Artificial intelligence remains one of the most widely discussed topics in the context of digital transformation. Yet many organisations still begin the conversation in the wrong place – by asking: “Which AI tool should we implement?” instead of “Which business problem are we trying to solve?”
This shift in perspective is more important than it may initially seem. The mere availability of technology does not guarantee that an organisation can use it safely, efficiently and in a way that aligns with its business objectives.
Drawing on the experience of practitioners specialising in AI transformation, solution scaling and AI governance, this article outlines the key questions organisations should answer before an AI implementation starts.
Don’t start with technology – identify the business problem first
A few years ago, organisations were still debating whether to invest in AI. Today, the question is often how quickly they can implement it. The problem is that the latter assumes the decision has already been made.
In practice, this is often the moment that determines the future success or failure of an initiative. If the AI conversation begins with selecting a tool, model or vendor, organisations can easily rush into action before answering the most important question of all: what business problem are we really trying to solve?
As Łukasz Migda, a digital transformation and AI leader with more than 20 years of experience, explains:
The biggest mistake is not experimenting with technology itself. The problem arises when organisations fail to connect technology with a specific business process, a business owner and a measurable outcome.
Migda describes this pattern very clearly:
Maurycy Moczulski, mathematician, deep learning specialist and Co-founder & CEO of GradientHouse, shares a similar perspective:
Before choosing a model, vendor or platform, organisations should first ask where technology can help solve a problem that is already visible in business performance, processes or customer and employee experience.
This is precisely the purpose of a well-structured Discovery process [linkowanie wewnętrzne]. At Infinity Group, we use it to align business objectives, processes, data and potential implementation scenarios. This enables organisations to determine not only how to implement AI, but more importantly whether AI is the right solution in the first place.
Why do so many AI pilots never make it into production?
One of the greatest paradoxes of modern AI initiatives is that many projects achieve their technical objectives and still never reach production.
Krzysztof Goworek, an advisor helping organisations transition from technology demonstrations to production-grade AI deployments, points out that pilots are often created in environments disconnected from everyday business reality:
This is why a proof of concept usually answers the question, “Does it work?”, but not “Will it work reliably, securely and cost-effectively once deployed?”
Goworek highlights several common reasons why projects stall:
Maurycy Moczulski points to a similar issue:
Another challenge is underestimating the cost of scaling. As Moczulski explains:
This is particularly evident with LLM-based solutions. What works for a small group of users may generate entirely different costs at production scale. Without considering these factors upfront, organisations risk finding themselves in a situation where technological success does not translate into a viable business case.
Moczulski emphasises that the quality of the data used in the pilot project is also significant:
Often, however, the problem lies neither with technology nor data. The organisation simply failed to define what success looks like. As Migda observes:
Moving from pilot to production is therefore less a test of technology and more a test of whether the organisation has genuinely prepared for implementation rather than demonstration.
How can organisations scale AI without losing control?
If so many AI initiatives stall at the pilot stage, it is worth asking the opposite question: what do organisations that successfully scale AI do differently?
According to Krzysztof Goworek, the sequence of actions matters greatly:
This is an important observation because many companies want to begin their AI journey with the most visible and impressive solutions. Mature AI implementations often start in a much less glamorous way: by organising internal processes, improving data quality and establishing decision-making frameworks.
This approach allows organisations to gradually build the capabilities, processes and data assets required for more sophisticated initiatives. Rather than striving for spectacular results from the very first implementation, organisations develop a sustainable ability to leverage AI over time.
Goworek also stresses the importance of clear ownership:
As AI adoption expands, control becomes increasingly important. In many organisations, AI adoption starts from the bottom up: employees begin using ChatGPT, Copilot and other generative AI tools to speed up their daily work. They create automations, workflows and informal AI agent experiments. This phenomenon, known as shadow AI, has evolved significantly. As Goworek explains:
The instinctive response is often to prohibit such behaviour, but Goworek believes this approach is doomed to fail:
At the same time, governance cannot remain a theoretical exercise:
Governance mechanisms should therefore be introduced from the very first implementations involving production data, rather than only after a project becomes successful.
What really determines AI success?
Ultimately, it is worth returning to the question that has surfaced throughout this article: why do some organisations remain stuck in the pilot phase while others unlock AI’s real potential?
The answer rarely lies in technology alone. Migda emphasises that successful AI implementation also requires the ability to guide people through organisational change:
Companies that do not embrace experimentation and learning from failure will struggle to build long-term AI capabilities. As Migda bluntly puts it:
Goworek, meanwhile, highlights the importance of developing competencies that go beyond simply using AI tools. He refers to this as AI fluency.
Increasingly, competitive advantage stems not from access to technology itself, but from an organisation’s ability to test hypotheses, measure outcomes and learn quickly.
Summary: Seven questions to ask before launching an AI initiative
If your organisation is considering AI, start by answering these seven fundamental questions. They will provide a far more accurate assessment of organisational readiness than evaluating tools or technologies alone.
- What specific business problem are we trying to solve? Can we describe it in terms of processes, costs, revenue, customer experience or operational efficiency?
- How will we know the project has succeeded? Have we defined measurable KPIs and a timeframe for evaluating them?
- Is AI truly the best solution? Could the same outcome be achieved through automation, process redesign or better use of existing tools?
- Who will own the initiative from a business perspective? Is there a person or team responsible not only for launching the project, but also for its adoption and long-term development?
- Do we have the right data to validate our assumptions? Is it complete, representative and available at a scale that reflects real business conditions?
- Is the organisation prepared for process change rather than simply tool deployment? Are we ready for new roles, new skills and new ways of working?
- What happens after the pilot phase? What conditions will determine whether the project is scaled, redirected or discontinued?
At Infinity Group, we have been helping organisations translate business goals into digital solutions for more than 25 years – from Discovery workshops and needs analysis, through solution design, to implementation and ongoing development.
That is why, before discussing specific AI tools or models, we focus on a far more important question: which business problem is worth solving, and which approach makes the most sense in a specific organisational context.
If your organisation is considering AI and you would like to explore potential directions, let’s start a conversation.
