How banks improve KYB conversion without cutting compliance corners

On this page
The best way for a bank to improve business onboarding conversion without cutting compliance corners is to stop treating speed and control as a trade-off. Verification gets risk-based. Data is pulled from authoritative sources instead of requested from applicants. Policy runs as code, so low-risk cases clear automatically and high-risk cases reach an analyst with the evidence already assembled. Every step writes to an audit trail.
This case study walks through that model as a bank would implement it on Duna. It is a reference scenario, not a profile of one named client. Where figures appear, they are either arithmetic you can check or outcomes Duna has published across its customers (mostly fintechs, not banks). The baseline table is a template your own data should replace.
The scenario: a multi-country bank losing business applicants inside KYB
Picture a mid-sized European bank that onboards small and mid-sized business customers across several countries. Most applicants arrive remotely, a minority through relationship managers. Business customers are the revenue priority, yet onboarding a company is a different job from onboarding a person. The bank must confirm the legal entity exists, identify the ultimate beneficial owners (UBOs), verify the people who can act for the company, and screen all of them.
The conversion problem shows up in predictable places:
Applicants abandon long forms that ask for data the bank could look up itself.
UBO details arrive incomplete or inconsistent with the registry, which triggers a follow-up.
Document uploads fail review for small discrepancies (a name variant, an expired extract), which triggers another.
Analysts handle low-risk and high-risk cases the same way, so queues grow and decision times stretch.
Each follow-up matters more than it looks. Duna's analysis of onboarding as a conversion problem puts the loss at roughly 15% of submitted customers per follow-up request. The arithmetic compounds. An application that needs three follow-ups has a 0.85 × 0.85 × 0.85 chance of surviving, or about 61%. The bank has done the acquisition work, the applicant has submitted everything asked for, and nearly four in 10 are still gone.
The constraint is audit pressure. The bank can't fix this by skipping checks. FATF's March 2020 Guidance on Digital Identity frames customer due diligence (CDD) under Recommendation 10 as a risk-based exercise. It asks institutions to assess whether a digital identity approach is sufficiently reliable and independent to mitigate AML/CFT risk. For remote onboarding in the EU, the EBA's Guidelines on the use of remote customer onboarding solutions (EBA/GL/2022/15) require institutions to assess the adequacy, reliability, and choice of their non-face-to-face identification tools. They also expect documented policies and processes. Any improvement has to survive both tests.
Baseline metrics to lock before changing anything
A conversion claim is only as good as its definitions. Before the first journey is rebuilt, the bank fixes a baseline window (a full quarter is usually enough to smooth seasonality) and agrees on how each KPI is counted.
KPI | Definition to agree on | Where friction usually hides |
|---|---|---|
Business onboarding conversion | Applications approved ÷ applications started (also track submitted ÷ started) | Form length, unclear data requests |
Time-to-decision | Application submitted to final accept/reject, in hours | Document review queues, follow-ups |
Follow-up rate | Share of cases needing at least one applicant request | UBO data, document discrepancies |
Analyst touch rate | Manual reviews per case | Uniform review regardless of risk |
Exception rate | Cases leaving the standard path | Registry gaps, inconclusive checks |
Straight-through rate | Cases decided with no analyst action | Rules that can't express policy |
Two details trip teams up. The first is the starting event for time-to-decision: counting from submission hides the abandonment that happens before it. The second is the denominator for conversion. Track "started" and "submitted" separately, because they point to different fixes. A gap between started and submitted is a UX problem. A gap between submitted and approved is a data and review problem.
What changed in the journey
Ask the applicant for less
The first intervention is cutting what the applicant has to type. In Duna Onboard, journeys are built without code, and the form adapts to the company. An applicant enters a company identifier and country. The platform pulls entity details from the relevant local registry, pre-fills the form, and asks the applicant only to confirm or correct it. Duna covers 210+ local registries and supports 7+ languages, which matters for a multi-country bank because localization gaps are a leading source of KYB data holes and the follow-ups that come with them.
Requests are also explained. If the bank needs a UBO's date of birth, the form says why. Intelligent reminders and in-platform messaging keep any remaining follow-up inside one auditable channel, not scattered across email threads.
Orchestrate checks instead of queuing them
Behind the form, checks run in parallel rather than in sequence: registry lookup, UBO register matching where available, document verification, and screening against PEP, sanctions, and adverse media lists. Duna's data platform routes each request across providers based on performance, cost, and coverage. If a provider fails or returns nothing, the request falls back to another source without an analyst stepping in. AI agents handle document verification, extracting and normalizing evidence so reviewers don't re-key it. Duna's piece on AI in KYC and AML covers how that extraction step cuts manual intake while governance stays intact.
