Negotiations are personal. Here's how we protect the people who practise on Discurso — and the research behind every score they get.
What we do with personal data, in plain words. The Privacy Policy has the full detail.
Discurso ApS is based in Aarhus, Denmark. Our app and our servers, including the ones that make call transcripts, run in the EU (Netherlands).
Each service that touches your data is listed in our Privacy Policy and works under a data protection agreement. Where data leaves the EU, the EU–US Data Privacy Framework or Standard Contractual Clauses apply.
Ask us to delete an account and its personal data is erased within 30 days: profile, messages, feedback, survey answers, call transcripts, notes, files and its place in every event.
On request we hand over a copy of everything we hold on a person, in a portable format.
Call audio is deleted as soon as it has been transcribed, or after 7 days if transcribing fails.
We never sell personal data or use it for advertising, and we don't share it with anyone beyond the providers named in our Privacy Policy.
The safeguards that keep negotiators' work between them and their instructors.
Everything travels over HTTPS, and video calls are encrypted on their way between participants. What we store is encrypted at rest by Google Cloud.
Google sign-in or email with a strong-password policy, with bot protection on sign-up and sign-in.
Recordings and transcripts are only handed out by our server, to the people allowed to see them. Nothing in our file storage is open to the public internet.
Our AI service only answers requests from the Discurso app itself, never from other websites.
Research exports replace names with anonymous IDs by default. Anything identifying is a separate, clearly labelled opt-in for the instructor.
Analytics cookies only run if you accept them, and you can change your mind any time from the cookie settings link.
AI plays your counterpart, coaches you between rounds and drafts your feedback. It runs on models from OpenAI and Google Gemini: what you write in a round is sent to them to produce a reply. Under their business terms, they do not train their models on your negotiations. Our coach keeps no logs of what you tell it, and live voice conversations are not traced by the provider.
The AI doesn't decide what good negotiating looks like. Each skill is graded against the published behaviours of the Negotiation Competency Model, so a score describes what happened in the transcript rather than the model's impression of it. And where grades matter, people stay in charge: for classes the AI only proposes grades, and the instructor reviews, adjusts and decides.
Discurso doesn't invent its own yardsticks. Feedback, surveys and outcome measures come from established negotiation science.
R. Smolinski & Xiong (2020) · Negotiation Journal
Every one of the 15 competency scores a negotiator gets is graded against this model's observable behaviours, from seriously deficient to expert.
Read the paper →Curhan, Elfenbein & Xu (2006) · Journal of Personality and Social Psychology
After each round both sides rate how they feel about the outcome, the process, themselves and the relationship — the relational half of every result.
Read the paper →Raiffa (1982) · Harvard University Press
The idea of joint gains and the efficient frontier: every deal is placed against all the deals the case allows, so you see how much value a pair created and how much was left on the table.
Read the paper →Kesting & R. Smolinski (2023) · Negotiation Journal
Blueprints for writing distributive, integrative, multi-issue and multiparty simulations that teach what they are meant to teach — the basis for how Discurso's cases are built.
Read the paper →Brett (2026) · Current Opinion in Psychology
A review of the evidence on AI in negotiation training: coaching and feedback after a negotiation help people learn, while live AI advice during one risks dependency. Discurso's feedback comes after the round for that reason.
Read the paper →Mo (2026) · Scientific Reports
A controlled semester-long experiment: students who practised against a generative-AI counterpart outperformed the control group on every negotiation competence measured, rated blind by humans.
Read the paper →The platform is also a research instrument. Here's what we're working on.
J. Smolinski, R. Smolinski, Kesting & Kröcher (2026) · Group Decision and Negotiation
An AI agent told to maximise gains will reach for manipulative tactics unless it is held back. This paper maps those risks and sets out guidelines for AI negotiation agents, from design to deployment.
Read the paper →Discurso research team
A round-robin benchmark of ten leading AI models negotiating against each other across eight conditions — value claimed, value created and how the counterpart felt.
Discurso research team
What happens when different AI agents share one arena and can adapt as they go? An experiment on which negotiating styles they pick, and which ones survive.
Discurso research team
Can AI grade negotiation skill? Human raters barely agree with each other on competency scores, so this paper makes the case for hybrid human–AI assessment.
Running a data-protection or IT review for your institution? Email support@discurso.ai and we'll answer your questionnaire.