On this page

This guide walks a bank's compliance and operations team through building a re-KYC program for business customers that runs across several countries. The end result is a documented playbook with five parts: event and scheduled triggers, risk tiers with review intervals, a maker-checker case workflow, automation thresholds, and an implementation checklist. It's written for compliance leads, risk owners, and onboarding or operations managers who already run KYB at onboarding and now need the same rigor for the rest of the customer lifecycle.

Expect two to four weeks to draft the playbook with your stakeholders, then roughly three to six months to pilot and scale it on a platform. Difficulty is moderate to high, mostly because of cross-jurisdiction policy alignment rather than technical complexity.

What you need before you start

Gather these inputs first:

  • Your current CDD/KYC policy and any local-law addenda for each country where you hold business customers.

  • A list of data sources already in use (company registries, UBO registers, screening providers, document stores).

  • Current case volumes, backlog size, and average time-to-decision for periodic reviews. You'll need these as baselines.

  • A named owner for each of the four roles in the table below.

Role

Responsibility in the re-KYC program

Compliance governance

Owns policy, risk appetite, and tier definitions; approves threshold changes

Model and risk

Calibrates risk scoring, validates automation criteria, reviews false-positive data

Operations and analysts

Work cases, request documents, apply the maker step of maker-checker

Technology and integration

Connects source systems, event feeds, and the audit store

Why multi-country re-KYC programs stall

Re-KYC, in a bank context, means two things running together: scheduled KYC/KYB updates driven by customer risk, and event-driven reassessment when something material changes. Examples of the second kind include a beneficial ownership change, a new sanctions or PEP match, adverse media, or a legal entity status change.

The FATF frames the foundation this way: "A risk-based approach means that countries, competent authorities, and banks identify, assess, and understand the money laundering and terrorist financing risk" (FATF, Risk-Based Approach for the Banking Sector). Recommendations 10, 24, and 25 are the reference points for risk-based CDD, ongoing monitoring, and periodic reassessment of beneficial ownership information. The recommendations tell you what to do. They don't tell you how to wire it up in 12 jurisdictions.

In practice, programs fail in four places:

  1. Fragmented vendor integrations, with each country using a different registry or screening provider and no shared data model.

  2. Inconsistent data recency, so one market refreshes ownership data monthly and another hasn't touched it in three years.

  3. Manual review backlogs, because every scheduled review lands in an analyst queue regardless of risk.

  4. Audit evidence gaps, where the decision exists but the reasoning, the data snapshot, or the approver doesn't.

Duna's own analysis of why reboarding has become the biggest compliance challenge makes the same point from the vendor side: KYB systems built only for onboarding break once the customer record has to stay current.

Step 1: Inventory your trigger sources and assign each an entry point

List every signal that should be able to start a review. Then decide where each one enters case management. Don't skip this step. Programs that begin with tiering and intervals tend to discover missing event feeds six months later.

Split triggers into two buckets.

Scheduled triggers fire from the risk-tier cadence (Step 3). Event-driven triggers fire from changes:

  • UBO or shareholder changes detected in a UBO register or registry filing

  • Legal entity status changes (dissolution, insolvency, name change, registered address change)

  • New adverse media on the entity or a related party

  • New PEP or sanctions matches, or discrepancies between customer-declared and registry data

  • Material shifts in transaction behavior flagged by your monitoring system

  • Customer-initiated changes, such as a new legal representative or new product

Build a simple inventory table with four columns: signal, source system, latency (real-time, daily, batch), and the case-management entry point. For screening, a daily cadence for business customers is a reasonable baseline. Duna's guide to ongoing PEP and sanctions screening for business customers covers watchlist cadence and matching logic in more detail.

Checkpoint: you should be able to point at each trigger source and say which queue or policy evaluates it. Any signal with no entry point is a gap.

Step 2: Define what the policy engine must output

A policy engine is where trigger signals become decisions about work. Define the outputs before you configure anything. For each signal, the engine should return five things:

  1. Review required, or no action.

  2. Risk-tier justification (which inputs moved the tier, and by how much).

  3. Data and document requirements for this entity, jurisdiction, and tier.

  4. Escalation level (analyst, senior compliance, independent reviewer).

  5. Automation eligibility (can the case close without a human?).

Writing policies as code matters because it makes each rule testable, versioned, and reviewable. Duna's policy engine translates compliance logic into code, supports back-testing of policy changes, and logs policy changes alongside decisions. That last point is what auditors ask about: not just what the rule is today, but what it was on the date of a given decision.

A sample rule, in illustrative pseudo-syntax (not Duna's configuration format):

policy: ubo_change_review
when:  event.type == "ubo_register_change"
       AND entity.tier IN ["medium", "high"]
then:
  review:               required
  scope:                [ubo_reverification, screening_refresh]
  due_in_days:          15
  escalation:           entity.tier == "high" ? "senior_compliance" : "analyst"
  automation_eligible:  false  IF new_ubo.jurisdiction_risk == "high"
  log:                  [inputs, rule_version, tier_before, tier_after]

Keep one global rule set with country-specific parameters (document lists, intervals, registry sources) rather than one rule set per country. That structure is what keeps calibration consistent across markets.

Step 3: Configure risk tiers and review intervals

Tier inputs for business customers usually include entity risk (type, sector, age), jurisdiction risk, ownership complexity (layers, nominee or bearer structures, cross-border holding chains), and screening outcomes. Score them with the same model everywhere. If Germany and Brazil use different weightings, the tiers aren't comparable and group-level reporting loses meaning.

