One question, two tracks #
My research asks one question: how do organizations and people meet AI, and who pays when the design gets it wrong?
I work on it along two tracks. The first is AI transformation in small and medium-sized enterprises. The second is generative AI as persuasive actors. By that I mean systems that do not only carry a message but adjust it to whoever is in front of them, while the conversation is still running.
The tracks meet in a case I keep coming back to. A small firm puts a chatbot on its website and wants it to do two things: advise visitors, and get them to act, whether that is a download, a booking or an order. Should the chatbot express emotions to do that job? What does it do to the visitor when it does? Those questions belong to both tracks at once, and they are what I am working on now.
AI transformation in SMEs #
The question #
Most SMEs have thin resources, uneven digital maturity and no slack for theory that cannot be turned into action. Most of what is written about digital and AI transformation quietly assumes a strategy team, a data foundation and a budget for mistakes. Small firms have none of those.
So I look for approaches that survive contact with that reality. What does a small firm actually need to learn? How can it choose technology without a specialist on staff? What does a transformation path look like when it has to be cheap to run and easy to correct once it turns out to be wrong, which it will?
Where it stands #
With Agnis Stibe I presented SMEDT at MobiWIS 2025 in Istanbul: an AI-first discovery of a capability-driven digital transformation framework for SMEs. It is a first version. The next stretch is testing it with real firms, which is slower and messier than building it was. Paper · Notes from Istanbul
Why it matters #
SMEs carry most of the European economy and they keep falling behind larger firms on AI. There is plenty of research on transformation. Very little of it is something an owner-manager can pick up on a Tuesday and use. That gap is where I want to work.
Generative AI as persuasive actors #
The question #
Persuasive technology used to mean designing a system to influence behaviour. Generative AI changes the subject of that sentence. A large language model reads the person it is talking to and adjusts tone, framing and emotional register as it goes. The persuader is no longer a designer somewhere upstream. It is the system itself, in the conversation, adapting.
That raises questions I find hard to put down. What happens to influence and consent when a system simulates empathy or encouragement while tuning its message to the signals we leak in conversation? The asymmetry bothers me most. The system can simulate emotion toward you to move you. You have nothing comparable pointed back at it.
My angle is SME-sized. What happens when a small firm with no data team and no ethics board puts one of these systems in front of its customers?
Current case #
I investigate how small and medium-sized enterprises can use persuasive chatbots for the two jobs they increasingly hand to them: advising customers, and moving those customers toward a call to action. The lever I am studying is synthetic emotions. Does a chatbot that expresses feeling advise better and convert better, and what does that cost the customer?
The first study tested the advising job. With Thong H. N. Dinh and Agnis Stibe I ran a randomized experiment with 126 participants in career guidance. Two chatbots gave the same guidance in different styles, one empathic and coaching, the other strategic and analytical. Style reshaped the early part of the path from understanding to intention. By the final step toward action the difference had disappeared, which was not what I expected. Presented at MobiWIS 2026 in Granada. Paper · Notes from Granada
The conceptual groundwork, on anthropomorphism and influence in AI-driven persuasion, I presented at the IADIS Information Systems conference 2026 in Zagreb. Paper · Notes from Zagreb
Next study #
The follow-up tests the call to action job in a setting any small firm will recognise. Participants, students acting as prospective customers of an SME, talk to one of three chatbots: a neutral one, an empathic one, and a rational, analytical one. Each has the same goal, to get the visitor to request more information by downloading a PDF. We measure which stance gets there most efficiently.
The study is in the design phase. If your organization would like to host a version of it, I would like to hear from you.
Why it matters #
Chatbots are becoming the front door between small firms and their customers, and they are sold as a plug-in with a tone setting. Warm or matter of fact is a dropdown now. That choice affects conversion, and it affects how free the customer is in the decision they end up making. I want to give SMEs an answer they can act on, backed by data, and some idea of where the ethical line runs.
Working with me #
I collaborate on both tracks. If you want to co-author, share data, or let your firm host a study, get in touch.
The complete list of papers, book chapters and patents is on the Publications page. The full record is also on Google Scholar and ORCID 0009-0009-9965-4269.