AI use case

AI assistant for digital account onboarding and KYC

A customer facing AI assistant that guides a new applicant, a person or a small merchant, through a digital account, card or relationship application: it collects and checks identity and supporting documents, orchestrates the know your customer and anti money laundering checks, prefills what it can and sends only the unclear cases to a human reviewer with a summary. The ownership research for complex corporate clients is a separate back office job.

By Len Debets · Last verified 27 September 2026 · 3 public deployments

30x
Reported productivity gain
M-DAQ Global, vendor claim.
USD 150,000 to USD 1.5 million
Indicative value per year
A retail bank that receives 100,000 digital account applications a year. Worked example, see how it is calculated.

What problem does it solve?

Onboarding is where a bank wins or loses a new customer, and it is also one of the most heavily regulated steps. The applicant has to choose a product, fill in long forms, photograph an identity document, pass a liveness check and sometimes explain where their money comes from. Every extra step or unclear question costs completions, and every applicant who stalls either leaves or calls the contact centre.

Behind the form, operations and compliance teams do the slow part by hand: checking documents that are blurred or expired, chasing missing information, screening names and, for wealth and business clients, researching source of wealth or the structure of a company. Rules differ by country, so a bank that operates in several markets maintains several processes. Fully manual review does not scale; fully automated rejection loses good customers and still misses sophisticated fraud such as synthetic identities and deepfakes.

How does it work?

  1. Help the applicant choose. The assistant asks a few needs questions and explains the products the applicant is eligible for, in plain language and the applicant's own language, without giving personal advice.
  2. Collect data once. It prefills from what the bank already knows (a lead form, an existing relationship or a government digital identity where one is available) and asks only for what is missing.
  3. Capture and verify identity. Document capture, authenticity checks, liveness and face matching run through the bank's identity verification provider; the assistant explains failures ("the photo is blurred, try again in better light") instead of ending the journey.
  4. Run the checks. An agentic workflow calls sanctions and politically exposed person screening, fraud signals and, where needed, source of wealth or company registry research, and records every result with its evidence.
  5. Decide the route. Clean cases continue straight to account opening under the bank's rules; unclear or high risk cases go to a human reviewer with a summary of what is missing or inconsistent. The approval decision on edge cases stays with a person.
Audience
Customer facing
Autonomy
Supervised agent
Adoption
Early adopters
Channels
Mobile app, Web chat, WhatsApp, Internal tools

What is it worth?

Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.

Value benchmarks for AI assistant for digital account onboarding and KYC
KPIMedianReported rangeData pointsClaimed by
Productivity gainToo few to pool
30x
11 vendor

Value drivers: Revenue growth, Speed and cycle time, Compliance quality, Lower cost to serve, Inclusion and access.

Indicative value

A retail bank that receives 100,000 digital account applications a year

USD 150,000 to USD 1.5 million

First year revenue from recovered applications per year

How this is calculated

Formula: applications * abandonRate * recoveredShare * valuePerAccount. The low scenario uses every low input, the high scenario every high input.

InputLowHighBasis
Digital applications started per year applications, applications per year100,000100,000The reference bank.
Share of started applications that are abandoned today abandonRate, fraction of applications0.30.5Editorial assumption, replace with your own funnel data.
Share of abandoned applications the assistant recovers recoveredShare, fraction of abandoned applications0.10.2Editorial assumption, deliberately conservative because no deployment on this page discloses a completion uplift.
First year revenue of a new account valuePerAccount, USD per account50150Editorial assumption, replace with your own product economics.

What it leaves out: Revenue from recovered applications only. It leaves out the review hours saved in operations, lower fraud losses, the cost of the identity verification provider and the platform, and the lifetime value of an account beyond the first year.

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.

Deutsche Bank

Germany · Banking · 2026

ProductionGrade C

Deutsche Bank Private Bank put an agentic AI solution live in its Singapore and Hong Kong booking centres that researches, documents and prepares Source of Wealth assessments, which the bank calls one of the most resource intensive parts of know your customer checks. It reads client documents alongside approved external data, flags gaps and inconsistencies, and hands the assessment to bank staff for review; relationship managers in Dubai use it for accounts booked in Singapore. The bank stresses that accountability stays with its people. No outcome figures are disclosed; the growth figures in the coverage are forecasts.

