Using GenAI is not the point. Benefiting from it is.
Prof. Dr. Tony Gorschek, SERL Sweden – Blekinge Institute of Technology (BTH)
Almost every organization I talk to has the same problem with generative AI, and it is not the technology. It is getting started. Everyone has access to the tools, and everyone has seen impressive demos. But between the demo and real measurable value, sit a lot of caveats. The output can be wrong while sounding right. Models, prices and vendors change every few months. The people who should use it are already busy. And nobody quite knows which parts of the business will benefit, and which will just get faster at producing things nobody cares about.
So the question I hear most often is not “should we use GenAI?” It is: how do we invest in this without losing money and time?
The answer: one person, going deep
Take on an industrial PhD student, together with a strong applied research group (SERL at Blekinge Institute of Technology, for example).
Not to build models. To have one person whose job is to go deep into your business and find where GenAI actually pays off. Treat it as ordinary business development, not as “now we do AI”. GenAI is one possible solution, just as “digitalisation” was before it. GenAI is not the purpose, it is a way to increase the efficiency and effectiveness of your core business. (In some more advanced cases it is enabling you to develop a new type of products, but that is a story for a separate discussion…).
What that person does
- Finds use cases where GenAI could make something faster, cheaper, better, or all three.
- Sets up measurement before anything changes. Boring, yes, and the step almost everyone skips. But it matters most: it tells you whether you succeeded, and it defines the goals, the actual requirements BEFORE you invest and take on costs.
- Introduces the change, tunes it, tests it, measures the effect (one small step at a time).
- Computes the delta: Δ = cost before − (cost after + cost of GenAI). If Δ is positive, it was a good idea: keep it. If not, roll it back. Either case you learn!
Then repeat across the business. The measured benefit decides, not hype or internal politics. Run 100 use cases and you may keep 40. In my experience, those 40 pays for ten PhD students and your cost is one…
Figure 1. The approach on one page.
Everyone wins
- The organization gets measured gains.
- The research group (that supervises the PhD student) gets results from real operations and research. One thing to remember. Since the expert doing this in your organization is an industrial PhD student hired by you, the focus is on your benefit, BUT it is research, which means the PhD student and the research group have to publish (scientifically) the results and *prove* benefit, not sell it and convince you. In other words you get objective measurement of benefit of your work and investment as a significant bonus!
- When the PhD is done, you hire that person as your own AI expert, someone who already knows your domain, data and people. A person that does not exist to just hire, but you helped create one for yourselves. Then this person keep improving and maintaining what works in your organization! And that is critical: GenAI solutions drift as models, data and vendors change. You cannot “buy” a quick solution. It is an investment that has to be maintained and evolved over time. This cannot be bought from external experts… at least not for a price that is less than the benefit you see.
Why not consultants or the AI vendors?
Both have their place, but the incentives differ.
Consultants are paid for hours, and a few months on site rarely reaches deep into how an organisation works. When they leave, the knowledge leaves with them.
The big AI platforms are paid per token: they earn when you use GenAI, not when you benefit from it.
Using GenAI is not the point. Benefiting from it is. That is the key.
Why measurement matters
The bottleneck has moved. AI generates far faster, but someone still must decide what counts as right. A 2025 study by METR found that experienced developers were 19% slower with AI tools while believing they were faster. Adoption is easy. Measured value is the hard part.
Using it well
Sweden did not invent the internet, but we built Spotify, Skype, Klarna and Minecraft on top of it. We can do the same with GenAI: by using it well, on purpose, and measuring what it gives us. The same is true for public sector and other organizations. Your main focus is the product or service you provide. GenAI can help you provide it faster and with higher quality for less costs if you do it correctly.
If you can fund one salary for four years, this is the lowest-risk, highest-return AI investment I know of. Get in touch if you want to talk about how it could work for your organisation.
About SERL Sweden
The Software Engineering Research Lab (SERL) at Blekinge Institute of Technology is ranked top 10 in the world and first in Europe in applied (empirical) Software Engineering. That is helping organizations improve efficiency and effectiveness of their core functions in any part that involves software intensive products and services. SERL studies how software-intensive products and services are conceived, designed, built, validated, deployed and maintained. Our research is empirical and applied, and built on long-term collaboration with industry across telecom, automotive, defense, fintech, manufacturing, entertainment and other sectors.
Two shifts shape our current work: using generative AI as an instrument of engineering work (AI for Software Engineering, AI4SE), and engineering products and services that contain AI components
