voi stands for "Voices of impact". An AI-powered software tool for impact measurement and reporting. voi is the first AI impact researcher. Describe your project and impact goals, and voi creates a scientific interview guide in minutes. It then interviews the people you serve at the scale of a survey, letting them answer by text, voice, or video, and asks follow-up questions the way a skilled researcher would. What comes back is verified evidence you can act on, share with funders, and stand behind.

A survey tool hands you a blank form and leaves the thinking to you. voi does the whole study. Before anyone is interviewed, it designs the research; builds a theory of change for your work; and selects validated indicators to measure against, the way a research team would. Then it interviews the people you serve at the scale of a survey, adapting its follow-up questions to each answer instead of collecting fixed responses.

When the interviews are done, it analyses everything and turns it into an impact report you can put in front of funders. A survey tool requires your capacity and gives you raw answers; voi gives you a finished study.

A traditional human-led impact study starts at €15,000 and takes months to complete because a firm does the work by hand. voi does the research powered by AI, so a study starts at €299 and a full report lands in days. The underlying research competency is comparable, built by researchers at leonardo impact and the Technical University of Munich. You stay in control of the questions and the conclusions. voi removes the cost by up to 95% and the wait that kept real impact research out of reach.

We acknowledge that voi is not for every project and we advise keeping a human researcher in the loop in certain cases. If you still want to work with a research agency, check out our impact studies offer or contact your partner of choice.  

A generic LLM helps at the edges of an impact study: it can suggest indicators before you start and help you make sense of data once you have it. The part it can't do is the middle: actually interviewing the people you serve. That gap is where the work really lives, and it's why using ChatGPT means exporting your questions into a survey tool, collecting responses there, then pasting the results back for analysis, losing structure and context at every handover. voi runs the whole study in one place, from Theory of Change and indicator selection through the interviews themselves to the finished report. Keeping it together pays off three times over: your analysis is built on the full context of every interview rather than a flattened export, so the output is better; you stop losing hours to moving files between tools and re-explaining your study to a model that has forgotten it; and the evidence stays traceable from a number in your report back to the voice that produced it.

‍voi's AI is built specifically for impact measurement, shaped by the Technical University of Munich and years of running studies with human researchers in the loop, so it applies impact research methods at every step rather than general helpfulness: it builds a Theory of Change that holds up to scrutiny, asks follow-up questions the way a trained interviewer would, probing an answer without leading the respondent, and analyses results with the caution a social scientist brings to what the data can and cannot support. The same discipline governs the numbers. Indicators come from a curated library where each one traces back to the published scale or framework that defined it and carries a fixed calculation, with the figures computed in code and the model writing only the prose around them. A score therefore means the same thing in your study this year, your study next year, and a partner's study in another country. For a funder that is what makes a portfolio comparable: the same indicator and the same calculation across every workspace it funds, with respondents told who is funding the study before they answer. A general LLM will happily invent a scoring approach on the spot, and a different one next time you ask.

‍voi also routes each task to whichever model is best at it, since strong researching, strong reasoning and strong interviewing are rarely the same model. And it is built to handle respondents' personal data lawfully and responsibly: your organisation stays the data controller under a Data Processing Agreement, respondents are told before the first question that an AI conducts the interview, all text processing stays in the EU, and no language model provider retains your data or trains on it. A consumer ChatGPT or Claude plan gives you none of that, keeps your data by default, and tells the people whose answers you paste in nothing at all.

Every number in a voi report links back to the interview that produced it. There are no self-reported claims taken on faith, and no black box. You can trace any finding back to the voice behind it. Respondents also see a clear notice before they take part, so participation is always informed.

voi is built for GDPR and EU AI Act compliant research. Respondents are told an AI conducts the interview before they answer. voi runs no emotion recognition, makes no automated decisions about respondents, sells no data, and uses no third-party trackers.

voi routes every AI request through a European gateway (Opper, in Sweden) to models from Anthropic, Google, and OpenAI. Three rules decide who we work with, and each one protects your respondents:
‍1. Everything runs in the EU.
2. Every provider operates under zero data retention, so no copy of an interview is kept once it's processed.
3. No provider ever trains its models on your data.

AI-powered transcription is a separate path that does not go through the gateway. Recordings go to ElevenLabs Scribe; the large language models never receive them. ElevenLabs Scribe runs on servers in the US, but cannot train on the study data either. We aim to replace transcription with a European alternative soon.

voi works wherever people are at the heart of the impact. voi is built for social enterprises and impact organisations who want real evidence of what they're changing. It also works for impact investors, foundations, and accelerators measuring across a whole portfolio, and for NGOs, researchers, and policymakers who need beneficiary voices at scale. If you're accountable for impact and tired of self-reported numbers, voi is for you.

A single study starts at €299 for 50 respondents, with Growth (150 respondents) and Pro (300 respondents) tiers for larger studies. Higher tiers also unlock more features, not just more respondents. You pay once per study with no subscription required. That's a fraction of the €15,000 to €25,000 a traditional study costs. See the pricing page for the full breakdown.

A single study is the simplest way to start. Pay once, run one study, and you're done. A plan makes sense if you'll run more than one study a year, or if you fund studies for others. A foundation or accelerator can run studies across its whole portfolio from one account and keep the results for its own reporting. Plans lower your cost per credit and can be raised or cancelled at any time. The pricing page compares both.