The RBI's periodic updation rules give one concrete regulatory reference. As summarized by BeFiSc from paragraph 42(1): high-risk customers at least once every two years, medium-risk customers at least once every eight years, low-risk customers at least once every ten years. These are minimum frequencies for India. A multi-country bank should apply the strictest rule among the jurisdictions where the customer operates, then set internal cadences tighter than the legal floor wherever risk appetite requires it.

Here's an illustrative schema to adapt. The internal cadences are example starting values, not regulatory requirements:

Risk tier

RBI outer limit (India)

Example internal cadence

Triggers that override cadence

Verification scope

SLA target (example)

High

2 years

12 months

UBO change, sanctions/PEP hit, adverse media, status change

Full KYB refresh, UBO re-check, EDD, senior sign-off

Close within 30 days of trigger

Medium

8 years

24 to 36 months

UBO change, new screening hit, status change

Registry refresh, UBO comparison, document refresh if expired

Close within 45 days

Low

10 years

60 months

Any screening hit, status change

Registry and screening refresh; auto-accept if criteria met

Close within 60 days, mostly automated

Add guardrails for re-tiering:

  • Recalculate the tier whenever new evidence arrives, not only at the scheduled review.

  • Require a recorded reason code for every tier change, and require four-eyes approval for any downgrade.

  • Never let a downgrade extend the next review date beyond the date set by the previous, higher tier until the downgrade is approved.

Checkpoint: draft the table above with your own values. Every cell, including SLA, should have an owner.

Step 4: Set automation thresholds so analysts see only exceptions

Automation threshold means the set of conditions under which a re-KYC case closes without human review. Everything else routes to an analyst. Define the conditions as explicit, testable criteria. For a low- or medium-tier entity, a starting set might be:

  • Registry status is active and legal name and registration number match the last verified record.

  • UBO set is unchanged against the last verified snapshot, or changes are fully confirmed against an official UBO register.

  • No new screening hits above your match-score threshold, and no unresolved prior hits.

  • Submitted documents pass verification above your confidence threshold, if documents were requested.

  • No open investigations or monitoring alerts on the entity.

If any criterion fails, route the case to a human. Treat the numeric thresholds (match scores, confidence levels) as parameters you calibrate during the pilot. Don't copy another institution's values.

Duna's AI agents handle document verification, periodic reviews, and automated acceptance within these policy boundaries, and the policy engine decides which cases are eligible. For a view of where scheduled reviews are heading, see the future of periodic reviews.

Keep false-positive reduction measurable. Tag every dismissed screening hit with a cause (name collision, outdated listing, transliteration variant, related-party noise) and review the tags monthly. Tuning matching logic against real dismissal data works better than loosening thresholds across the board.

Step 5: Build the case workflow with maker-checker controls

Use a fixed set of case states so reporting is consistent across countries:

  1. New: trigger fired and policy evaluated.

  2. Data request: documents or declarations requested from the customer.

  3. Pending verification: data received, checks running.

  4. Review: analyst (maker) assesses the evidence.

  5. Escalated review: senior compliance or independent reviewer takes over.

  6. Decision: checker approves or rejects.

  7. Closed: accepted, rejected, or exited, with a reason code.

The four-eye review applies at the points where a single mistake carries the highest cost:

  • Any acceptance of a high-tier entity

  • Any tier downgrade

  • Any manual override of an automation-eligible or automation-blocked outcome

  • Any rejection or exit recommendation

  • Any adjudication of a screening hit on a high-tier entity

The maker and checker must be different people, and the system should enforce that rather than rely on procedure. Escalate when you see high-risk flags, UBO chains with more than your defined number of layers or with nominee structures, repeated hits on the same entity that analysts keep dismissing, or customers in jurisdictions on your elevated-risk list. Each escalation needs a written rationale and the evidence snapshot attached.

Keep customer communication inside the case. Duna's in-platform messaging holds requests, reminders, and replies in the same record as the decision, which removes the common audit problem of evidence sitting in email. Duna's lifecycle module runs re-KYC and re-KYB as automated workflows with policies applied across products, partners, and jurisdictions.

Checkpoint: write a one-page runbook listing roles, case states, escalation rules, and the evidence captured at each state (data snapshot, rule version, reviewer identity, timestamps, reason code).

Step 6: Connect data sources and apply local coverage

Consistent re-KYC depends on consistent data refresh. Connect these layers:

  • Source systems: core banking, CRM, and onboarding records that hold the customer master.

  • Lifecycle event feeds: customer-initiated changes and internal status changes.

  • Screening signals: PEP, sanctions, and adverse media results, ideally daily.

  • Registry and UBO data: official company registries and UBO registers per country.

  • Document capture and verification.

  • Case management and an immutable audit store.

Local coverage is the practical constraint. Duna connects to 210+ local registries and supports UBO registers, local VAT checks, and 7+ languages. That lets you run one refresh logic across countries while using each country's authoritative source. Smart routing across data providers, with fallback when a provider fails, avoids a stalled review queue when a single registry is down.

If you're comparing platforms for this, the requirements list in Duna's bank-grade KYB RFP guide is a useful starting checklist: SOC 2, UBO resolution, sanctions screening, and audit trails.

Step 7: Pilot, measure, then scale in phases

A phased rollout limits the cost of calibration errors. This reference timeline fits most multi-country banks; adjust it to your volumes and regulatory calendar.

