Erik Grönroos: Making Sense of Data Before It Makes Decisions for Us

Brings clarity, restraint, and human thinking to analytics and AI!

Every digital product today generates data, yet very few organisations truly understand what to do with it. Dashboards grow crowded, tools multiply, and numbers move faster than meaning. In this space, Erik Grönroos offers a clear solution: use data with intention, test ideas carefully, and apply analytics only where it genuinely improves decisions and experiences.

Erik works as Chief Analytics and AI Engineer at the Finnish digital consultancy 8-bit-sheep, where he helps organisations solve complex digital challenges with calm judgement rather than blind adoption of technology. His work focuses on improving user experience, usability, and measurable business outcomes by grounding analytics and AI in practical reality. For him, data earns its place only when it creates clarity.

His interest in technology began early. As a child, he spent time with early home computers such as the Commodore 64 and Amiga 500, learning how systems respond to human input. That curiosity deepened in the mid-nineties, when he built his first website using a computer at a local library. That experience revealed how the internet shapes behaviour, choice, and interaction, and it set the direction for his career.

Over more than twenty-five years, Erik has worked across web development, analytics, business and process development, data science, and machine learning. He has led analytics and AI initiatives in both consulting and corporate roles, always returning to one belief. Data should support decisions, not overwhelm them.

At 8-bit-sheep, founded in 2017 as a digital and data strategy boutique, Erik helps strong brands integrate data into products and operations in ways that feel natural and useful. The company draws inspiration from both technology and culture, reflecting the idea that logic and creativity work best together. Its mission centres on delivering high-quality solutions for clients in Finland, the Nordic region, and wider markets.

Erik also contributes to the broader analytics community. He is the founder and editor-in-chief of Analyytikko magazine, Finland’s first expert publication focused on analytics, data science, and machine learning. He maintains the widely used R package googleAnalyticsR, which allows professionals to work more effectively with analytics data.

What defines Erik’s approach is restraint. He believes AI can solve some problems well, while others require traditional analysis, automation, or careful experimentation grounded in statistics. By promoting a data-driven experimental culture, he helps organisations choose tools with purpose.

Through his work, Erik persists in showing that good analytics begins with understanding people, and that progress comes from knowing when to measure, when to test, and when to keep things simple.

Human Values in a Data-Driven World

At the heart of Erik’s work lies a belief that technology should never exist in isolation from human values. That philosophy is reflected in the identity of 8-bit-sheep itself, a company shaped by the idea that meaningful progress happens when art, logic, and ethics coexist. The name draws inspiration from Philip K. Dick’s Do Androids Dream of Electric Sheep, symbolising the tension and balance between humanity and machines.

This mindset was articulated early on by the company’s co-founder, Sami Kallinen, who once described himself as, “I’m a humanist that dreams of 8-bit sheep.” The statement captures the spirit that continues to guide the company’s direction.

Founded in 2017, 8-bit-sheep was established as a Digital and Data Strategy Boutique with a clear purpose: to help strong brands solve complex digital and data-related business challenges. The firm operates as a consulting agency focused primarily on data-driven business problems, with artificial intelligence treated as one component rather than a universal solution.

A defining characteristic of Erik’s approach is restraint. AI is never positioned as the answer by default. Instead, the team evaluates whether AI is appropriate at all, recognising that in many cases traditional analytics, data automation, or established data science methods deliver better outcomes. This ability to distinguish between what is technically possible and what is practically valuable has become one of the company’s core strengths.

8-bit-sheep works with leading brands across Finland, the Nordic region, and international markets, helping them embed data directly into products and operations. The objective is not experimentation for its own sake, but measurable improvement in customer experience and long-term competitive advantage.

Privacy Before Performance

As AI adoption accelerates, Erik has consistently emphasised that data legality and privacy must come before speed or scale. One of the most significant risks associated with generative AI is the ease with which sensitive information can be exposed to third parties, often without organisations fully realising it.

This risk is particularly severe for EU-based companies. When data is transferred outside the EU or EEA, frequently to the United States, where many popular generative AI services are hosted, organisations may unintentionally violate GDPR requirements. These are not theoretical risks but operational ones, with real legal and reputational consequences.

