Skip to content

Why Off-the-Shelf AI Isn’t Enough: Lessons from AtoZ GPT

Our year-long AtoZ GPT development project has now been completed. Before going any further, I want to thank all the #AtoZfellows who contributed to the project, as well as Technology Industries of Finland for partially funding it. The project required many of us to learn new things and, more importantly, to challenge our own assumptions. We also deviated slightly from the original plan along the way. Nevertheless, the outcome provides us with a very strong foundation for delivering AI solutions to our customers.

Even before the project officially started, we already had an internal version of AtoZ GPT in use. It was trained on our own materials and helped streamline everyday work for our employees and HR. In particular, it sped up information retrieval, reduced repetitive questions, and supported our experts in finding answers in daily situations.

The Challenges of AtoZ GPT 1.0

The biggest issue with the original AtoZ GPT was that the solution relied heavily on third-party components and services. As a result, we could not genuinely guarantee data security or the long-term sustainability of the solution. For internal use, this was just barely acceptable, but even then we constantly had to consider what kind of information could be safely fed into the system and what could not.

When data security cannot be fully guaranteed, usage inevitably becomes cautious. It also quickly became clear that such a solution could not be offered to customers as-is. This realization became the true starting point of the entire development project. If AI is to be used responsibly and at scale, its foundation must be fully under your own control.

From an Internal Tool to a Platform for Customer Projects

Over the course of the year-long development project, we did not end up building a new off-the-shelf product. Instead, we built a platform. A foundation on top of which very different AI implementations can be created to meet real customer needs.

AtoZ GPT is not a push-button, ready-made solution. It is a technical and operational foundation that enables the use of AI in areas such as information management, expert work, quality assurance, and other data-intensive processes. This approach provides flexibility and, most importantly, ensures that each solution is built to fit the customer’s specific context.

A Platform-Based Approach Saves Time and Money

In practice, a platform-based approach means that customers do not have to pay for the same groundwork over and over again. Document processing, indexing, vector search, language model utilization, user interfaces, and a secure base architecture are already in place and ready to be tailored.


room Room based data input

When these fundamentals are done properly, the focus can shift to what truly creates value. Are we solving information discovery challenges, improving quality, or supporting expert decision-making? Building on a solid foundation is lighter, faster, and significantly more cost-effective than starting from scratch every time.

Security and Cost Efficiency Go Hand in Hand

Data security was a starting point throughout the development process, not something added afterwards. We built AtoZ GPT so that language and embedding models can be run flexibly either in AtoZ’s own environments or in the customer’s infrastructure and hardware. This makes it possible to choose the most suitable setup for each use case without forcing everything into the cloud.


Hallusinaatiovapaa käsittely ja käyttöliittymädemonstraatio

In practice, this means two things for the customer. Data remains firmly under control, and costs become more predictable. Not every AI solution requires continuous cloud computing to work effectively. With a well-designed architecture, models can be run locally when it is sensible and secure, while cloud resources are used only when they provide clear added value.


Esimerkkinä datan anonymisointi - Tietokannasta LLM putkenläpi työaikaraportoinnin analyysiin


Flexibility as a Key Enabler

Flexibility is one of the main reasons why AtoZ GPT serves as a strong foundation for customer-specific AI initiatives. The solution adapts to the customer’s reality, not the other way around.

The Impact of the Project on AtoZ and Our Customers

Over the past year, AtoZ’s AI capabilities have grown significantly. We now have a controlled, secure AI architecture and hands-on experience from real implementations built on top of our platform. Going forward, AtoZ GPT will serve as the foundation for customer projects and enables us to deliver AI solutions faster, more reliably, and with greater confidence than before.

What’s Next?

The AtoZ GPT project has reached its conclusion, but the real work starts now. We will leverage the platform more broadly in our projects and continue developing it based on real-world use cases. We will also be launching a dedicated AtoZ GPT section on our website soon, where you can learn more and book a demo.

If you are considering how to leverage AI and want to avoid heavy and expensive experiments, let’s talk. Often, a single conversation is enough to clarify where AI makes sense and where it does not. That alone can save time, money, and quite often a fair amount of frustration.