Phase

Timing

Activities

Exit criteria

Design

Days 0 to 30

Trigger inventory, tier model, intervals, thresholds, runbook; capture baselines

Policy signed off by compliance governance

Pilot

Days 31 to 90

One country or one entity segment; run automation in shadow mode first, then live for low-tier

Automation precision and analyst feedback within agreed limits

Regional scale

Days 90 to 180

Add countries in the same regulatory region; add medium tier; tune country parameters

Backlog trend falling; audit sampling clean

Global standardization

After day 180

Unify rule set, retire local workarounds, centralize reporting

Single policy repository, per-country parameters only

Run automation in shadow mode first. The platform evaluates every case and records what it would have decided, while analysts continue to work normally. Compare the two. Any case where the engine would have auto-accepted and the analyst found a problem points to a threshold that's too loose.

Track these metrics against your Day 0 baselines:

Metric

What to measure

Analyst efficiency

Manual reviews per analyst per week; median time-to-decision

False-positive reduction

Share of screening alerts dismissed, by cause

Coverage

Jurisdictions and entity types under policy-driven re-KYC

Control quality

Share of cases with complete audit trails; share of reviews completed within SLA

For context, Duna reports 10.6x faster onboarding, a 37% conversion increase, and 4.8x analyst efficiency across its customer base. Those figures cover onboarding and review operations broadly, not re-KYC alone, so set your own re-KYC targets from your baselines rather than assuming they transfer.

After the first 60 days of live data, expect to adjust the same things in most programs: match-score thresholds that produce repeat dismissals, document lists that differ by country in ways the first policy missed, and escalation criteria that send too many medium-tier cases to senior reviewers.

Troubleshooting common failure modes

Stale registry data. If a registry updates slowly, the platform can confirm "no change" when a change has happened. Record the registry's last-updated date on every check and add a rule that blocks auto-acceptance when the source is older than your freshness limit.

Missing evidence at decision time. Cases get closed with a rationale but no data snapshot. Make the snapshot a required field in the Decision state so the case can't close without it.

Inconsistent document requirements by country. Keep one document policy with country parameters. If analysts in one market request extra documents informally, the rule set has a gap. Add the requirement to the policy and remove the workaround.

Analyst overload despite automation. Check the automation-eligible share by tier. If it's near zero for low-tier entities, a criterion is probably too strict, or a data source returns incomplete fields. Batch low-risk re-verifications by jurisdiction so analysts work similar cases together.

Audit gaps on policy changes. If you can't show which rule version applied to a past decision, versioning isn't attached to case records. Fix it before the next audit cycle.

Misconfigured tiering. Symptoms include almost everyone in the medium tier or high-tier counts that differ sharply between countries with similar customer bases. Re-run the scoring on a sample, compare weights across jurisdictions, and back-test any change before you apply it.

What you've built and where to go next

You now have a re-KYC playbook with a trigger inventory, a policy engine output specification, a tier and interval table anchored to a regulatory reference, automation criteria, a maker-checker case workflow, and a phased rollout with metrics. Together, these cover KYB re-verification, beneficial ownership reassessment, and ongoing screening as one lifecycle instead of three separate projects.

Next steps:

  • Run the trigger inventory with each country's compliance lead this month and close entry-point gaps.

  • Capture baselines for backlog, time-to-decision, and alert dismissal rates before configuring anything.

  • Choose the pilot segment, ideally a high-volume, low-to-medium-tier population in one jurisdiction.

  • Schedule a back-test of your first policy set against historical cases before going live.

  • Ask Duna for a demo of the policy engine and case management on your own tier model and two or three of your jurisdictions.

This guide walks a bank's compliance and operations team through building a re-KYC program for business customers that runs across several countries. The end result is a documented playbook with five parts: event and scheduled triggers, risk tiers with review intervals, a maker-checker case workflow, automation thresholds, and an implementation checklist. It's written for compliance leads, risk owners, and onboarding or operations managers who already run KYB at onboarding and now need the same rigor for the rest of the customer lifecycle.

Expect two to four weeks to draft the playbook with your stakeholders, then roughly three to six months to pilot and scale it on a platform. Difficulty is moderate to high, mostly because of cross-jurisdiction policy alignment rather than technical complexity.

What you need before you start

Gather these inputs first:

  • Your current CDD/KYC policy and any local-law addenda for each country where you hold business customers.

  • A list of data sources already in use (company registries, UBO registers, screening providers, document stores).

  • Current case volumes, backlog size, and average time-to-decision for periodic reviews. You'll need these as baselines.

  • A named owner for each of the four roles in the table below.

Role

Responsibility in the re-KYC program

Compliance governance

Owns policy, risk appetite, and tier definitions; approves threshold changes

Model and risk

Calibrates risk scoring, validates automation criteria, reviews false-positive data

Operations and analysts

Work cases, request documents, apply the maker step of maker-checker

Technology and integration

Connects source systems, event feeds, and the audit store

Why multi-country re-KYC programs stall

Re-KYC, in a bank context, means two things running together: scheduled KYC/KYB updates driven by customer risk, and event-driven reassessment when something material changes. Examples of the second kind include a beneficial ownership change, a new sanctions or PEP match, adverse media, or a legal entity status change.

