Duna

Models carry real work here, from structuring company documents and real time risk scoring to agents that gather evidence for periodic reviews. An onboarding orchestration platform would survive their removal as a saleable product, so the AI sits on top of a decision engine rather than constituting it.

Agents collect evidence while analysts make final judgments, and Duna Decide adds four eyes review on cases. Automated approvals are also offered, with no threshold, sampling or rule for which cases skip a person, on a decision that gates a business's access to a bank account or marketplace.

Decision logic written as policy as code is inspectable and versioned, giving a validator real rules to examine. Accuracy, validation evidence and false positive rates are unpublished, and the figures shown measure speed and volume rather than how often the system was right.

KYB reviews at seQura run 16.3 times faster with a 76 percent lift in first time right decisions, per a named case study. Homepage aggregates of 10.6 times faster onboarding and a 37 percent conversion increase carry no customer attribution.

A digital passport for every business, shareable across institutions for one click onboarding, is the stated destination, and who owns it, what consent applies and how it is revoked are unaddressed. Training use of customer data is not addressed either.

The privacy policy, updated May 25, 2026, reaches the businesses and individuals being onboarded, with Duna as processor for onboarding data and access, deletion and export rights stated. Retention runs as long as needed without fixed periods, and service providers are pointed to a trust page rather than listed.

Encryption, penetration testing and a bug bounty program sit on the trust page beside ISO 27001 certification and a completed SOC 2 Type II audit. Dates, scope and report availability are not given.

Know your business, customer due diligence and anti money laundering are named repeatedly, with daily screening against sanctions and politically exposed person lists. Neither EU anti money laundering instruments nor national supervisors appear, and no step is mapped to the obligation it discharges.

Wrongful exclusion is tracked as a business outcome, a real credit, yet nothing published shows which businesses get refused. Registry gaps, transliteration and adverse media coverage weaken in underserved jurisdictions, so a conversion average can hide uneven error rates.

Logged case actions let a decision be contested inside one institution, but no guarantee or accuracy commitment stands behind it. Today a refused business can try another institution, and a passport shared across a network would remove that practical appeal, with no correction or challenge route described.

Registry, screening and adverse media sources sit in the decision path for every check, and none is named on the homepage or trust page. Whether company documents and ownership records reach an outside model provider during extraction is unstated.

Smart routing picks registry, screening and enrichment sources per jurisdiction and entity type across more than 210 local registries, taking real integration work off the buyer. Outbound links are thinner, with no core banking platform or external case system named.

All customer data is stored in the EU on AWS under GDPR, with backups across multiple EU locations. A private deployment is not described, nor is where a reused passport lives once an institution outside the EU picks it up.

Pricing, packaging and basis of charge are unpublished, with contact running through a demo request. Data provider calls carry their own per check cost in this category, which a buyer inherits on top of whatever the platform charges.

Plaid, Moss and Bol sit among named customers spanning payments infrastructure, spend management and marketplaces, for a stated buyer set of banks, fintechs and regulated platforms. Both registry depth and the customer base are European, with the company in Amsterdam, and comparable depth in North America or Asia is not shown.