From Fieldwork to Foresight: How a Traditional Research Agency Rebuilt Itself as a Research Tech Company
By Demoskop
- article
- AI
- Artificial Intelligence
- AI Agents
- Full Service Research
Summarise with AI
For decades, the economics of market research were easy to understand: more projects required more people. More interviews meant more fieldwork. More data meant more analysts. More clients meant more PowerPoint decks.
AI has broken that equation.
At Demoskop, a Swedish market and opinion research company founded in 1989, we have spent the past three years testing what happens when AI is not treated as a tool added to the old agency model, but as infrastructure underneath the entire research process.
Today, Demoskop looks and operates like a different company. We haven’t changed our name, clients, or commitment to methodological rigour. What we’ve changed is how the work gets done, and, as a result, what we can offer clients, how we’re organised, and where we sit in the value chain.
This is the story of how we integrated AI into every workflow in the business, why that forced us to rethink our organisation and our business model rather than just our tooling, and what it has meant for growth, productivity, and our relationship with clients.
The Problem With “Adding AI” to a Traditional Agency
Like most agencies, our first instinct wasn’t wrong, simply too small. We piloted AI tools for transcription, coding open-ends, and for the odd chart. It saved a few hours here and there, but it didn’t change the economics of the business, because the underlying model hadn’t changed: we were still selling human hours, project by project.
The real shift happened when we stopped asking “where can we bolt on an AI tool?” and started asking “if we were building a research company from scratch today, what would it look like?”. That question led us to build Stella Research Engine™ – our own AI platform for analysis and insight, designed from day one to sit underneath every stage of the research process rather than beside it, built on multiple language models so the user always gets the best tool for the task without switching platforms.
Integrating AI Into Every Workflow – Not Just the Analysis Stage
The mistake many agencies make is treating AI as a point solution for one bottleneck – usually analysis or reporting. We took the opposite approach and mapped every stage of the research value chain to identify where AI could remove friction, not just save time:
- Project scoping & questionnaire design: AI-assisted drafting against a knowledge base of prior studies, so every new brief starts from institutional memory rather than a blank page.
- Data processing & coding: our quantitative open-text analysis module turns large volumes of free text into themes and concrete findings researchers can use to build decisions and reports.
- Analysis & problem-solving: a flexible multichat assistant our analysts use for analysis, text production, and problem-solving, alongside transcription and voice tools that turn audio into text for analysis and text back into voice for communication and presentations.
- Content & storytelling: an AI-assisted media creator that produces and edits images and graphics quickly for reports and presentations.
- Client delivery: AI Personas built on a client’s own data or research we’ve conducted for them, so stakeholders can chat directly with their target segments, stress-test messaging, and get instant feedback on tone and arguments – with personas continuously retrained as new data comes in.
Crucially, none of this replaced our researchers. It replaced the time our researchers spent on tasks that weren’t research – data wrangling, first-pass coding, formatting, repetitive querying – freeing them to spend more time on the parts of the job that require human judgment: framing the right questions, interpreting nuance, and advising clients on what the numbers actually mean for their decisions.
The Result: Growth and Productivity, Funded From Our Own Cash Flow
The percentages tell a story clearly on their own. Over the past year, revenue grew by roughly a quarter, while average headcount grew by less than a tenth. That gap is the real evidence that the AI layer is absorbing volume growth that would previously have required proportional hiring: revenue per employee (our best available measure of productivity) improved by well over 10% in the same period.
Underlying profitability – that is, profitability from ongoing operations, before the cost of building our own technology platform – has increased. We made a deliberate choice to fund the development of Stella Research Engine and our platform migration entirely from our own cash flow, rather than raising external capital. That choice meant absorbing a temporary cost in the short term, but it kept us in full control of our own technology roadmap and our own equity – something we think matters a great deal for an insights business whose product is, ultimately, trust.
Since then, we’ve seen the trend continue in the right direction: strong order intake, and the AI initiative and our reformed way of working attracting new customer segments, both nationally and internationally.
Building the Organisation Around AI, Not Around Titles
None of this happened by simply installing AI. It required us to change how the company itself is organised.
We created a Head of AI and Transformation – a role whose sole mandate is to redesign workflows around the technology, not just deploy tools into old processes. This person sits between the technical build of Stella and the day-to-day reality of research delivery, making sure every workflow redesign is co-built with the researchers who’ll actually use it, which means framing AI explicitly as a way to elevate their role, not threaten it.
We also introduced a new role that didn’t exist in a traditional agency structure: the Research Automation Engineer. This person sits at the intersection of software development and analytical production, and the job spans four distinct responsibilities:
- Contributing directly to the Stella Research Engine development team, working in a DevOps environment with CI/CD pipelines and system integration.
- Identifying and automating the manual steps still buried inside our analytical production process.
- Owning data quality and fraud-detection controls across our production pipeline.
- Ensuring every automated solution still meets GDPR and ICC/ESOMAR standards.
It’s a genuinely new job description for our industry – part engineer, part research operations specialist – and one we expect more agencies will need to create as automation moves deeper into analytical production.
Alongside these new roles, we restructured how we work day to day around client teams. Everyone on a client team meets the client directly – regardless of whether their formal role is research, technology, or project management. That’s a deliberate departure from the traditional agency hierarchy, where client contact was concentrated in a small number of senior roles while the rest of the team stayed invisible to the client. Full client exposure across the whole team means faster feedback loops, and it means the people building and refining our AI tooling hear directly from the people using its output – which has itself become one of our fastest drivers of further product development.
Moving Up the Value Chain: From Supplier to Strategic Partner
The organisational and technological changes above have let us do something a traditional agency model never allowed: go further in delivery and work as genuine partners to our clients, rather than as an external supplier handing over a finished report.
Concretely, clients who work with us today can get direct access to the same platform and the same panels we use ourselves. Where it makes sense, we build AI agents together with the client, inside Stella Research Engine, tailored to their specific questions and market. And because the platform is model-agnostic and built to be configured to how an organisation actually works with insight – not a generic, one-size-fits-all AI tool – clients can even bring in and analyse competitors’ data inside the same environment. That’s not a threat to our business model; it’s irrelevant to it. Our value was never in gatekeeping data. It’s in the judgment we bring to interpreting it.
That shift has moved us higher up the value chain. We’re no longer just a supplier of surveys and reports; we’re strategic advisors who understand AI and who understand our clients’ competitive environment well enough to help them ask better questions of their own data – whichever data that turns out to be.
The Bigger Lesson for the Insights Industry
The research industry’s conversation about AI has largely focused on efficiency: doing the same work faster and cheaper. That’s real, and our productivity numbers reflect it. But the more interesting shift is that AI, properly integrated, changes what an agency’s product actually is – from a delivered report to a living, queryable insight capability co-owned with the client – and that changes the organisation, the business model, and ultimately how high up the value chain an agency gets to sit.
We didn’t set out to become a “research tech” company. We set out to solve a workflow problem. Solving it properly meant rebuilding the organisation, the business model, and the client relationship underneath it – funded, deliberately, from our own resources rather than someone else’s capital. That’s the transformation I’d encourage every traditional agency leader to actually pursue, rather than stopping at the first AI tool that saves a few hours a week.