To address this, 8-bit-sheep focused on building controlled infrastructure rather than relying solely on external platforms. As part of this effort, the company’s AI Engineer, Mark Edmondson, developed an internal AI platform called Sunholo Multivac. Designed as a GenAI and LLMOps cloud infrastructure, it allows AI solutions to be deployed locally or within specific regions of choice.

Sunholo Multivac was built to solve what Erik often describes as the “last mile” problem in AI, moving applications from experimentation into secure, production-ready environments. This approach reflects the company’s broader strategy: preparing for growing AI demand while maintaining a strict privacy-first mindset. Data protection is treated as a foundational requirement, not a trade-off.

The Uncomfortable Questions Around AI

The public release of ChatGPT marked a turning point in how society interacts with AI. It removed barriers to entry and brought advanced language models into everyday use almost overnight. While this shift unlocked new possibilities, it also exposed unresolved challenges that Erik believes cannot be ignored.

Intellectual property is one of them. Generative AI systems are trained on vast amounts of human-created content, much of it protected by intellectual property rights. At the same time, these systems function as aggregators of human knowledge, raising difficult questions about where the line should be drawn. Erik has noted that humans routinely learn from protected content without legal issue, which complicates the argument when machines perform similar processes.

Ethical concerns weigh even heavier. Generative AI systems can unintentionally reinforce cultural, social, and democratic biases. In more serious cases, they may even support harmful or self-destructive thought patterns. These risks extend beyond misuse and point to structural issues in how models are trained and deployed.

Misinformation represents another critical threat. As AI-generated responses become more fluent and authoritative in tone, people increasingly trust outputs without verification. This becomes especially dangerous in analytics-driven environments. When conversational AI tools are used to inform business decisions without expert interpretation, false or misleading results can directly lead to harmful outcomes.

Speed Without Discipline

The rise of AI-assisted development tools has dramatically lowered the barrier to creating software. So-called “vibe coding” tools enable rapid prototyping and experimentation, and Erik recognises their value when used responsibly. In the hands of experienced developers, these tools can accelerate ideation and even contribute to production-ready systems.

The risk emerges when speed replaces skill. When individuals without sufficient programming experience deploy AI-generated code directly into production environments, the result is often insecure, unstable, and vulnerable systems. In such cases, the issue is not the tool itself, but the absence of review, governance, and process.

In response, Erik has shifted focus away from model comparisons and toward workflow control. Emphasis is now placed on structured processes, feedback loops, and experience-driven frameworks.

Tools such as Ralph Wiggam Loops and Antigravity have shown promise not because they are novel, but because they support disciplined, repeatable practices. The centre of gravity has moved from chasing models to building systems that manage risk.

Perspective on Intelligence and the Road Ahead

Despite how convincing modern AI systems can appear, Erik remains clear about their limitations. Human intelligence operates on a vastly different scale. The human brain contains roughly 90 billion neurons, while even the largest AI models today operate with only a fraction of that capacity. This gap underscores how far artificial systems remain from true machine intelligence.

At the same time, there is no illusion about the direction of travel. The pace of development makes reversal unrealistic. AI systems used in the latter part of this decade will be significantly more advanced than today’s models. Erik expects rapid iteration, with future versions of leading models evolving far beyond current capabilities.

This combination of realism and foresight defines his outlook. AI is neither dismissed nor romanticised. It is treated as a powerful tool that must be handled with technical depth, ethical awareness, and long-term responsibility.

Leadership Built on Technical Depth

Erik’s leadership style has been shaped less by titles and more by years of hands-on technical work. His background as a software developer goes back over two decades, beginning with JavaScript at a time when the language was still evolving.

That early choice proved durable. Today, JavaScript remains embedded in the vast majority of websites worldwide, making it a natural foundation for digital analytics and data-driven systems aligned with its expertise.

From JavaScript, his work expanded into PHP, where early analytical tasks were carried out using a combination of server-side scripting and Microsoft Excel. As data volumes increased, the limitations of spreadsheet-based analysis became impossible to ignore. Scaling challenges forced a shift toward more suitable tools, leading him to R, a language he has continued to use extensively ever since.

This progression formed a strong belief that leadership in technical domains requires firsthand experience. Having built applications, solved production-level problems, and worked across both business and software layers allows him to understand issues deeply rather than abstractly. In his view, it is difficult to guide complex initiatives without having personally navigated the details that underpin them.