Encode the policy, including when to escalate
This is the intervention that makes speed defensible. The bank's compliance team defines what an accepted customer looks like, and Duna's policy engine turns that definition into code. Policy changes can be back-tested before they go live, and compliance can update them without waiting on engineering.
A simplified version of the rules:
A low-risk company with a registry match, verified UBOs, a clean screening result, and consistent documents is accepted automatically.
A company with a complex ownership chain, a high-risk jurisdiction exposure, or a PEP hit moves to enhanced due diligence (EDD). The policy requests additional data from the applicant at that point, not from everyone upfront.
A company where a check is inconclusive routes to an analyst with the evidence attached.
The result is risk-based customer onboarding in a literal sense. Data collection scales with risk, so a low-risk applicant never sees the questions an EDD case needs.
Govern the remote identity checks
Remote verification of legal representatives uses document verification with liveness detection and document integrity checks. The governance around those tools matters as much as the tools. Following the EBA guidelines, the bank documents why each tool is adequate for the risk it addresses, how its reliability is tested, and what happens when it fails. The same reasoning applies to the FATF reliability and independence test: the bank can show that identity evidence comes from sources independent of the applicant. Deepfakes and synthetic documents are part of that risk assessment now, and integrity checks sit in the policy as explicit conditions, not assumptions.
One EBA expectation shapes the design: institutions shouldn't process a customer before initial CDD is complete. Speed has to come from completing CDD faster, not from deferring it. The journey is built accordingly, with account activation gated on the accept decision.
Before and after
Step | Before | After |
|---|---|---|
Data entry | Applicant types all entity and UBO details | Registry pre-fill; applicant confirms |
Verification | Sequential, mostly manual | Parallel, orchestrated with provider fallback |
Document review | Analyst reviews every file | AI extraction; analyst reviews exceptions |
Routing | Same review path for all risk levels | Policy-driven: auto-accept, EDD, or manual |
Follow-ups | Email threads | In-platform messaging, logged |
Record | Spread across systems | One case file with a full audit trail |
Compliance safeguards and auditability
What the audit trail records
Every case produces a single record. It captures the applicant's inputs, the data returned by each provider and registry, the policy version applied, each rule outcome, the final decision, timestamps, and every analyst action. Policy changes are logged too, so an examiner asking "why was this company accepted in March?" gets the data, the rule version, and the decision path, not a reconstruction. Because customer messages live inside the platform, communications are part of the same file.
Where humans stay in the loop
Automation doesn't remove analysts. It moves them to the cases where judgment matters. The bank sets thresholds for mandatory review (risk score bands, screening hits, ownership complexity) and enables four-eye review for high-risk decisions. Analysts work from automated case management queues, with evidence already assembled, so a review takes minutes of judgment rather than an hour of collection. The bank can also sample auto-accepted cases for periodic QA, which gives compliance a measurable check on false negatives.
What happens when a check can't conclude
Inconclusive results are the real test of a compliance design. If a registry is unavailable, smart routing tries an alternate provider. If no provider can answer, the case escalates to manual review instead of defaulting to approval. If a liveness or document integrity check fails, the policy decides between a retry, an alternative evidence path, and rejection. These fallbacks are written into the bank's documented procedures, which is what the EBA's governance expectations look for. Duna's KYB verification guide explains why verification breaks down without this kind of redundancy.
Results: what Duna customers report, and how to read them
Duna doesn't publish bank-specific figures for this reference scenario, and this article won't invent them. What exists is Duna-reported performance across its customer base: 10.6x faster onboarding, a 37% conversion increase, and 4.8x analyst efficiency. Plaid, Moss, and SeQura are among the customers behind the speed and productivity figures cited on Duna's fintech page. These are fintech results, so a bank should treat them as a directional benchmark rather than a promise.
Here is how a bank should translate those benchmarks into its own scorecard:
Conversion should be measured against the baseline definitions above, with submitted-to-approved and started-to-submitted reported separately.
Time-to-decision should fall most for low-risk cases, because those are the ones that stop waiting for a human. If high-risk cases get faster too, check that EDD isn't being skipped.
Straight-through rate is the lever behind both analyst efficiency and cycle time. Track it by risk tier.
Compliance quality needs its own proof: QA pass rates on sampled auto-accepts, rework rates, and the share of escalations analysts confirm as warranted. If speed rises and these hold steady, the controls held.
A bank that reports only the speed numbers hasn't finished the case study. The compliance proxies are what an auditor will ask for.