The FATF frames the foundation this way: "A risk-based approach means that countries, competent authorities, and banks identify, assess, and understand the money laundering and terrorist financing risk" (FATF, Risk-Based Approach for the Banking Sector). Recommendations 10, 24, and 25 are the reference points for risk-based CDD, ongoing monitoring, and periodic reassessment of beneficial ownership information. The recommendations tell you what to do. They don't tell you how to wire it up in 12 jurisdictions.

In practice, programs fail in four places:

  1. Fragmented vendor integrations, with each country using a different registry or screening provider and no shared data model.

  2. Inconsistent data recency, so one market refreshes ownership data monthly and another hasn't touched it in three years.

  3. Manual review backlogs, because every scheduled review lands in an analyst queue regardless of risk.

  4. Audit evidence gaps, where the decision exists but the reasoning, the data snapshot, or the approver doesn't.

Duna's own analysis of why reboarding has become the biggest compliance challenge makes the same point from the vendor side: KYB systems built only for onboarding break once the customer record has to stay current.

Step 1: Inventory your trigger sources and assign each an entry point

List every signal that should be able to start a review. Then decide where each one enters case management. Don't skip this step. Programs that begin with tiering and intervals tend to discover missing event feeds six months later.

Split triggers into two buckets.

Scheduled triggers fire from the risk-tier cadence (Step 3). Event-driven triggers fire from changes:

  • UBO or shareholder changes detected in a UBO register or registry filing

  • Legal entity status changes (dissolution, insolvency, name change, registered address change)

  • New adverse media on the entity or a related party

  • New PEP or sanctions matches, or discrepancies between customer-declared and registry data

  • Material shifts in transaction behavior flagged by your monitoring system

  • Customer-initiated changes, such as a new legal representative or new product

Build a simple inventory table with four columns: signal, source system, latency (real-time, daily, batch), and the case-management entry point. For screening, a daily cadence for business customers is a reasonable baseline. Duna's guide to ongoing PEP and sanctions screening for business customers covers watchlist cadence and matching logic in more detail.

Checkpoint: you should be able to point at each trigger source and say which queue or policy evaluates it. Any signal with no entry point is a gap.

Step 2: Define what the policy engine must output

A policy engine is where trigger signals become decisions about work. Define the outputs before you configure anything. For each signal, the engine should return five things:

  1. Review required, or no action.

  2. Risk-tier justification (which inputs moved the tier, and by how much).

  3. Data and document requirements for this entity, jurisdiction, and tier.

  4. Escalation level (analyst, senior compliance, independent reviewer).

  5. Automation eligibility (can the case close without a human?).

Writing policies as code matters because it makes each rule testable, versioned, and reviewable. Duna's policy engine translates compliance logic into code, supports back-testing of policy changes, and logs policy changes alongside decisions. That last point is what auditors ask about: not just what the rule is today, but what it was on the date of a given decision.

A sample rule, in illustrative pseudo-syntax (not Duna's configuration format):

policy: ubo_change_review
when:  event.type == "ubo_register_change"
       AND entity.tier IN ["medium", "high"]
then:
  review:               required
  scope:                [ubo_reverification, screening_refresh]
  due_in_days:          15
  escalation:           entity.tier == "high" ? "senior_compliance" : "analyst"
  automation_eligible:  false  IF new_ubo.jurisdiction_risk == "high"
  log:                  [inputs, rule_version, tier_before, tier_after]

Keep one global rule set with country-specific parameters (document lists, intervals, registry sources) rather than one rule set per country. That structure is what keeps calibration consistent across markets.

Step 3: Configure risk tiers and review intervals

Tier inputs for business customers usually include entity risk (type, sector, age), jurisdiction risk, ownership complexity (layers, nominee or bearer structures, cross-border holding chains), and screening outcomes. Score them with the same model everywhere. If Germany and Brazil use different weightings, the tiers aren't comparable and group-level reporting loses meaning.

The RBI's periodic updation rules give one concrete regulatory reference. As summarized by BeFiSc from paragraph 42(1): high-risk customers at least once every two years, medium-risk customers at least once every eight years, low-risk customers at least once every ten years. These are minimum frequencies for India. A multi-country bank should apply the strictest rule among the jurisdictions where the customer operates, then set internal cadences tighter than the legal floor wherever risk appetite requires it.

Here's an illustrative schema to adapt. The internal cadences are example starting values, not regulatory requirements:

Risk tier

RBI outer limit (India)

Example internal cadence

Triggers that override cadence

Verification scope

SLA target (example)

High

2 years

12 months

UBO change, sanctions/PEP hit, adverse media, status change

Full KYB refresh, UBO re-check, EDD, senior sign-off

Close within 30 days of trigger

Medium

8 years

24 to 36 months

UBO change, new screening hit, status change

Registry refresh, UBO comparison, document refresh if expired

Close within 45 days

Low

10 years

60 months

Any screening hit, status change

Registry and screening refresh; auto-accept if criteria met

Close within 60 days, mostly automated

Add guardrails for re-tiering:

  • Recalculate the tier whenever new evidence arrives, not only at the scheduled review.

  • Require a recorded reason code for every tier change, and require four-eyes approval for any downgrade.

  • Never let a downgrade extend the next review date beyond the date set by the previous, higher tier until the downgrade is approved.

Checkpoint: draft the table above with your own values. Every cell, including SLA, should have an owner.

Step 4: Set automation thresholds so analysts see only exceptions

Automation threshold means the set of conditions under which a re-KYC case closes without human review. Everything else routes to an analyst. Define the conditions as explicit, testable criteria. For a low- or medium-tier entity, a starting set might be:

  • Registry status is active and legal name and registration number match the last verified record.