Problem-solving has always been a central driver. From an early age, intellectual challenges held his attention. Exposure to logical reasoning and structured thinking in childhood laid the groundwork for a long-standing interest in data, mathematics, and systems. Data and AI naturally followed, not because they were fashionable, but because they presented difficult problems with measurable outcomes.

One defining moment came with the first successful predictive model he built. Seeing future outcomes predicted with meaningful accuracy demonstrated the tangible power of data-driven methods. While no model is flawless, achieving high levels of predictive reliability reinforced the practical value of disciplined modelling and careful training.

How Success is Measured

For Erik, success is not measured by visibility or marketing metrics. It is reflected in client behaviour. When clients return with additional projects without prompting, it signals that value has been delivered. Repeat engagement is treated as the most honest indicator of impact.

Direct feedback mechanisms support this perspective. Periodic Net Promoter Score surveys provide another layer of validation. Achieving a consistent score of 100 reflects trust built through reliability, clarity, and outcomes rather than promises.

Across industries, AI use cases continue to expand. Healthcare organisations use AI to support surgical planning, while space agencies apply it to identify previously unseen objects in the universe. These examples demonstrate AI’s potential when applied with purpose and domain expertise.

At the same time, Erik remains cautious about emerging trends such as Agentic AI. While promising, these systems are still in early stages and lack the reliability required for serious deployment. Their evolution, however, is expected to unlock new creative possibilities as maturity improves.

The Cultural Gap Holding AI Back

Despite technological advances, the primary barrier to effective AI adoption is rarely technical. Erik consistently observes that the real challenge lies in organisational culture and leadership commitment. Many companies approach AI as a series of isolated experiments rather than as part of a long-term data strategy.

Without sustained investment in data capabilities, governance, and usage, AI initiatives fail to deliver meaningful returns. Successful organisations identify concrete business cases, develop data infrastructure patiently, and commit to continuous improvement over years rather than months.

Experimentation still plays an important role, but it must be systematic. In data-driven environments, meaningful insights often emerge only after extensive testing. Large-scale A/B testing programmes, sometimes involving hundreds of experiments annually, are necessary to uncover a small number of genuinely valuable use cases.

Teams remain motivated when leadership treats data responsibilities seriously. When commitment fades, talent follows. Retaining skilled data professionals depends on whether organisations demonstrate long-term intent rather than short-lived curiosity.

A Flat Model by Design

8-bit-sheep operates with a deliberately flat structure. There is no Chief Executive Officer. Instead, the organisation functions as a single-level model where consultants operate independently as solopreneurs, forming small teams when necessary.

This structure keeps focus where it matters: client work. Consultants integrate directly into client development teams, contributing as peers rather than external advisors. Autonomy is paired with accountability, creating an environment where expertise carries more weight than hierarchy.

Equality, diversity, and ethics are not treated as abstract values. They influence internal operations, client selection, and data and AI development practices. A formal due diligence process ensures that projects do not include unethical features or applications that conflict with these principles.

Despite being recognised as a data and AI expert, Erik consistently resists positioning AI as a universal solution. In his view, AI is one tool among many. Overuse or misapplication can be as damaging as underuse.

The focus remains on advancing the client’s business, not promoting AI adoption for its own sake. When AI is unnecessary or counterproductive, he actively advises against it. This willingness to push back is rooted in responsibility rather than scepticism.

The analogy he often uses is simple: no one builds an entire house using only a hammer. The same logic applies to AI. Effective solutions require choosing the right tools for the right problems.

Word of Wisdom – In His Words

It depends so much on your goals and industry, but if you’re aiming for a tech company, my advice is this: First, learn to code and solve software problems for a few years. This is because this way you will gain a basic understanding of how applications and systems work, and how they do and don’t provide data.

Then, study and practice Statistics to understand how numbers work. This took me 5 years, so be patient and persistent. Once you have these fundamentals in hand, it’s time to start modelling your data and experimenting with Machine Learning and AI models. I prefer R, but most people prefer Python.

In the end, it’s all about just choosing a language, French or German? It’s your call. Don’t listen to people who say R can’t be used in production; it can and is used. And my most important advice is, always be nice and easy to work with, always.