Lessons that apply to any bank
Ask less, not later. Most conversion loss comes from requests, not from checks. Pre-filling from registries and requesting EDD data only when the policy calls for it removes follow-ups without removing diligence. Visa's "Modernizing SMB onboarding to drive growth" argues for the same move from rigid, compliance-first flows to adaptive, risk-based ones. The difference is that a bank can measure it. Duna's piece on why onboarding belongs on the COO agenda makes the same case from the operations side.
Route exceptions by policy, not by habit. When every case gets the same review, analysts become the bottleneck and low-risk customers wait as long as risky ones. Written risk tiers and thresholds fix that, and they give compliance a single place to adjust when risk appetite changes.
Build the audit trail first. Teams that bolt logging on afterward discover gaps during the first examination. Deciding up front what gets logged, with policy versions and analyst actions included, makes the system defensible from the first live case.
Test the fallbacks before launch. Provider outages and inconclusive results will happen. A pilot that deliberately triggers them shows whether cases escalate or leak through.
Implementation timeline and next steps
Durations depend on how many integrations and policy variants a bank has, so treat these as a planning template, not a guarantee. Duna Onboard's no-code configuration shortens the journey-building phases considerably.
Phase | Typical focus | Planning range |
|---|---|---|
Discovery | Baseline KPIs, current policy, data sources, risk tiers | 2 to 3 weeks |
Policy and journey design | Rules, thresholds, EDD triggers, UX with pre-fill | 3 to 4 weeks |
Integrations | Core banking handoff, providers, screening, fallbacks | 3 to 6 weeks |
Pilot | One segment or country, QA sampling, fallback testing | 4 to 6 weeks |
Rollout | Additional countries and segments, dashboards, training | Staged |
By the end of rollout the bank has live onboarding journeys, orchestrated data routing, automated case management with reviewer queues, monitoring dashboards, and audit trail outputs ready for supervisory requests.
The next steps after launch are mostly about breadth and quality. Expand registry and language coverage into new markets. Tune rules using back-testing against actual outcomes, which raises the straight-through rate without relaxing standards. Then extend the same policy logic to the customer lifecycle: daily screening, periodic reviews, and policy-driven re-KYC, so the discipline applied at onboarding continues afterward. Banks exploring this can see how Duna approaches compliance infrastructure for banks and how Duna Onboard fits into a business onboarding program.
The best way for a bank to improve business onboarding conversion without cutting compliance corners is to stop treating speed and control as a trade-off. Verification gets risk-based. Data is pulled from authoritative sources instead of requested from applicants. Policy runs as code, so low-risk cases clear automatically and high-risk cases reach an analyst with the evidence already assembled. Every step writes to an audit trail.
This case study walks through that model as a bank would implement it on Duna. It is a reference scenario, not a profile of one named client. Where figures appear, they are either arithmetic you can check or outcomes Duna has published across its customers (mostly fintechs, not banks). The baseline table is a template your own data should replace.
The scenario: a multi-country bank losing business applicants inside KYB
Picture a mid-sized European bank that onboards small and mid-sized business customers across several countries. Most applicants arrive remotely, a minority through relationship managers. Business customers are the revenue priority, yet onboarding a company is a different job from onboarding a person. The bank must confirm the legal entity exists, identify the ultimate beneficial owners (UBOs), verify the people who can act for the company, and screen all of them.
The conversion problem shows up in predictable places:
Applicants abandon long forms that ask for data the bank could look up itself.
UBO details arrive incomplete or inconsistent with the registry, which triggers a follow-up.
Document uploads fail review for small discrepancies (a name variant, an expired extract), which triggers another.
Analysts handle low-risk and high-risk cases the same way, so queues grow and decision times stretch.
Each follow-up matters more than it looks. Duna's analysis of onboarding as a conversion problem puts the loss at roughly 15% of submitted customers per follow-up request. The arithmetic compounds. An application that needs three follow-ups has a 0.85 × 0.85 × 0.85 chance of surviving, or about 61%. The bank has done the acquisition work, the applicant has submitted everything asked for, and nearly four in 10 are still gone.
The constraint is audit pressure. The bank can't fix this by skipping checks. FATF's March 2020 Guidance on Digital Identity frames customer due diligence (CDD) under Recommendation 10 as a risk-based exercise. It asks institutions to assess whether a digital identity approach is sufficiently reliable and independent to mitigate AML/CFT risk. For remote onboarding in the EU, the EBA's Guidelines on the use of remote customer onboarding solutions (EBA/GL/2022/15) require institutions to assess the adequacy, reliability, and choice of their non-face-to-face identification tools. They also expect documented policies and processes. Any improvement has to survive both tests.