  • UBO set is unchanged against the last verified snapshot, or changes are fully confirmed against an official UBO register.

  • No new screening hits above your match-score threshold, and no unresolved prior hits.

  • Submitted documents pass verification above your confidence threshold, if documents were requested.

  • No open investigations or monitoring alerts on the entity.

If any criterion fails, route the case to a human. Treat the numeric thresholds (match scores, confidence levels) as parameters you calibrate during the pilot. Don't copy another institution's values.

Duna's AI agents handle document verification, periodic reviews, and automated acceptance within these policy boundaries, and the policy engine decides which cases are eligible. For a view of where scheduled reviews are heading, see the future of periodic reviews.

Keep false-positive reduction measurable. Tag every dismissed screening hit with a cause (name collision, outdated listing, transliteration variant, related-party noise) and review the tags monthly. Tuning matching logic against real dismissal data works better than loosening thresholds across the board.

Step 5: Build the case workflow with maker-checker controls

Use a fixed set of case states so reporting is consistent across countries:

  1. New: trigger fired and policy evaluated.

  2. Data request: documents or declarations requested from the customer.

  3. Pending verification: data received, checks running.

  4. Review: analyst (maker) assesses the evidence.

  5. Escalated review: senior compliance or independent reviewer takes over.

  6. Decision: checker approves or rejects.

  7. Closed: accepted, rejected, or exited, with a reason code.

The four-eye review applies at the points where a single mistake carries the highest cost:

  • Any acceptance of a high-tier entity

  • Any tier downgrade

  • Any manual override of an automation-eligible or automation-blocked outcome

  • Any rejection or exit recommendation

  • Any adjudication of a screening hit on a high-tier entity

The maker and checker must be different people, and the system should enforce that rather than rely on procedure. Escalate when you see high-risk flags, UBO chains with more than your defined number of layers or with nominee structures, repeated hits on the same entity that analysts keep dismissing, or customers in jurisdictions on your elevated-risk list. Each escalation needs a written rationale and the evidence snapshot attached.

Keep customer communication inside the case. Duna's in-platform messaging holds requests, reminders, and replies in the same record as the decision, which removes the common audit problem of evidence sitting in email. Duna's lifecycle module runs re-KYC and re-KYB as automated workflows with policies applied across products, partners, and jurisdictions.

Checkpoint: write a one-page runbook listing roles, case states, escalation rules, and the evidence captured at each state (data snapshot, rule version, reviewer identity, timestamps, reason code).

Step 6: Connect data sources and apply local coverage

Consistent re-KYC depends on consistent data refresh. Connect these layers:

  • Source systems: core banking, CRM, and onboarding records that hold the customer master.

  • Lifecycle event feeds: customer-initiated changes and internal status changes.

  • Screening signals: PEP, sanctions, and adverse media results, ideally daily.

  • Registry and UBO data: official company registries and UBO registers per country.

  • Document capture and verification.

  • Case management and an immutable audit store.

Local coverage is the practical constraint. Duna connects to 210+ local registries and supports UBO registers, local VAT checks, and 7+ languages. That lets you run one refresh logic across countries while using each country's authoritative source. Smart routing across data providers, with fallback when a provider fails, avoids a stalled review queue when a single registry is down.

If you're comparing platforms for this, the requirements list in Duna's bank-grade KYB RFP guide is a useful starting checklist: SOC 2, UBO resolution, sanctions screening, and audit trails.

Step 7: Pilot, measure, then scale in phases

A phased rollout limits the cost of calibration errors. This reference timeline fits most multi-country banks; adjust it to your volumes and regulatory calendar.

Phase

Timing

Activities

Exit criteria

Design

Days 0 to 30

Trigger inventory, tier model, intervals, thresholds, runbook; capture baselines

Policy signed off by compliance governance

Pilot

Days 31 to 90

One country or one entity segment; run automation in shadow mode first, then live for low-tier

Automation precision and analyst feedback within agreed limits

Regional scale

Days 90 to 180

Add countries in the same regulatory region; add medium tier; tune country parameters

Backlog trend falling; audit sampling clean

Global standardization

After day 180

Unify rule set, retire local workarounds, centralize reporting

Single policy repository, per-country parameters only

Run automation in shadow mode first. The platform evaluates every case and records what it would have decided, while analysts continue to work normally. Compare the two. Any case where the engine would have auto-accepted and the analyst found a problem points to a threshold that's too loose.

Track these metrics against your Day 0 baselines:

Metric

What to measure

Analyst efficiency

Manual reviews per analyst per week; median time-to-decision

False-positive reduction

Share of screening alerts dismissed, by cause

Coverage

Jurisdictions and entity types under policy-driven re-KYC

Control quality

Share of cases with complete audit trails; share of reviews completed within SLA

For context, Duna reports 10.6x faster onboarding, a 37% conversion increase, and 4.8x analyst efficiency across its customer base. Those figures cover onboarding and review operations broadly, not re-KYC alone, so set your own re-KYC targets from your baselines rather than assuming they transfer.

After the first 60 days of live data, expect to adjust the same things in most programs: match-score thresholds that produce repeat dismissals, document lists that differ by country in ways the first policy missed, and escalation criteria that send too many medium-tier cases to senior reviewers.

Troubleshooting common failure modes

Stale registry data. If a registry updates slowly, the platform can confirm "no change" when a change has happened. Record the registry's last-updated date on every check and add a rule that blocks auto-acceptance when the source is older than your freshness limit.