No outcome disclosed.

Albo

Mexico · Banking · 2025

ProductionGrade C

Albo, a Mexican neobank, uses Gemini models in Albot, a chatbot that handles customer onboarding and offers support and financial guidance around the clock to millions of first time banking users. Google Cloud lists the deployment and says it supports financial inclusion and regulatory compliance; no outcome figures are disclosed.

No outcome disclosed.

M-DAQ Global

Singapore · Payments and cards · 2024

ProductionGrade C

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

  • Product eligibility rules and approved product descriptions per market
  • The bank's know your customer and anti money laundering policy, including risk ratings and document lists per jurisdiction
  • Funnel data showing where applicants abandon today
  • Labelled historical cases of manual review outcomes, to test routing

Systems to integrate

  • Identity verification provider (document authenticity, liveness, face match)
  • Sanctions, politically exposed person and adverse media screening
  • Government digital identity schemes where they exist
  • Customer onboarding or account origination platform and core banking
  • Case management for manual review, with the assistant's summary attached

Complexity: High

The conversation is the easy part. The work is in orchestrating identity verification, screening and core account opening systems, keeping the rules configurable per jurisdiction, and producing an audit trail that satisfies compliance and the regulator.

  1. 1

    Map the funnel and the rules

    Take the current application funnel and mark where applicants drop out and why. Write the verification rules per product and jurisdiction as configuration, not as prompts, so compliance can own and change them.

  2. 2

    Start with guidance and document help

    The first release explains products, answers questions and helps applicants get their documents right. It changes no decisions, so it can ship early and still move completion.

  3. 3

    Automate the checks behind the form

    Add the agentic workflow that runs screening and collects evidence, with every tool call logged. Measure how often the reviewer agrees with the prepared summary before you let any case skip review.

  4. 4

    Set the routing thresholds with compliance

    Agree which cases may proceed without a human and which must be reviewed (high risk countries, politically exposed persons, mismatched data, low verification confidence). Keep human approval for every edge case.

  5. 5

    Test against fraud as well as friction

    Build a test set with genuine applicants who struggle (poor lighting, unusual names, foreign documents) and with attack patterns (edited documents, replayed selfies, synthetic identities), and run it on every change.

Guardrails

  • Verification rules are configuration owned by compliance, never free text instructions to a model
  • The assistant cannot approve an edge case; ambiguous or high risk cases always go to a person
  • Personal data and identity images are masked in logs and never used for model training without consent
  • Clear disclosure of what data is collected, why, and that the applicant is talking to AI
  • No personal financial advice during product selection, only eligibility and product facts

KPIs to instrument

  • Application completion rate per step and per channel, before and after
  • Median time from start to account open
  • Share of cases sent to manual review, and reviewer agreement with the prepared summary
  • Fraud found after onboarding in accounts that went straight through
  • Applicant satisfaction and contact centre calls about applications

Human in the loop

Onboarding analysts review every case the rules mark as unclear or high risk, with the assistant's summary and evidence in front of them, and they own the approval. Compliance samples straight through cases every week and signs off every change to thresholds.

Common failure modes

Fast onboarding for fraudsters
Straight through rules tuned for conversion let synthetic identities in. Track fraud on new accounts by route and tighten thresholds when it rises.
Silent unfair rejection
Document or liveness checks fail more often for some groups of applicants, who then give up. Monitor failure rates by document type, age band and language and offer an assisted route.
One rulebook for many countries
A single process copied across markets breaks local rules. Keep verification logic configurable per jurisdiction with a named owner.
Unexplainable decisions
A reviewer or regulator cannot see why a case was routed. Log every check, its result and the rule that applied.

What are the risks and rules?

