A new customer decides to open an account with your bank. They start the application on their phone, hit a request for a document they don’t have to hand, and close the tab. In most institutions, that’s where the story quietly ends: the half-finished application sits in a queue, a generic reminder email goes out days later, and the applicant has already moved on — often to a competitor. The frustrating part is that nothing about the bank’s compliance obligations caused the loss. The applicant didn’t fail identity verification. They simply gave up on a process nobody was there to help them through.

This is the gap an AI onboarding agent is built to close. Not by touching the regulated parts of Know Your Customer (KYC) — those stay firmly with the bank’s compliance systems and people — but by owning the conversation around them: explaining what’s needed, answering questions at 2am, chasing incomplete applications across every channel, and handing the bank a complete, auditable record of every interaction.

Onboarding abandonment is a measurable, expensive problem

Banks don’t lose most applicants at the credit decision or the identity check. They lose them earlier, in the fog between “I’d like an account” and “here is everything you need from me”.

The research on this is stark. Signicat’s Battle to Onboard study found that 68% of European consumers abandoned a financial services application within the previous year — up from 63% in its 2020 edition. The same research found the average applicant gave up after just 18 minutes and 53 seconds, and that 38% of respondents abandoned an application because they didn’t have the right identity credentials to hand, such as a passport or digital identity. A further 21% walked away because they were asked for too much personal information, and 30% simply found the process complicated.

Read those numbers together and a pattern emerges. Applicants aren’t abandoning because banks ask for identity documents — they’re abandoning because nobody explains why a document is needed, what counts as acceptable, or how to finish later when they’ve found it. A static web form can’t answer “will my expired passport work?” or “can I use my Medicare card instead?”. A branch can, but only during business hours, in one language, for people willing to queue.

That’s a conversation problem, not a compliance problem. And conversation problems are exactly what autonomous agents are good at.

What KYC actually requires — and what it doesn’t

Before looking at what an agent can do, it’s worth being precise about what KYC involves, because the boundary between “helping an applicant” and “performing regulated verification” is one a bank cannot afford to blur.

At a high level, KYC (a core part of customer due diligence) is jurisdiction-specific in detail but broadly consistent in shape: before a financial institution provides regulated services to a new customer, it must collect identifying information, verify that the person is who they claim to be using reliable and independent sources, assess the risk the customer presents, and keep records demonstrating how it did all of the above.

In Australia, these obligations sit under the Anti-Money Laundering and Counter-Terrorism Financing Act 2006, overseen by AUSTRAC. AUSTRAC’s guidance requires reporting entities to complete initial customer due diligence before providing a designated service — establishing on reasonable grounds that a customer is who they claim to be, based on information such as full name, date of birth and residential address. A bank cannot begin providing the service until it is satisfied of the customer’s identity on reasonable grounds. Record-keeping obligations are equally serious: AUSTRAC requires records showing how initial customer due diligence was carried out, retained for seven years after the business relationship ends. And the bar is rising, not falling — reforms to Australia’s AML/CTF regime commenced for existing reporting entities on 31 March 2026, with newly regulated “tranche 2” sectors following from 1 July 2026.

Notice what’s in that list: collection, verification against independent sources, risk assessment, approval, record-keeping. These are the bank’s regulated obligations, executed by the bank’s compliance systems and accountable humans. Nothing in the AI onboarding agent pattern replaces any of it.

What the regulations don’t prescribe is how pleasant, fast or multilingual the journey around those checkpoints should be. A bank can meet every obligation with a hostile, confusing process — most do — or it can meet the same obligations with an applicant who was guided, informed and gently chased to completion. Compliance is fixed; experience is a choice.

The pattern: a conversational layer in front of compliance, never instead of it

Zian’s Bank Registration Onboarding (KYC) agent is one of a family of niched agents built on the same foundation as its autonomous AI agents platform. The pattern has four jobs, and a deliberately hard boundary around all of them.

1. Walk applicants through requirements. The agent explains, in plain language, what the bank needs at each step: which identity documents are acceptable, what a certified copy means, why date of birth and residential address are being collected, and what happens next. It draws on research, web and knowledge-base lookups, so answers reflect the bank’s own documented policies rather than generic guesses.

2. Answer questions 24/7, in 30+ languages. Applicants apply at night, on weekends, and in whatever language they think in. Like Zian’s AI customer support agents, the onboarding agent is always on and operates in more than 30 languages — which matters enormously in a market like Australia, where a large share of new-to-bank customers are also new to the country and to its identity-document conventions.

3. Chase incomplete applications across channels. When an application stalls, the agent follows up by live phone, SMS, email and WhatsApp — choosing channel and timing intelligently rather than firing one templated email into the void. More on this below, because it’s where most of the recovered revenue lives.

4. Hand over a complete, auditable record. Every interaction the agent has with an applicant is logged. When the file reaches the bank’s compliance team, it arrives with the full conversation history attached: what was asked, what was explained, what was promised, and when.

And the boundary: the agent does not verify identity documents, does not screen applicants against watchlists, does not assess money-laundering risk, and does not approve or decline anyone. Those functions remain with the bank’s own verification systems and its human compliance officers, exactly as regulators expect. The agent’s output is a better-prepared applicant and a better-documented file — the decision inputs, never the decision.

