cons of low code / no code AI development. DigitalDefynd. https://digitaldefynd.com/IQ/pros-cons-of-low-code-no-code-ai-development/. Access date: 24 November 2025. DigitalDefynd Education
EIT Manufacturing. (2025). How SMEs can harness AI: Key skills for growth. EIT Manufacturing. https://www.eitmanufacturing.eu/news-events/news/how-smes-can-harness-ai-key-skills-for-growth/. Access date: 24 November 2025. EIT Manufacturing
Khan, U. A., Kudryavtsev, D., Kauttonen, J., Joutsenniemi, A., Leskinen, A., Remes, J., Yangarber, R., Pivovarova, L., & Wu, Y. (2025c). Evaluating generative AI technology choices and software frameworks for developing AI solutions in business. Proceedings of the 26th European Conference on Knowledge Management.
McKinsey Global Institute. (2025). Tech faces a talent bottleneck: Here is what to do about it. McKinsey & Company. https://www.mckinsey.com/mgi/media-center/tech-faces-a-talent-bottleneck-heres-what-to-do-about-it. Access date: 24 November 2025. McKinsey & Company
MIT. (2025). The GenAI divide: State of AI in business 2025 (reported summary). Times of India technology section. https://timesofindia.indiatimes.com/technology/tech-news/mit-study-finds-95-of-generative-ai-projects-are-failing-only-hype-little-transformation/articleshow/123453071.cms. Access date: 24 November 2025. The Times of India
Pandium. (2024). The hidden limitations of low code and no code integration platforms. Pandium. https://www.pandium.com/blogs/the-hidden-limitations-of-low-code-and-no-code-integration-platforms. Access date: 24 November 2025. pandium.com
Umair Ali Khan
Senior Researcher
Finnish AI Region
+358 294471413
umairali.khan@haaga-helia.fi
Finnish AI Region
2022-2025.
Media contacts