Baseline metrics to lock before changing anything
A conversion claim is only as good as its definitions. Before the first journey is rebuilt, the bank fixes a baseline window (a full quarter is usually enough to smooth seasonality) and agrees on how each KPI is counted.
KPI | Definition to agree on | Where friction usually hides |
|---|---|---|
Business onboarding conversion | Applications approved ÷ applications started (also track submitted ÷ started) | Form length, unclear data requests |
Time-to-decision | Application submitted to final accept/reject, in hours | Document review queues, follow-ups |
Follow-up rate | Share of cases needing at least one applicant request | UBO data, document discrepancies |
Analyst touch rate | Manual reviews per case | Uniform review regardless of risk |
Exception rate | Cases leaving the standard path | Registry gaps, inconclusive checks |
Straight-through rate | Cases decided with no analyst action | Rules that can't express policy |
Two details trip teams up. The first is the starting event for time-to-decision: counting from submission hides the abandonment that happens before it. The second is the denominator for conversion. Track "started" and "submitted" separately, because they point to different fixes. A gap between started and submitted is a UX problem. A gap between submitted and approved is a data and review problem.
What changed in the journey
Ask the applicant for less
The first intervention is cutting what the applicant has to type. In Duna Onboard, journeys are built without code, and the form adapts to the company. An applicant enters a company identifier and country. The platform pulls entity details from the relevant local registry, pre-fills the form, and asks the applicant only to confirm or correct it. Duna covers 210+ local registries and supports 7+ languages, which matters for a multi-country bank because localization gaps are a leading source of KYB data holes and the follow-ups that come with them.
Requests are also explained. If the bank needs a UBO's date of birth, the form says why. Intelligent reminders and in-platform messaging keep any remaining follow-up inside one auditable channel, not scattered across email threads.
Orchestrate checks instead of queuing them
Behind the form, checks run in parallel rather than in sequence: registry lookup, UBO register matching where available, document verification, and screening against PEP, sanctions, and adverse media lists. Duna's data platform routes each request across providers based on performance, cost, and coverage. If a provider fails or returns nothing, the request falls back to another source without an analyst stepping in. AI agents handle document verification, extracting and normalizing evidence so reviewers don't re-key it. Duna's piece on AI in KYC and AML covers how that extraction step cuts manual intake while governance stays intact.
Encode the policy, including when to escalate
This is the intervention that makes speed defensible. The bank's compliance team defines what an accepted customer looks like, and Duna's policy engine turns that definition into code. Policy changes can be back-tested before they go live, and compliance can update them without waiting on engineering.
A simplified version of the rules:
A low-risk company with a registry match, verified UBOs, a clean screening result, and consistent documents is accepted automatically.
A company with a complex ownership chain, a high-risk jurisdiction exposure, or a PEP hit moves to enhanced due diligence (EDD). The policy requests additional data from the applicant at that point, not from everyone upfront.
A company where a check is inconclusive routes to an analyst with the evidence attached.
The result is risk-based customer onboarding in a literal sense. Data collection scales with risk, so a low-risk applicant never sees the questions an EDD case needs.
Govern the remote identity checks
Remote verification of legal representatives uses document verification with liveness detection and document integrity checks. The governance around those tools matters as much as the tools. Following the EBA guidelines, the bank documents why each tool is adequate for the risk it addresses, how its reliability is tested, and what happens when it fails. The same reasoning applies to the FATF reliability and independence test: the bank can show that identity evidence comes from sources independent of the applicant. Deepfakes and synthetic documents are part of that risk assessment now, and integrity checks sit in the policy as explicit conditions, not assumptions.
One EBA expectation shapes the design: institutions shouldn't process a customer before initial CDD is complete. Speed has to come from completing CDD faster, not from deferring it. The journey is built accordingly, with account activation gated on the accept decision.
Before and after
Step | Before | After |
|---|---|---|
Data entry | Applicant types all entity and UBO details | Registry pre-fill; applicant confirms |
Verification | Sequential, mostly manual | Parallel, orchestrated with provider fallback |
Document review | Analyst reviews every file | AI extraction; analyst reviews exceptions |
Routing | Same review path for all risk levels | Policy-driven: auto-accept, EDD, or manual |
Follow-ups | Email threads | In-platform messaging, logged |
Record | Spread across systems | One case file with a full audit trail |
Compliance safeguards and auditability
What the audit trail records
Every case produces a single record. It captures the applicant's inputs, the data returned by each provider and registry, the policy version applied, each rule outcome, the final decision, timestamps, and every analyst action. Policy changes are logged too, so an examiner asking "why was this company accepted in March?" gets the data, the rule version, and the decision path, not a reconstruction. Because customer messages live inside the platform, communications are part of the same file.
