Account opening for people who have just moved to a new country. An established bank brings in AI agents to onboard newcomers faster; this is how I’d decide what the agents do, where a person steps in, and how to keep it fair.
Most people who start an application online never finish it, and newcomers have the most reasons to stop.
The first account unlocks everything else: salary, rent, a phone contract.
AI agents for the routine, a named person at the moments that decide, and safeguards around both.
Time to a working account, a first salary on time, and no gaps between groups.
A fictional established retail bank wants to win people in their first week, before a digital bank does. It is bringing AI agents into account opening.
Employers bring people in all year. Each of them needs an account before their first salary.
Competitors open an account and put a card in the phone wallet in minutes. The established bank has trust, branches and products, but a slower door.
The bank is rolling out AI agents across onboarding. Done well, they remove waiting. Done badly, they trap people who need a person.
The Head of Retail’s brief“Use AI to open accounts faster for people who have just arrived, without losing the ones who need a person. And make sure we can explain every decision it makes.”
The traps I’d rule out on day one, so speed never costs trust.
The newcomer opens the account. The employer needs it working. The frontline catches whatever the AI can’t finish.
Just landed. Residency and ID may still be in progress, there is no local credit history, and the first salary is weeks away.
What they needStuck at 80% of an application, with a salary that can’t be paid and nobody to ask.
Brings in new hires every month and needs every one of them paid on time.
What they needA new hire who can’t be paid in their first month.
Branch and contact-centre staff who pick up the cases the AI can’t finish.
What they needA customer who has to explain everything again, and is already angry.
Newcomers often settle on a main bank in their first weeks. A slow or confusing start can send them to a competitor.
Most of it happens before and after the form, not in it. These are the moments the journey has to answer.
An address to open the account, an account to rent the flat. One missing piece stalls everything.
Lists differ by bank, by branch and by the person answering.
The ID card hasn’t arrived yet, and the salary can’t wait for it.
The hardest questions come in someone else’s language, at the moment that matters most.
Frontline staff are trained to be cautious, and rarely have room to make an exception.
Everyone pays with their phone; a newcomer waiting days for a card can’t buy groceries.
The real deadline isn’t the application. It’s the first payday.
Transfers to family from week one, and no local credit history for anything else.
In a 2025 central bank study of 722 frontline staff, about a third said they couldn’t open an account for someone at least monthly, and 76% were encouraged to be cautious. Recent migrants were among the groups most affected.*
Digital wallets made up 33% of in-store spending and 56% of online spending worldwide in 2025.* A newcomer expects to pay with the phone on day one.
More than half of people who start a digital account application never finish it; in one poll of 14,000+ customers, more than half dropped out at ten questions or more.*
The bank already sees where people stop. The evidence just sits with different teams.
Where people stop, by step, device and language.
Where the agent loops, guesses or can’t answer.
What people call about after starting online.
What they say in public, and to the regulator.
Who is stuck, and what it costs them.
On day 1 and day 30, in the customer’s language.
Hypotheses, to be tested against the bank’s own data. Nothing will be named.
Each part of the journey answers a hypothesis. If the first 90 days confirm it, this is what changes. If they don’t, the journey changes too.
Sends the document checklist through the employer, in the newcomer’s language.
HR has a contact at the bank for anyone with an unusual case.
Guides the application on a phone, scans documents and fills in what it can read.
If the newcomer asks, or gets stuck twice on the same step.
Checks documents and flags anything missing, with a reason.
Every exception: residency in progress, a mismatch, anything the agent isn’t sure of.
Adds a digital card to the phone wallet, so they can pay from day one, and sends the account details for payroll.
A welcome call for anyone who asked for one. A plastic card only if they want it.
Confirms the salary arrived; flags it if it didn’t.
If the salary is missing, a person calls the same day.
Asks what still feels hard, in two questions.
A follow-up for anyone who says something is wrong.
The lens I bring from hospitality: an arrival is judged in the first hours. At Hotel Paris, a person was always there when it mattered; here, the AI is the front desk and a person is never more than one tap away.
I’m the bank’s AI assistant. Your residency permit is still being processed, so I can’t finish this step on my own.
I’ve passed your application to an onboarding specialist, with everything you’ve shared. They speak your language and will call you today at 15:00. You won’t need to start again.
And explains why it asks for each document.
With an honest wait time, in the customer’s language.
The person receives the full case before saying hello.
The person has the authority to solve it.
The same person owns the case until the account works.
In a 2026 survey of 2,500+ consumers, 81% wanted some human involvement when they called their bank, and fewer than 1 in 5 would trust AI to make financial recommendations on its own.*
In another 2026 study of 3,000 consumers, 55% trusted AI customer service when a clear path to a person existed, against about a quarter when it didn’t.*
The agent says it is AI, explains what it needs and why, and shows what it decided.
Asks only for what regulation requires, once, and reuses it with consent.
Tested in every supported language; completion is tracked by group, and any gap has an owner.
No application is rejected by AI alone. A person reviews every “no”.
Every agent has a named owner, an audit trail, and regular reviews with compliance.
When the agent isn’t sure, it hands over. It never guesses on identity or money.
Run the journey with one employer’s new hires, in two or three languages, and listen on day 1 and day 30 before scaling.
An agentic AI product I designed from 170+ interviews, where the AI suggests and the person decides.
Read the case