How to adopt AI in your company: from Discovery to first tangible results

There’s a question that comes up frequently in the projects we work on: “How can we use AI?”

It’s a legitimate question. The market is moving fast, competition is intense, and artificial intelligence is already present in processes, products and customer expectations. And yet, in most cases, this question starts from the wrong place.

Because the real issue is understanding whether the organization is truly ready to transform its data and information into systems capable of generating useful and measurable value.

The most common mistake: focusing on algorithms instead of data

When companies talk about AI adoption, the focus is usually on tools:

• which model to use
• which platform to integrate
• which AI feature to add to the product
• which automation to implement

That’s understandable: tools are tangible, easy to communicate and often become the center of the narrative.

But a sophisticated model built on fragmented or inconsistent data produces unreliable outputs. A predictive system trained on incomplete datasets generates forecasts that may look credible, but aren’t. An AI-supported decision-making process that ignores the quality of input information does not optimize anything, it simply amplifies existing errors.

And before being a technical issue, this is a strategic one.

The real question: how mature is your data ecosystem?

When we work with organizations looking to integrate AI into their products or processes, the first step is not evaluating which model to use. It’s understanding the maturity level of their data ecosystem.

Here are some key questions that help frame that assessment.

What data are you collecting today, and how structured is it?

Excel files, disconnected legacy systems, information scattered across emails, CRMs and different tools: this situation is far more common than it seems. It is the real starting point for many organizations.

Understanding this is essential because it defines what is realistically possible.

Does the data you collect reflect what you actually want to optimize?

Many organizations collect the data that is easiest to gather, not the data that is truly useful for decision-making.

A company trying to reduce churn may have detailed transactional data but no structured information about actual product usage behavior. AI cannot compensate for that absence.

Is there a clear data governance process?

Who is responsible for data quality? How are duplicates, anomalies or inconsistent versions managed? Any AI system built on weak governance will inevitably be fragile.ì

How quickly does your business data change?

The frequency with which a model needs to evolve depends on how quickly the reality it describes changes. In fast-moving markets, a system trained on outdated data can quickly shift from competitive advantage to operational limitation.

AI is an evolving system

The path toward AI adoption is almost always, first and foremost, a process of data maturity.

Organizations that achieve tangible results with AI are those that:

• understand what they want to optimize
• collect information intentionally
• treat data as a continuous asset
• build measurable systems
• connect AI, processes and outcomes

A concrete example of this approach is AUTOMA, a project designed to support the monitoring of invasive marine species.

The initial challenge was highly operational: the growing number of invasive species made it increasingly difficult for researchers to verify and analyze large volumes of images and videos coming from different sources. In this case, the goal was not simply to “add AI,” but to build a system capable of evolving over time together with the collected data.

For this reason, we started by designing an evolutionary AI roadmap based on the progressive growth of datasets.

The first phase focused on developing a Machine Learning and Computer Vision pipeline to accelerate invasive species recognition, reduce researchers’ manual work and improve image analysis. The model was trained on datasets directly validated by the researchers themselves.

But the most important aspect is that the architecture and data collection processes were designed from the beginning to support future evolution.

Once a significant amount of historical data is consolidated, the platform will be able to evolve toward predictive models capable of identifying high-risk areas and supporting preventive decision-making. In a later phase, the goal will be to introduce generative AI tools to allow researchers to quickly query large volumes of scientific data collected over time, simplifying research and consultation.

Looking even further ahead, the system could integrate intelligent orchestration of IoT devices and underwater robotic systems, depending on the evolution of the operational infrastructure and hardware context.

And this is the key point: AI was not treated as an isolated feature, but as an evolving system built around the collection, quality and interpretation of data.

Because the real source of value is not the model itself, but the system’s ability to grow, adapt and generate increasingly useful insights over time.

From discovery to first tangible results

Talking about AI adoption also means talking about discovery, processes, governance, data quality, interoperability, decision-making capabilities and measurability.

And this is where many AI initiatives stop: not because the technology does not work, but because there is no structure capable of sustaining it over time. For this reason, in our projects, the starting point is never simply “implement AI”.

It is understanding:

• where the data is located
• what shape it takes
• how reliable it is
• which processes it describes
• which decisions it could improve

Only then does it make sense to design systems, workflows or products that integrate AI components. That is where AI can deliver its best results:

• improving operational visibility
• supporting faster decisions
• reducing complexity
• helping prioritize actions
• enabling sustainable automation
• making systems more adaptive

At 20tab, we work with organizations at different stages of digital maturity.

Some already have solid infrastructures and are looking for the most effective way to integrate AI into their products. Others are still building systems capable of collecting, interpreting and understanding the data they already generate every day.

In both cases, our approach starts from the same principle: understanding the context before the technology.

Because the goal is not adopting AI for the sake of following the market. The goal is building digital systems capable of generating real, sustainable and measurable value over time.