Where humans stay in the loop
Automation doesn't remove analysts. It moves them to the cases where judgment matters. The bank sets thresholds for mandatory review (risk score bands, screening hits, ownership complexity) and enables four-eye review for high-risk decisions. Analysts work from automated case management queues, with evidence already assembled, so a review takes minutes of judgment rather than an hour of collection. The bank can also sample auto-accepted cases for periodic QA, which gives compliance a measurable check on false negatives.
What happens when a check can't conclude
Inconclusive results are the real test of a compliance design. If a registry is unavailable, smart routing tries an alternate provider. If no provider can answer, the case escalates to manual review instead of defaulting to approval. If a liveness or document integrity check fails, the policy decides between a retry, an alternative evidence path, and rejection. These fallbacks are written into the bank's documented procedures, which is what the EBA's governance expectations look for. Duna's KYB verification guide explains why verification breaks down without this kind of redundancy.
Results: what Duna customers report, and how to read them
Duna doesn't publish bank-specific figures for this reference scenario, and this article won't invent them. What exists is Duna-reported performance across its customer base: 10.6x faster onboarding, a 37% conversion increase, and 4.8x analyst efficiency. Plaid, Moss, and SeQura are among the customers behind the speed and productivity figures cited on Duna's fintech page. These are fintech results, so a bank should treat them as a directional benchmark rather than a promise.
Here is how a bank should translate those benchmarks into its own scorecard:
Conversion should be measured against the baseline definitions above, with submitted-to-approved and started-to-submitted reported separately.
Time-to-decision should fall most for low-risk cases, because those are the ones that stop waiting for a human. If high-risk cases get faster too, check that EDD isn't being skipped.
Straight-through rate is the lever behind both analyst efficiency and cycle time. Track it by risk tier.
Compliance quality needs its own proof: QA pass rates on sampled auto-accepts, rework rates, and the share of escalations analysts confirm as warranted. If speed rises and these hold steady, the controls held.
A bank that reports only the speed numbers hasn't finished the case study. The compliance proxies are what an auditor will ask for.
Lessons that apply to any bank
Ask less, not later. Most conversion loss comes from requests, not from checks. Pre-filling from registries and requesting EDD data only when the policy calls for it removes follow-ups without removing diligence. Visa's "Modernizing SMB onboarding to drive growth" argues for the same move from rigid, compliance-first flows to adaptive, risk-based ones. The difference is that a bank can measure it. Duna's piece on why onboarding belongs on the COO agenda makes the same case from the operations side.
Route exceptions by policy, not by habit. When every case gets the same review, analysts become the bottleneck and low-risk customers wait as long as risky ones. Written risk tiers and thresholds fix that, and they give compliance a single place to adjust when risk appetite changes.
Build the audit trail first. Teams that bolt logging on afterward discover gaps during the first examination. Deciding up front what gets logged, with policy versions and analyst actions included, makes the system defensible from the first live case.
Test the fallbacks before launch. Provider outages and inconclusive results will happen. A pilot that deliberately triggers them shows whether cases escalate or leak through.
Implementation timeline and next steps
Durations depend on how many integrations and policy variants a bank has, so treat these as a planning template, not a guarantee. Duna Onboard's no-code configuration shortens the journey-building phases considerably.
Phase | Typical focus | Planning range |
|---|---|---|
Discovery | Baseline KPIs, current policy, data sources, risk tiers | 2 to 3 weeks |
Policy and journey design | Rules, thresholds, EDD triggers, UX with pre-fill | 3 to 4 weeks |
Integrations | Core banking handoff, providers, screening, fallbacks | 3 to 6 weeks |
Pilot | One segment or country, QA sampling, fallback testing | 4 to 6 weeks |
Rollout | Additional countries and segments, dashboards, training | Staged |
By the end of rollout the bank has live onboarding journeys, orchestrated data routing, automated case management with reviewer queues, monitoring dashboards, and audit trail outputs ready for supervisory requests.
The next steps after launch are mostly about breadth and quality. Expand registry and language coverage into new markets. Tune rules using back-testing against actual outcomes, which raises the straight-through rate without relaxing standards. Then extend the same policy logic to the customer lifecycle: daily screening, periodic reviews, and policy-driven re-KYC, so the discipline applied at onboarding continues afterward. Banks exploring this can see how Duna approaches compliance infrastructure for banks and how Duna Onboard fits into a business onboarding program.
Continue reading
Industries
Customers
Company

Industries
Customers
Company

Industries
Customers
Company