Missing evidence at decision time. Cases get closed with a rationale but no data snapshot. Make the snapshot a required field in the Decision state so the case can't close without it.

Inconsistent document requirements by country. Keep one document policy with country parameters. If analysts in one market request extra documents informally, the rule set has a gap. Add the requirement to the policy and remove the workaround.

Analyst overload despite automation. Check the automation-eligible share by tier. If it's near zero for low-tier entities, a criterion is probably too strict, or a data source returns incomplete fields. Batch low-risk re-verifications by jurisdiction so analysts work similar cases together.

Audit gaps on policy changes. If you can't show which rule version applied to a past decision, versioning isn't attached to case records. Fix it before the next audit cycle.

Misconfigured tiering. Symptoms include almost everyone in the medium tier or high-tier counts that differ sharply between countries with similar customer bases. Re-run the scoring on a sample, compare weights across jurisdictions, and back-test any change before you apply it.

What you've built and where to go next

You now have a re-KYC playbook with a trigger inventory, a policy engine output specification, a tier and interval table anchored to a regulatory reference, automation criteria, a maker-checker case workflow, and a phased rollout with metrics. Together, these cover KYB re-verification, beneficial ownership reassessment, and ongoing screening as one lifecycle instead of three separate projects.

Next steps:

  • Run the trigger inventory with each country's compliance lead this month and close entry-point gaps.

  • Capture baselines for backlog, time-to-decision, and alert dismissal rates before configuring anything.

  • Choose the pilot segment, ideally a high-volume, low-to-medium-tier population in one jurisdiction.

  • Schedule a back-test of your first policy set against historical cases before going live.

  • Ask Duna for a demo of the policy engine and case management on your own tier model and two or three of your jurisdictions.

This guide walks a bank's compliance and operations team through building a re-KYC program for business customers that runs across several countries. The end result is a documented playbook with five parts: event and scheduled triggers, risk tiers with review intervals, a maker-checker case workflow, automation thresholds, and an implementation checklist. It's written for compliance leads, risk owners, and onboarding or operations managers who already run KYB at onboarding and now need the same rigor for the rest of the customer lifecycle.

Expect two to four weeks to draft the playbook with your stakeholders, then roughly three to six months to pilot and scale it on a platform. Difficulty is moderate to high, mostly because of cross-jurisdiction policy alignment rather than technical complexity.

What you need before you start

Gather these inputs first:

  • Your current CDD/KYC policy and any local-law addenda for each country where you hold business customers.

  • A list of data sources already in use (company registries, UBO registers, screening providers, document stores).

  • Current case volumes, backlog size, and average time-to-decision for periodic reviews. You'll need these as baselines.

  • A named owner for each of the four roles in the table below.

Role

Responsibility in the re-KYC program

Compliance governance

Owns policy, risk appetite, and tier definitions; approves threshold changes

Model and risk

Calibrates risk scoring, validates automation criteria, reviews false-positive data

Operations and analysts

Work cases, request documents, apply the maker step of maker-checker

Technology and integration

Connects source systems, event feeds, and the audit store

Why multi-country re-KYC programs stall

Re-KYC, in a bank context, means two things running together: scheduled KYC/KYB updates driven by customer risk, and event-driven reassessment when something material changes. Examples of the second kind include a beneficial ownership change, a new sanctions or PEP match, adverse media, or a legal entity status change.

The FATF frames the foundation this way: "A risk-based approach means that countries, competent authorities, and banks identify, assess, and understand the money laundering and terrorist financing risk" (FATF, Risk-Based Approach for the Banking Sector). Recommendations 10, 24, and 25 are the reference points for risk-based CDD, ongoing monitoring, and periodic reassessment of beneficial ownership information. The recommendations tell you what to do. They don't tell you how to wire it up in 12 jurisdictions.

In practice, programs fail in four places:

  1. Fragmented vendor integrations, with each country using a different registry or screening provider and no shared data model.

  2. Inconsistent data recency, so one market refreshes ownership data monthly and another hasn't touched it in three years.

  3. Manual review backlogs, because every scheduled review lands in an analyst queue regardless of risk.

  4. Audit evidence gaps, where the decision exists but the reasoning, the data snapshot, or the approver doesn't.

Duna's own analysis of why reboarding has become the biggest compliance challenge makes the same point from the vendor side: KYB systems built only for onboarding break once the customer record has to stay current.

Step 1: Inventory your trigger sources and assign each an entry point

List every signal that should be able to start a review. Then decide where each one enters case management. Don't skip this step. Programs that begin with tiering and intervals tend to discover missing event feeds six months later.

Split triggers into two buckets.

Scheduled triggers fire from the risk-tier cadence (Step 3). Event-driven triggers fire from changes:

  • UBO or shareholder changes detected in a UBO register or registry filing

  • Legal entity status changes (dissolution, insolvency, name change, registered address change)

  • New adverse media on the entity or a related party

  • New PEP or sanctions matches, or discrepancies between customer-declared and registry data

  • Material shifts in transaction behavior flagged by your monitoring system

  • Customer-initiated changes, such as a new legal representative or new product

Build a simple inventory table with four columns: signal, source system, latency (real-time, daily, batch), and the case-management entry point. For screening, a daily cadence for business customers is a reasonable baseline. Duna's guide to ongoing PEP and sanctions screening for business customers covers watchlist cadence and matching logic in more detail.

