What problem does it solve?
Onboarding a company is slower and harder than onboarding a person. The bank has to establish who the company is, who controls it and who ultimately owns it, which means pulling filings from registries in several countries, reading articles of association and trust deeds, following ownership through layers of holding companies, and screening every entity and person found. Much of the data is in PDFs, in other languages, or missing from registries with no digital access.
Analysts can spend more of their time building the file than judging it, clients receive repeated document requests, and cases wait while documents are chased. Opaque structures are exactly where the financial crime risk sits, so shortcuts are not an option. AI can build the file and the ownership graph quickly and consistently, as long as every link in the chain stays traceable to a source and a person makes the decision.
How does it work?
- Fetch. From the company name or registration number, the agent pulls registry records, filings and ownership data from the available registers and data providers.
- Read the documents. Document AI extracts officers, shareholders, percentages and control rights from articles, share registers, trust deeds and client submissions.
- Resolve and map. Entity resolution links the same company or person across sources, and a graph of ownership and control is built up to the ultimate beneficial owners, with the calculated effective ownership per person.
- Screen. Every entity and person is screened for sanctions, politically exposed persons and adverse media, and gaps (no registry, missing documents) are listed.
- Score and summarise. The case is risk scored against the bank's policy and summarised in plain language, with every finding linked to its source.
- Decide. A compliance analyst reviews, overrides where needed and decides; low risk cases that meet the policy still get a person's lighter review before the decision is confirmed.
- Audience
- Back office
- Autonomy
- Supervised agent
- Adoption
- Emerging
- Channels
- Internal tools, API and system to system
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Automation rate | Too few to pool | 25% | 1 | 1 organization |
| Cycle time reduction | Too few to pool | 20% | 1 | 1 organization |
| Productivity gain | Too few to pool | 30x | 1 | 1 vendor |
| Accuracy | Too few to pool | Not pooled: up to 99% | 0plus 1 up to | 1 vendor |
Value drivers: Speed and cycle time, Compliance quality, Lower cost to serve, Risk and loss reduction.
Indicative value
A bank onboarding 3,000 business clients a year
USD 180,000 to USD 960,000
Analyst time released, valued at loaded cost per year
How this is calculated
Formula: cases * hoursPerCase * reduction * hourlyCost. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Business onboarding cases per year cases, cases per year | 3,000 | 3,000 | The reference bank. |
| Analyst hours to build and review a KYB file hoursPerCase, hours per case | 4 | 8 | Editorial assumption, replace with your own time study. Complex structures take far longer. |
| Share of analyst time saved reduction, fraction of time | 0.3 | 0.5 | Editorial assumption, replace with your own pilot results. |
| Loaded cost of an analyst hour hourlyCost, USD per hour | 50 | 80 | Editorial assumption, replace with your own loaded cost. |
What it leaves out: Values analyst time only. It leaves out data subscription costs, revenue from clients onboarded sooner, fewer clients lost to slow onboarding, and the reduced risk of missing a hidden owner.
Who already uses it?
3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
BNY
United States · Capital markets · 2026
BNY built its own AI platform, Eliza, on Microsoft Azure and Microsoft Foundry. In a Microsoft customer story, Saed Shonnar, Head of AI Enablement at BNY, says twenty five percent of the bank's new client onboardings this year were assisted with AI, resulting in a twenty percent faster onboarding process on average. A second BNY executive describes the research element of onboarding, the document processing and decision making behind verifying a new client, as the common challenge the AI addresses. The same story describes a digital employee that repairs incomplete payment instructions and faster processing of client settlement inquiries, which are reported as separate use cases.
- Automation rate: 25%, new client onboardings this year
"Twenty-five percent of all of our new onboardings were assisted with AI this year and that has resulted in a twenty percent faster onboarding process on average for these clients,"
Claimed by: organization - Cycle time reduction: 20%, new client onboardings this year, average
"Twenty-five percent of all of our new onboardings were assisted with AI this year and that has resulted in a twenty percent faster onboarding process on average for these clients,"
Claimed by: organization
Incore Bank
Switzerland · Banking · 2026
Incore Bank, a Swiss bank that serves other banks, financial intermediaries and corporates rather than retail customers, completed a proof of concept with Kyndryl and Google Cloud that applies agentic AI, built on Kyndryl's Agentic AI Framework and Google's Gemini models, to the know your customer checks it runs on prospective and existing business clients. Several AI agents extract and validate customer information from documents, internal systems and external sources, identify risk factors, produce an explainable risk score and create an auditable decision record for compliance staff to review. Kyndryl reports the proof of concept reached up to 99 percent accuracy extracting data from onboarding documents; the further claim that the approach could cut onboarding time from months to days is stated as a demonstrated potential, not a measured result.
- Accuracy: up to 99%, automated extraction of data from customer onboarding documentation, during the proof of concept
"During the proof of concept, the solution achieved up to 99% accuracy in the automated extraction of data from customer onboarding documentation and demonstrated potential to reduce onboarding time from months to days."
Claimed by: vendor
M-DAQ Global
Singapore · Payments and cards · 2024
M-DAQ Global, a fintech group headquartered in Singapore that specialises in foreign exchange and cross border payments, runs a Know Your Business compliance solution on Vertex AI and Google Kubernetes Engine that uses natural language processing to automate the verification work behind onboarding business customers. The vendor reports a productivity gain of 30 times and shorter onboarding times.
- Productivity gain: 30x, compliance tasks, as reported by the vendor
"The natural language processing-based system automates compliance tasks and improves productivity by 30 times, reducing onboarding times and eliminating manual bottlenecks in customer verification."
