Đọc bản tiếng Việt tại đây: 9 chiến lược AI đã chứng minh hiệu quả – Bài học từ CTO Orient Software
AI is reshaping every industry, but not every company is transforming in the right direction.
A recent report from MIT’s NANDA Project, titled The GenAI Divide: State of AI in Business 2025, uncovers a sobering reality. Despite more than 30 billion USD poured into generative AI, only 5% of AI projects generate real profit, while 95% show almost zero financial return.
The issue, the report suggests, does not lie in model quality, but in how companies approach and deploy AI.
So, how can a business move beyond the pilot phase and achieve meaningful transformation?
To explore what successful adoption truly looks like, ITviec sat down with Øyvind Forsbak, CTO and Co-founder of Orient Software and a member of the Forbes Technology Council. With more than 25 years of experience leading global software and AI/Data projects, Øyvind shared an open and insightful perspective on the common traps businesses fall into during the AI hype, how to spot mistakes early, and how to build an AI strategy that keeps you on the right path – one that drives real, lasting business value instead of short-term hype.
Here are 9 proven AI strategies from our talk with Øyvind:
Start with the problem, not the technology
As AI dominates the headlines with endless promises of transformation, Øyvind believes many companies feel pressured to “do something with AI”, and that’s often their first mistake.
This pressure drives them to rush into technology investments without clearly defining the problem they want to solve or the value they hope to achieve.
“I think it is going to fail in most cases, because even a very strong engineering team if they don’t have direction, they will not be able to solve the problem.”
He shared a real example: the Orient Software team once built a “smart search” feature for a recruitment app. The model performed well in the test environment, but when applied to real data, the results were far from smart. It’s because the AI team hadn’t studied how real recruiters actually search for candidates.
Øyvind called it “a classic example of technology before the problem” – the project didn’t fail because of the wrong tools, it failed because the team asked the wrong question at the start.
He noted that technology waves come and go, but business problems remain:
“If you look back two years ago, everyone was building platforms around RAG or semantic search. But those only solved a small part of the problem. Then the focus shifted to agents, and even to agents talking to other agents, yet these still didn’t deliver the kind of business value people were hoping for.
So if you buy into the hype without applying critical thinking, your solution or pilot won’t be strong enough to create real value. Users won’t adopt it, and the project will simply end there.”
His conclusion was simple but sharp: “AI is just a tool to solve problems.” The right approach, he said, is to identify the business problem first, then explore whether AI, at its current capability, can help solve it effectively.
Prepare your data ready for AI
If strategy is your map, then data is the fuel.
According to Gartner, poor data quality is one of the top reasons most GenAI projects fail right after the pilot stage. Øyvind strongly agrees:
“If you have a good strategy but you don’t have the data to solve your problem, then you will fail.”
He explained that when data is fragmented or of poor quality, AI tries to “fill in the blanks” – inventing information that leads to errors, loss of trust, and zero real value.
So what does AI-ready data mean? Øyvind believes it’s not about building flashy data infrastructure, it’s about centralizing and organizing your data effectively.
How to make your data AI-ready?
Unify your data into one platform
Most companies have a large amount of data in CRM systems, emails, SharePoint, or financial platforms, but everything is scattered. Disconnected data makes it hard for AI engineers to access, train, or secure effectively. Bringing all your data into one unified platform and cleaning it regularly is the foundation of “AI readiness.”
Define clear and secure access rights
One often-overlooked risk of integrating siloed data systems into a unified platform is that AI agents may unintentionally expose information to users who shouldn’t have access. So when combining multiple systems into one platform, access control must be carefully managed.
Use your unique data as a competitive advantage
According to Øyvind, proprietary data has become a strategic asset for creating real business value that competitors simply cannot copy.
“If you have some key data that your competitors don’t have, then you have a business advantage.
Large language models, most of them commodities. You can buy access through an API, and your competitors can do the same. But the data you have is what really differentiates your success from another company’s success. So I think it’s very important that you keep experimenting and learning what your data can actually do for you.”
Invest in data early, don’t wait for the technology
AI is still in its early stages, and it’s evolving fast. Øyvind advised companies to start investing in data now, even if they are not ready to deploy AI yet.
Building a solid data platform takes time and doesn’t show quick returns, but it’s an inevitable step and a long-term investment. When AI matures, those with ready data will move faster and benefit sooner.
Manage expectations in the AI productivity hype
Øyvind also pointed out another common mistake: many C-level leaders have unrealistic expectations about what AI can achieve.
“Many companies say they can double or triple their productivity with AI tools. But in our experience, that’s not true. At this point, we can probably improve things by around 30%.”
He added that leaders shouldn’t expect AI to deliver instant profit either:
“Sometimes the return will not come because the AI industry is not mature enough for what we’re trying to do. It’s changing so fast means it’s not mature.”
Many of the “benchmark” numbers for AI performance come from ideal lab conditions. When businesses believe in those numbers too strongly, they overinvest and later lose faith in AI
That’s why the key lies in managing expectations.
“If you have the wrong expectations, you’ll be disappointed. You might overinvest, and you probably won’t succeed.
But if you set the right expectations and understand the problem you’re trying to solve, you’ll likely allocate the right amount of money, and you’ll also be okay if it doesn’t succeed. Because that’s just part of the learning process, you learn from it, you adjust, and you continue.”