Checkpoint: you should be able to point at each trigger source and say which queue or policy evaluates it. Any signal with no entry point is a gap.

Step 2: Define what the policy engine must output

A policy engine is where trigger signals become decisions about work. Define the outputs before you configure anything. For each signal, the engine should return five things:

  1. Review required, or no action.

  2. Risk-tier justification (which inputs moved the tier, and by how much).

  3. Data and document requirements for this entity, jurisdiction, and tier.

  4. Escalation level (analyst, senior compliance, independent reviewer).

  5. Automation eligibility (can the case close without a human?).

Writing policies as code matters because it makes each rule testable, versioned, and reviewable. Duna's policy engine translates compliance logic into code, supports back-testing of policy changes, and logs policy changes alongside decisions. That last point is what auditors ask about: not just what the rule is today, but what it was on the date of a given decision.

A sample rule, in illustrative pseudo-syntax (not Duna's configuration format):

policy: ubo_change_review
when:  event.type == "ubo_register_change"
       AND entity.tier IN ["medium", "high"]
then:
  review:               required
  scope:                [ubo_reverification, screening_refresh]
  due_in_days:          15
  escalation:           entity.tier == "high" ? "senior_compliance" : "analyst"
  automation_eligible:  false  IF new_ubo.jurisdiction_risk == "high"
  log:                  [inputs, rule_version, tier_before, tier_after]

Keep one global rule set with country-specific parameters (document lists, intervals, registry sources) rather than one rule set per country. That structure is what keeps calibration consistent across markets.

Step 3: Configure risk tiers and review intervals

Tier inputs for business customers usually include entity risk (type, sector, age), jurisdiction risk, ownership complexity (layers, nominee or bearer structures, cross-border holding chains), and screening outcomes. Score them with the same model everywhere. If Germany and Brazil use different weightings, the tiers aren't comparable and group-level reporting loses meaning.

The RBI's periodic updation rules give one concrete regulatory reference. As summarized by BeFiSc from paragraph 42(1): high-risk customers at least once every two years, medium-risk customers at least once every eight years, low-risk customers at least once every ten years. These are minimum frequencies for India. A multi-country bank should apply the strictest rule among the jurisdictions where the customer operates, then set internal cadences tighter than the legal floor wherever risk appetite requires it.

Here's an illustrative schema to adapt. The internal cadences are example starting values, not regulatory requirements:

Risk tier

RBI outer limit (India)

Example internal cadence

Triggers that override cadence

Verification scope

SLA target (example)

High

2 years

12 months

UBO change, sanctions/PEP hit, adverse media, status change

Full KYB refresh, UBO re-check, EDD, senior sign-off

Close within 30 days of trigger

Medium

8 years

24 to 36 months

UBO change, new screening hit, status change

Registry refresh, UBO comparison, document refresh if expired

Close within 45 days

Low

10 years

60 months

Any screening hit, status change

Registry and screening refresh; auto-accept if criteria met

Close within 60 days, mostly automated

Add guardrails for re-tiering:

  • Recalculate the tier whenever new evidence arrives, not only at the scheduled review.

  • Require a recorded reason code for every tier change, and require four-eyes approval for any downgrade.

  • Never let a downgrade extend the next review date beyond the date set by the previous, higher tier until the downgrade is approved.

Checkpoint: draft the table above with your own values. Every cell, including SLA, should have an owner.

Step 4: Set automation thresholds so analysts see only exceptions

Automation threshold means the set of conditions under which a re-KYC case closes without human review. Everything else routes to an analyst. Define the conditions as explicit, testable criteria. For a low- or medium-tier entity, a starting set might be:

  • Registry status is active and legal name and registration number match the last verified record.

  • UBO set is unchanged against the last verified snapshot, or changes are fully confirmed against an official UBO register.

  • No new screening hits above your match-score threshold, and no unresolved prior hits.

  • Submitted documents pass verification above your confidence threshold, if documents were requested.

  • No open investigations or monitoring alerts on the entity.

If any criterion fails, route the case to a human. Treat the numeric thresholds (match scores, confidence levels) as parameters you calibrate during the pilot. Don't copy another institution's values.

Duna's AI agents handle document verification, periodic reviews, and automated acceptance within these policy boundaries, and the policy engine decides which cases are eligible. For a view of where scheduled reviews are heading, see the future of periodic reviews.

Keep false-positive reduction measurable. Tag every dismissed screening hit with a cause (name collision, outdated listing, transliteration variant, related-party noise) and review the tags monthly. Tuning matching logic against real dismissal data works better than loosening thresholds across the board.

Step 5: Build the case workflow with maker-checker controls

Use a fixed set of case states so reporting is consistent across countries:

  1. New: trigger fired and policy evaluated.

  2. Data request: documents or declarations requested from the customer.

  3. Pending verification: data received, checks running.

  4. Review: analyst (maker) assesses the evidence.

  5. Escalated review: senior compliance or independent reviewer takes over.

  6. Decision: checker approves or rejects.

  7. Closed: accepted, rejected, or exited, with a reason code.

The four-eye review applies at the points where a single mistake carries the highest cost:

  • Any acceptance of a high-tier entity

  • Any tier downgrade

  • Any manual override of an automation-eligible or automation-blocked outcome

  • Any rejection or exit recommendation

  • Any adjudication of a screening hit on a high-tier entity

The maker and checker must be different people, and the system should enforce that rather than rely on procedure. Escalate when you see high-risk flags, UBO chains with more than your defined number of layers or with nominee structures, repeated hits on the same entity that analysts keep dismissing, or customers in jurisdictions on your elevated-risk list. Each escalation needs a written rationale and the evidence snapshot attached.

