Catalyst? Products with AI

The Rise of AI: From Data to Value

The rise of Artificial Intelligence (AI) is an irreversible shift that is fundamentally turning the product world upside down. It is a complex phenomenon with both enormous benefits and significant points of concern. In our previous whitepaper, we looked at what this means for the Product Owner (PO); today, we take a different perspective: what can a product achieve when it is "powered" by AI?

Nothing New Under the Sun?

Although not a day goes by without a news article about how AI has changed the world, using (large amounts of) data to create value is actually an existing phenomenon. In the beginning, it was quite simple: data was converted into information via logging and reporting, primarily describing what had happened. Soon after, conclusions could be drawn, and machines began to explain why it happened. Humans took in the input, made a decision, and then acted to create value.

‍

AI only enters the frame later, when prediction becomes possible alongside analysis and classification. It started with chess computers, but soon it could detect fraud and make recommendations ("Other customers also bought..."). This was achieved through decision trees, SVMs[2], and neural networks—and yes, this was already possible in the previous millennium. The downside was that datasets had to be neat, structured, and labeled, and the AI was very task-specific, such as a self-driving car.

GenAI is special in the sense that it can generate new content (hence GenAI) based on what it has learned. Essentially, the thinking was: "What would happen if we threw as much electricity and money at AI as we could find?[3]" What resulted was an AI that could recognize patterns in unstructured data. Whereas "classic" AI required data scientists, machine learning experts, and other scarce knowledge, suddenly everyone could work with the system thanks to Natural Language Processing (NLP).

This makes it possible for systems to take decisions and adapt themselves without human intervention. In terms of "Agility," product people have been asking for 20 years: "Does being agile actually make us faster?" We are now swapping that for: "This is going too fast—how do I keep up?"

The agile wave of change over the last 20 years primarily aimed to respond more quickly to new market insights. Pre-1990 processes tried to prevent change, while Agile tried to put people at the center to better react to new insights. But what happens if the human disappears from the chain?

AI as Part of a Product

AI has an enormous impact on creating new products along two dimensions. Firstly, it influences how we make new products. In this sense, AI can be a catalyst for Product Owners. (See the whitepaper "AI for Product Owners: Finding the Balance," where we use the Oxygen model to look at how the world of POs and product organizations is changing).

The second dimension covers what happens when GenAI becomes an integral part of your product. To be clear: we aren't talking about "classic" AI here—that has since become "standard."

For ASITO, for example, we integrated a classic AI with a chat interface. Employees can ask questions in natural language about collective labor agreements (CAO), vacation days, etc., and the system answers in understandable language via WhatsApp. What it doesn't do is generate new employment conditions, which would, of course, be undesirable.

As Figure 1 shows, GenAI in particular offers the opportunity to develop a new kind of product that grows autonomously with the data in multiple dimensions. While a classic recommender grows with the product dataset, a GenAI product develops in dimensions you cannot yet predict—that is the gain, and also the risk.

What Does This Mean for Value?

It is likely that future software products (read: all products) will increasingly contain AI components. Simplified, the model will look something like this:

‍

About 10 years ago, cloud computing became the dominant standard. Driven largely by COVID, running one's own hardware became less common, meaning scaling performance and costs were suddenly linked to usage. We are now on the threshold of a similar opportunity regarding interpreting data and making decisions based on it.

The "Model" layer will likely consist of various models working together to provide an answer. General AI is currently a Large Language Model (LLM), which is expensive to train—costing millions. The chance of this being an in-house development is virtually zero, though there are developments in Small Language Models (SLM)[4]” On top of that, domain-specifically trained models will provide context (think of a legal publisher with a vast amount of specific context[5]), followed by localization of the problem domain.

The application the end-user sees benefits from this through faster systems (as the system grows "automatically"), performance (the AI layer can adapt to usage), security (the system can be automatically patched), and user-friendliness (natural language is becoming increasingly intuitive). General systems with an embedded AI component will likely perform much better than their traditional counterparts.

Implications and Challenges

"Every advantage has its disadvantage," as a famous Amsterdam philosopher once said. Traditional software-dominant products are generally deterministic in nature. That is: the Product Owner has an idea, it’s translated into a requirement, something technical happens, and the idea lands in the hands of the customer[6].

‍

With non-deterministic products, a large portion of the features and data are generated by the parameters the PO sets (context) plus what is inside the model.

This raises the question of how we will test this and guarantee the quality of underlying components. These are self-learning and act in real-time. The complex interactions within non-deterministic components make this even more complicated. For example, when generating legal or clinical advice, it is desirable for the product to give increasingly better answers, but simultaneously unacceptable for it to give incorrect answers.[7].

What Does This Mean for the Product Owner/Manager?

As a PO or PM, you are at the start of this value chain, meaning several things will fundamentally change for you and your product development:

• Requirements: Specifying requirements in advance or testing them afterward becomes harder and partially impossible because the product’s behavior can change as it learns more[8]‍
• Role & Responsibility: If the product reacts to the user and adapts itself, what is the role of the PO? Who is responsible? What does this mean for prioritizing work?‍
• Compliance: This is now a real-time phenomenon. Where we are compliant now, we might not be in a second.‍
• Team & IP: If your teams "consist" of 50% or more AI, how do you handle turnover and intellectual property? How can I remain distinctive if I use the same "employees" as my competitor?‍
• Costs: Usage is currently subsidized, and costs are not proportional to revenue. What happens to the price when the market "shakeout" is over and only a few players remain[9]?

However, it’s not all "doom and gloom." AI-powered teams will undoubtedly be a catalyst:

• The speed from idea to working feature will be exponentially faster[10].
• The merger between UX and PM will accelerate as ideas can be tested with real users faster than ever and at almost no cost.
• All that collected data (user, log, product data, etc. [11]) will finally be used to make good decisions automatically.
• Domain models will eventually be standardized, making domain-specific applications cheaper to use.
• The gap between those with the idea and those who can realize it is shrinking, reducing information loss.

Conclusion: Explorers Wanted

The future of product development is in flux, and that offers opportunities. While much is uncertain, sitting still is not an option. You must do "something" with AI, which means searching for potential treasures like an explorer, while keeping an eye on the risks. As a Product Owner, the key to success in the AI era is finding the balance between AI-driven speed and human insight.

‍

‍

‍