Treat AI as R&D – keep testing and refining
To manage expectations well, Øyvind suggests viewing AI adoption as an ongoing R&D process rather than a one-time project.
“AI is like doing research & developing and adjusting continuously as you get more data.”
Companies need to track both technical metrics (like recall rate and accuracy) and real user experience (adoption, satisfaction, and business impact).
He also emphasized that not every failure is bad. If an AI project solves 70-75% of the business problem, that’s still valuable learning. What matters is keeping a clear goal, an open mindset, and continuous improvement, rather than buying into the hype.
Build vs. Buy: Create agentic workflows, not just buying platforms
There are countless AI tools on the market, but Øyvind warned that many are “just shiny wrappers” over the same large language models.
“Most of them are just wrappers around large language models (LLMs) and provide only shallow value. They might look impressive in demos, but when you apply them to real business problems, the results are usually not as expected, and often, very expensive.”
His three key lessons:
- Don’t build your own AI model from scratch.
Training a model yourself is almost impossible and extremely expensive. Even open-source models require huge resources but deliver limited returns.
“Owning your own model doesn’t really give you a true competitive advantage.”
- Use models through APIs and build your own workflows.
Instead, companies should rent access to models (like GPT or Claude) via API and create their own agentic workflows – AI processes that can handle multiple steps and adapt to specific business needs. - Invest in people more than tools.
AI technology changes quickly, and off-the-shelf tools can become outdated in months. The sustainable approach is to invest in teams that can learn, research, and adapt, because they are the real long-term advantage.
Turning AI anxiety into curiosity
AI can’t run without people, but people are often the hardest part.
In any organization, some are eager to try new things, while others prefer stability and the familiar ways of working. The fear of “AI taking jobs” can also make employees feel unsafe or unwilling to cooperate.
Øyvind shared that even at a tech company like Orient Software, this challenge still exists. The company’s task, he said, is to make AI less intimidating and to help people see it as a partner, not a threat.
He suggested a few practical approaches:
- Share a positive perspective.
Remind people that as tools become more powerful, humans also gain the opportunity to take on higher-value, more creative work. - Encourage collective learning.
Instead of making AI adoption a top-down mandate, turn it into a space to experiment, share results, and learn together. At Orient Software, AI isn’t introduced as “a new technology to learn,” but as a shared journey of exploration across the whole company. - Build critical thinking and soft skills.
According to Øyvind, a good AI engineer needs to communicate like a consultant, understand business logic, and ask why, not just follow instructions. These soft skills are essential for tech professionals to grow sustainably alongside AI.
Measure success by business value, not just technical metrics
Once strategy, data, and people are in place, the next challenge is how to measure success.
From Øyvind’s perspective, a successful AI project should be defined and measured by business value, tied to organizational goals or OKRs, not just technical benchmarks.
He shared a real example:
“In the past, we developed a fairly sophisticated model. But because of poor change management, users kept doing things the old way. Even though they had access to the new approach, they didn’t apply it.
So the problem wasn’t really about the AI model or how good it was, it was about how you implement it in a real business environment.”
Look for “early wins”, but follow through
In the early stages, Øyvind recommends focusing on metrics that are easy to measure yet still show real progress. Here are 2 examples:
- User adoption rate: Even the best tool will fail if people don’t actually use it.
- Process automation impact: For example, how much time is saved or how workflows improve through automation.
These “low-hanging fruits” help demonstrate tangible results early and convince leadership to continue investing in long-term AI initiatives.
But Øyvind also cautioned that half-done success isn’t real success. Real transformation only happens when projects move beyond pilots and deliver value in production.
“You need to approach the problem not as a pilot, but as something you’ll keep working on, until you either succeed and create real business value, or learn that what you’re trying to do just isn’t feasible.”
Finding the balance between AI agility and control
Øyvind pointed out that in today’s fast-changing AI world, moving too slowly means falling behind, but rushing all-in can waste huge budgets.
The safest path, he said, is to keep experimenting and learning while maintaining realistic expectations and controlled investments.
“There isn’t necessarily a rush to adopt AI today, but make sure your business is ready for it when AI itself becomes ready.”
Final Thoughts: AI itself is not the advantage – Mindset, Data, and People are
Øyvind’s lessons aren’t a list of quick wins. It’s a foundation for long-term, intelligent, and sustainable AI adoption that helps companies lead change, no matter how technology evolves.
His core message is simple:
“AI models are getting smarter every day — but that advantage is available to everyone.
The real winners will be the companies that understand their own problems and keep building their own set of tools to solve them.”
When technology becomes accessible to everyone, the only true difference lies in vision, data, and people. Companies that understand their problems deeply, invest in clean and structured data, and empower people who are ready to adapt, these are the ones who will turn the AI hype into real business value.
From ITviec’s Perspective: Øyvind’s insights don’t just apply to companies, they also speak to IT professionals building their careers in the AI era:
- Businesses now need professionals who understand both technology and business problems.
- IT professionals should look for places where they can experiment, learn, and grow continuously – that’s how to build lasting strength in the age of AI.
————————
This insightful interview marks the collaboration between ITviec and Orient Software to bring forward authentic conversations with leading tech leaders in AI and Data. Together, we aim to help Vietnam’s IT professionals gain practical perspectives on AI, Data, real stories and challenges in emerging technologies — bridging the gap between industry vision and career growth.
For more insight from other leaders and latest job opportunities in AI/Data field, visit ITviec’s AI/Data Segment now!
