The shape of survey analysis is changing. Recense is built for what comes next.

Recense is a survey analysis tool built around one bet: the next decade of insights work is a collaboration between a researcher and a frontier AI model, and the software for it has not been built yet.

The problem the tools were built for is no longer the problem

Modern survey analysis tools changed the work once already. Before them, analysing a survey meant briefing a data-processing company, waiting for crosstabs in Excel, and emailing back a list of corrections. Tools like SPSS and Displayr collapsed that loop: researchers could self-serve crosstabs from a .sav file with a bit of setup. The time between a question and an answer dropped from days to minutes. Faster iteration meant better analysis, better storytelling, more impact.

Working with a frontier model on a real, complex U&A deep-dive in late 2025 made it clear something similar was happening again. The bottleneck has moved further downstream — from generating crosstabs to interpreting them and building a narrative the business can act on. The model could propose analyses, draft commentary, challenge weak threads, and ladder up to a story that was actually interesting. The work shifted from running tabs to directing an agent: which hypotheses to test, which threads to pull, which insights to lead with.

Software that centred the work on building crosstabs made sense for the previous era. It does not match the way the best researchers are starting to work now.

What Recense is

Recense is built for human-AI collaboration in market research. Import an SPSS file, work with the data in a fast, drag-and-drop canvas, and brief an agent that operates the same surfaces you do. The agent can build tables, apply filters, derive variables, and draft the narrative — in your workspace, on your data, with your choice of model.

The architecture follows the bet. Computation runs in your browser via a Rust engine compiled to WebAssembly, so local files stay local and tabulation is measured in milliseconds. The agent surface is wired through a designed-for-it protocol, not bolted onto a chatbot. Model choice is open: bring your own key, swap providers, move to whatever is best in the next six months.

What Recense isn't

Recense isn't trying to replace researchers. The interesting questions, the business context, the judgement about what matters — these are the parts of the job AI hasn't earned. They probably won't for a while. Recense is built to amplify those parts, by taking the friction out of everything else.

Recense isn't a synthetic-population engine and doesn't pretend to be. Synthetic microdata can extend real signal in specific, transparent ways. It is not a substitute for actually asking people.

And Recense isn't a thin wrapper around a single embedded model. The pace of frontier model improvement consistently outpaces what any closed implementation can keep up with. Letting users swap in the latest models — directly, today — is a feature, not a concession.

Who's behind it

Built by Matt Harris — a former market researcher who spent years using the tools described above, then spent 2024 and 2025 working with frontier models on the kind of analysis they are starting to do well, and decided the next-generation tool needed to exist. Background in consumer insights at scale. UK-based.

If you'd like to read the longer version of the why now argument, the post that prompted this work is on LinkedIn.

Will Recense be here in five years?

Recense Ltd is founder-funded today, with a deliberately small cost base — the engine runs in your browser, and the team is one person. The plan is to grow on revenue from researchers and teams who find the tool genuinely useful, not on an outsized burn-and-pivot path. The commitment goes both ways: your data stays in formats you already own (SPSS .sav, CSV) and the export path is always available, so you can leave at any time.