Keep customer communication inside the case. Duna's in-platform messaging holds requests, reminders, and replies in the same record as the decision, which removes the common audit problem of evidence sitting in email. Duna's lifecycle module runs re-KYC and re-KYB as automated workflows with policies applied across products, partners, and jurisdictions.

Checkpoint: write a one-page runbook listing roles, case states, escalation rules, and the evidence captured at each state (data snapshot, rule version, reviewer identity, timestamps, reason code).

Step 6: Connect data sources and apply local coverage

Consistent re-KYC depends on consistent data refresh. Connect these layers:

  • Source systems: core banking, CRM, and onboarding records that hold the customer master.

  • Lifecycle event feeds: customer-initiated changes and internal status changes.

  • Screening signals: PEP, sanctions, and adverse media results, ideally daily.

  • Registry and UBO data: official company registries and UBO registers per country.

  • Document capture and verification.

  • Case management and an immutable audit store.

Local coverage is the practical constraint. Duna connects to 210+ local registries and supports UBO registers, local VAT checks, and 7+ languages. That lets you run one refresh logic across countries while using each country's authoritative source. Smart routing across data providers, with fallback when a provider fails, avoids a stalled review queue when a single registry is down.

If you're comparing platforms for this, the requirements list in Duna's bank-grade KYB RFP guide is a useful starting checklist: SOC 2, UBO resolution, sanctions screening, and audit trails.

Step 7: Pilot, measure, then scale in phases

A phased rollout limits the cost of calibration errors. This reference timeline fits most multi-country banks; adjust it to your volumes and regulatory calendar.

Phase

Timing

Activities

Exit criteria

Design

Days 0 to 30

Trigger inventory, tier model, intervals, thresholds, runbook; capture baselines

Policy signed off by compliance governance

Pilot

Days 31 to 90

One country or one entity segment; run automation in shadow mode first, then live for low-tier

Automation precision and analyst feedback within agreed limits

Regional scale

Days 90 to 180

Add countries in the same regulatory region; add medium tier; tune country parameters

Backlog trend falling; audit sampling clean

Global standardization

After day 180

Unify rule set, retire local workarounds, centralize reporting

Single policy repository, per-country parameters only

Run automation in shadow mode first. The platform evaluates every case and records what it would have decided, while analysts continue to work normally. Compare the two. Any case where the engine would have auto-accepted and the analyst found a problem points to a threshold that's too loose.

Track these metrics against your Day 0 baselines:

Metric

What to measure

Analyst efficiency

Manual reviews per analyst per week; median time-to-decision

False-positive reduction

Share of screening alerts dismissed, by cause

Coverage

Jurisdictions and entity types under policy-driven re-KYC

Control quality

Share of cases with complete audit trails; share of reviews completed within SLA

For context, Duna reports 10.6x faster onboarding, a 37% conversion increase, and 4.8x analyst efficiency across its customer base. Those figures cover onboarding and review operations broadly, not re-KYC alone, so set your own re-KYC targets from your baselines rather than assuming they transfer.

After the first 60 days of live data, expect to adjust the same things in most programs: match-score thresholds that produce repeat dismissals, document lists that differ by country in ways the first policy missed, and escalation criteria that send too many medium-tier cases to senior reviewers.

Troubleshooting common failure modes

Stale registry data. If a registry updates slowly, the platform can confirm "no change" when a change has happened. Record the registry's last-updated date on every check and add a rule that blocks auto-acceptance when the source is older than your freshness limit.

Missing evidence at decision time. Cases get closed with a rationale but no data snapshot. Make the snapshot a required field in the Decision state so the case can't close without it.

Inconsistent document requirements by country. Keep one document policy with country parameters. If analysts in one market request extra documents informally, the rule set has a gap. Add the requirement to the policy and remove the workaround.

Analyst overload despite automation. Check the automation-eligible share by tier. If it's near zero for low-tier entities, a criterion is probably too strict, or a data source returns incomplete fields. Batch low-risk re-verifications by jurisdiction so analysts work similar cases together.

Audit gaps on policy changes. If you can't show which rule version applied to a past decision, versioning isn't attached to case records. Fix it before the next audit cycle.

Misconfigured tiering. Symptoms include almost everyone in the medium tier or high-tier counts that differ sharply between countries with similar customer bases. Re-run the scoring on a sample, compare weights across jurisdictions, and back-test any change before you apply it.

What you've built and where to go next

You now have a re-KYC playbook with a trigger inventory, a policy engine output specification, a tier and interval table anchored to a regulatory reference, automation criteria, a maker-checker case workflow, and a phased rollout with metrics. Together, these cover KYB re-verification, beneficial ownership reassessment, and ongoing screening as one lifecycle instead of three separate projects.

Next steps:

  • Run the trigger inventory with each country's compliance lead this month and close entry-point gaps.

  • Capture baselines for backlog, time-to-decision, and alert dismissal rates before configuring anything.

  • Choose the pilot segment, ideally a high-volume, low-to-medium-tier population in one jurisdiction.

  • Schedule a back-test of your first policy set against historical cases before going live.

  • Ask Duna for a demo of the policy engine and case management on your own tier model and two or three of your jurisdictions.