EU AI Act

Depends on design

The conversational assistant falls under the Article 50 transparency duty. Biometric verification whose sole purpose is to confirm that a person is who they claim to be is excluded from the Annex III biometric category. The system becomes high risk when the same journey assesses creditworthiness or a credit score of a natural person, for example for a credit card or overdraft (Annex III point 5(b)).

Guidance

Controls to put in place

  • Documented verification rules per jurisdiction with a named compliance owner
  • Audit trail of every check, tool call and routing decision per application
  • Bias monitoring of verification failure rates across applicant groups
  • Human approval for every case outside the straight through rules
  • AI disclosure and a data collection notice at the start of the journey

Frequently asked questions

Can AI approve new customers without a human?
A bank can let its rules approve clean, low risk cases straight through, with the AI preparing the evidence. Unclear or high risk cases should always reach a person. Deutsche Bank describes its Source of Wealth agent this way: it prepares assessments for staff to review and accountability stays with people.
Is identity verification with face matching high risk under the EU AI Act?
Not by itself. Annex III excludes biometric verification whose only purpose is to confirm that someone is who they claim to be. The journey becomes high risk if it also scores the applicant's creditworthiness.
Where does AI help most in onboarding?
In two places: helping applicants finish (explaining products and fixing document problems in the moment) and preparing the checks for reviewers. The deployments on this page show both, from a neobank chatbot for first time bank users to automated Know Your Business and Source of Wealth research.

How to cite this page

Blits.ai AI Use Case Library, "AI assistant for digital account onboarding and KYC", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/digital-onboarding-assistant. Licensed under CC BY 4.0. Method: how we verify use cases.

Changelog
  • 27 September 2026: First published

Related use cases

BankingPayments and cards

AI for business onboarding (KYB) and beneficial ownership discovery

An AI agent that builds the know your business (KYB) due diligence file for a new or reviewed corporate client, before any account is opened: it collects registry, incorporation and ownership documents, resolves the entity across sources, maps the ownership chain through holding companies, nominees and trusts to the ultimate beneficial owners, screens the entity and its owners, and presents a risk scored case for a compliance analyst to decide.

Deployments
3 public, best grade C
Reported automation rate
25%
BNY, organization claim
BankingPayments and cards

AI for application and identity fraud detection

AI that checks incoming account and loan applications for forged or AI generated documents, synthetic and stolen identities, and coordinated application rings, by analysing documents, device and application data across the whole queue and cross checking against bureau and official sources.

Deployments
6 public, best grade B
Reported detection improvement
2.5x
Department for Work and Pensions, organization claim
BankingPayments and cards

AI for perpetual KYC and event driven customer due diligence

AI that keeps each customer's due diligence file current by replacing calendar driven KYC reviews with continuous, event driven refreshes: it watches for trigger events such as a change of ownership, address, behaviour or a new adverse finding, refreshes the file automatically where it can, and involves an analyst only when something material has changed. The risk rating itself and the first file for a new business client are separate use cases.

Deployments
5 public, best grade B
Reported cost reduction
40%
JPMorgan Chase, organization claim
BankingPayments and cards

AI for PEP and adverse media screening

AI that continuously scans news, court records, registries and other open sources in many languages for negative information and political exposure linked to customers, counterparties and beneficial owners, discards look alikes, and summarises credible risk for the analyst with the sources attached.

Deployments
7 public, best grade B
Reported handling time reduction
at least 60%
Save the Children, vendor claim
BankingPayments and cards

AI agent for account and card servicing

An AI agent that resolves routine account and card requests end to end, such as balances, statements, card blocks and replacements, PIN resets and limit changes, across app, web, messaging and phone, and hands anything sensitive or unusual to a human with the full context.

Deployments
2 public, best grade B
Reported containment rate
about 90%
DBS Bank, organization claim
Banking

Conversational AI for loan application intake

A conversational assistant on web, app, messaging or voice that explains loan products, captures the application through dialogue in the customer's language, checks documents and basic eligibility rules, and hands a complete, structured application to origination, without making the credit decision.

Deployments
5 public, best grade C
Autonomy
Supervised agent