Claimed by: vendor
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Access to company registries and commercial ownership data for the bank's markets
- The bank's KYB policy, risk model and beneficial ownership thresholds per jurisdiction
- Screening lists for sanctions, politically exposed persons and adverse media
Systems to integrate
- Client lifecycle management or onboarding case system
- Registry and data provider APIs
- Screening engine
- Client portal for document requests
- CRM
Complexity: High
Registry coverage and quality vary widely by country, ownership calculations through circular or layered structures are tricky, and the output feeds a regulated decision that examiners review.
- 1
Codify the policy first
Write down, per jurisdiction and client type, which documents are required, the ownership threshold, and what makes a case low, medium or high risk, before any automation.
- 2
Build the file, not the decision
Start with the agent assembling the case file and ownership graph for analysts, and measure time and quality, before letting any case move forward with lighter review.
- 3
Show confidence and gaps
Make the agent state where registry data is missing or weak, and route those cases to manual research instead of filling the gap.
- 4
Keep the chain examinable
Store for every ownership link the document or record it came from, so an examiner can follow the chain from the client to each beneficial owner.
- 5
Tune on real cases
Compare agent built files with analyst built files on a sample of past cases, including complex structures, and fix systematic misses before scaling.
Guardrails
- Every ownership link and screening hit is traceable to a source document or record
- Gaps and low confidence are surfaced, never silently filled
- A compliance officer decides onboarding, exits and enhanced due diligence
- Low risk cases that meet written criteria still get lighter review, with a person confirming the onboarding decision and quality assurance sampling the outcomes
- Personal data of owners and directors is used only for the compliance purpose and retained per policy
KPIs to instrument
- Elapsed days from application to decision, by risk level
- Analyst hours per case, by structure complexity
- Share of cases with complete files on first review
- Ownership errors and missed owners found in quality assurance
- Document requests sent to clients per case
Human in the loop
Analysts review every case above the low risk threshold and can override any finding. Low risk cases still get a lighter review, and a compliance officer decides onboarding and exits in every case. Quality assurance samples cases that passed with lighter review, and the model owner reviews screening and scoring performance.
Common failure modes
- False completeness
- The graph looks complete because a registry returned nothing about a layer. Show coverage per jurisdiction and flag layers without data.
- Wrong entity match
- Two companies or people with similar names are merged. Use identifiers where available and send low confidence matches to an analyst.
- Opaque scoring
- Analysts cannot explain why a case scored high. Show the factors behind each score and keep the model documented and validated.
What are the risks and rules?
EU AI Act
Depends on design
Customer due diligence on legal entities is not listed in Annex III, and an internal analyst tool usually carries no Article 50 transparency duty, so the system is usually minimal risk. The design decides the rest: biometric verification that only confirms a director is who they claim to be is excluded from Annex III point 1(a), but remote biometric identification (one to many matching) is high risk, and so is any use of the output to assess the creditworthiness of the natural persons involved (point 5(b)). GDPR applies to the personal data of owners and directors throughout. Keep biometric and credit steps in separately assessed components.
Rules that apply
Guidance
- Regulation (EU) 2024/1624 on the prevention of the use of the financial system for money laundering or terrorist financing (European Union, Europe). The EU Anti Money Laundering Regulation applies from 10 July 2027. It sets the customer due diligence and beneficial ownership rules the case file must meet, including the ownership and control tests. Article 76(5) requires meaningful human intervention in every decision to enter, refuse or maintain a business relationship with a customer, and in every decision to raise or lower the customer due diligence measures applied, so no such decision can be left to straight through automation.
- CDD Final Rule (FinCEN, North America). The Bank Secrecy Act rule that requires US banks and other covered institutions to identify and verify the beneficial owners of legal entity customers when those companies open accounts. FinCEN ruling FIN-2026-R001 grants exceptive relief from repeating this at each new account opening.
- MAS Guidelines for Artificial Intelligence (AI) Risk Management (Monetary Authority of Singapore, Asia Pacific). Consultation paper of 13 November 2025 proposing supervisory expectations on AI oversight, AI inventories, risk materiality, human oversight, testing and monitoring, including for AI agents. Relevant where AI output supports financial crime decisions.
Controls to put in place
- Written KYB policy per jurisdiction that the agent is configured and tested against
- Source record for every ownership link and screening result, retained with the case
- Model inventory, validation and monitoring of entity resolution and risk scoring
- Quality assurance sampling, with higher sampling for straight through cases
Frequently asked questions
- Can AI decide whether to onboard a company?
- It should not. AI can build the file, map ownership and score risk, but a compliance officer decides onboarding and exits, and a person confirms every decision. Low risk cases that meet written criteria only get a lighter review, with sampling.
- How does AI find hidden beneficial owners?
- By extracting owners from filings and documents, linking the same entities and people across sources, and calculating effective ownership through every layer. It still depends on registry coverage, so gaps must be flagged rather than guessed.
- Are there published results?
- Two vendor case studies name a real deployment. Google Cloud lists M-DAQ Global, a fintech group in foreign exchange and cross border payments, whose KYB compliance system on Vertex AI improves productivity by 30 times. Kyndryl reports that a proof of concept it ran with Incore Bank, a Swiss bank whose customers are other banks, financial intermediaries and corporates, and Google Cloud reached up to 99 percent accuracy extracting data from onboarding documents, with a further, unmeasured claim that onboarding time could fall from months to days. Both are vendor claims, one from a fintech and one from a bank, and neither has independent oversight behind it. Treat such figures as a starting point and test them in a pilot.
How to cite this page
Blits.ai AI Use Case Library, "AI for business onboarding (KYB) and beneficial ownership discovery", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/business-onboarding-and-ubo-discovery. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published