Traditional onboarding vs agent-assisted onboarding

Dimension	Traditional form-and-queue onboarding	Agent-assisted onboarding
Guidance	Static forms, help pages and hold music; applicants interpret requirements alone	Conversational walkthrough of each requirement, with questions answered in the moment
Availability	Branch and call-centre hours; queues at peak times	24/7, including nights, weekends and public holidays
Language	Usually English-first, with limited interpreter access	30+ languages, in the applicant’s preferred channel
Incomplete applications	One or two templated reminder emails, then silence	Persistent, intelligently paced follow-up across phone, SMS, email and WhatsApp
Compliance decisions	Bank’s compliance systems and staff	Unchanged — still the bank’s compliance systems and staff
Record of the journey	Fragmented notes across branch, call centre and email	Complete, logged conversation record handed to the compliance team
Data location	Bank-controlled systems	Bank-controlled — private model deployment on the bank’s own infrastructure is supported

The follow-up layer: where abandoned applications get recovered

If 68% of applicants abandon and the average one gives up in under 19 minutes, the single highest-leverage capability in onboarding isn’t a prettier form — it’s what happens in the hours and days after someone stalls.

Most banks’ answer is a reminder email or two. The agent pattern treats an incomplete application the way a disciplined sales team treats a warm lead: as something to be followed up persistently, politely and across channels until it resolves one way or the other. Zian’s SmartReach AI™ orchestrates the message, the channel and the timing by country, industry and profile, with intelligent follow-up pacing — so an applicant who ignored an email might get a WhatsApp message the next morning, then a short call a few days later, each one referencing exactly where they left off and exactly what’s still needed.

The persistence gap this closes is enormous. Across Zian’s platform, teams have seen a 926% increase in follow-ups and 28x more contact attempts compared with their previous manual processes. Those figures come from sales deployments, but the mechanism transfers directly: abandoned applications, like unworked leads, mostly die of neglect rather than rejection. An applicant who abandoned because they couldn’t find their passport on Tuesday is often perfectly happy to finish on Saturday — if anyone remembers to ask, in a channel they actually read.

PrecisionPitch AI™ adds a second layer: continuous split-testing of scripts and approaches, optimised for real completion outcomes rather than opens or clicks. Over time, the agent learns which explanations, which channels and which pacing actually get applications finished — per segment, not on average.

The auditable conversation record — and who holds the data

For a compliance team, an AI that talks to applicants raises an immediate and entirely fair question: what exactly did it say? Regulated onboarding is not a place for unlogged, unreviewable conversations.

This is why the auditable record is a first-class part of the pattern rather than an afterthought. Every interaction — every call, SMS, email and WhatsApp exchange — is logged and attached to the applicant’s file. When a human compliance officer picks up the application, they can see the full history: what the applicant was told about document requirements, what questions they asked, what follow-ups were sent and when. If a regulator or internal auditor later asks how an applicant was handled, the answer is a record, not a reconstruction. That sits comfortably alongside the record-keeping discipline AUSTRAC already demands of reporting entities.

The second half of the answer is data control. Zian supports private model deployment on the customer’s own infrastructure — meaning a bank can run the agent inside its own environment, under its own security policies and data-residency arrangements, rather than routing applicant conversations through shared external systems. For institutions whose risk teams treat data location as non-negotiable, that capability is usually the difference between “interesting” and “deployable”. On the systems side, the platform offers API and CRM integrations (HubSpot, Salesforce, HighLevel, Zapier) for connecting conversation records and application status into existing workflows.

The same conversational vetting pattern shows up elsewhere in Zian’s niched agent family — from professional recruitment screening to university application vetting — because the underlying problem is identical: a structured process that humans must decide, wrapped in a conversation that machines can carry.

Frequently asked questions

Does the AI make KYC decisions or verify identity?

No. The AI onboarding agent handles the conversation and follow-up layer only: explaining requirements, answering applicant questions and chasing incomplete applications. Identity verification, watchlist screening, risk assessment and final KYC approval decisions remain entirely with the bank’s own compliance systems and human compliance team. The agent hands over a complete conversation record as an input to those decisions — it never makes them.

How does the bank keep control of applicant data?

Zian supports private model deployment on the customer’s own infrastructure, so a bank can run the onboarding agent inside its own environment under its own security and data-residency policies. Every interaction is logged, giving the bank a complete, auditable record of each applicant conversation within systems it controls.

Which channels can the onboarding agent use to reach applicants?

The agent operates across live phone, SMS, email and WhatsApp. It chooses the channel and timing intelligently for each applicant, so someone who ignores an email reminder can be reached by SMS or a call instead, with each message referencing exactly what is still outstanding on their application.

Can the agent help applicants who don’t speak English?

Yes. The agent operates in more than 30 languages, 24 hours a day. Applicants can ask questions about identity and document requirements in their preferred language at any time, which is particularly valuable for new migrants who are unfamiliar with local identity-document conventions.

Why do so many people abandon bank onboarding in the first place?

Signicat’s Battle to Onboard research found 68% of European consumers abandoned a financial services application within a year, with the average applicant giving up after under 19 minutes. Leading causes included not having the right identity credentials to hand (38%), being asked for too much personal information (21%) and finding the process complicated (30%) — largely conversation and guidance failures rather than compliance failures.

Is the Zian onboarding agent available today?

Zian is currently in waitlist beta. Banks and financial institutions interested in the Bank Registration Onboarding (KYC) agent pattern can join the waitlist at zian.ai to be contacted as access opens up.

If your onboarding funnel leaks applicants your compliance process never even got to assess, the fix isn’t weaker checks — it’s a better conversation around them. Join Waitlist to see how Zian’s Bank Registration Onboarding (KYC) agent guides applicants to complete, well-documented applications while your compliance team keeps every decision.