Feature Operational Risk
The Second Line's Key Person Problem: Mapping Who Actually Knows the Control
Key person dependency risk in second-line compliance controls: a control leverage map, cold-execution drills, and the evidence examiners ask for.
Table of Contents
TL;DR
- For most second-line controls, the real single point of failure isn’t a role — it’s the artifact. One person knows which query pulls the population, which sample size survives challenge, and where the sign-off lives.
- The OCC wrote the control-level version of this into Citibank’s 2020 consent order: Article X requires a staffing assessment that names gaps in skills, dual roles, and matrix reporting, with documented annual board review.
- Map controls by leverage, not by org chart. Then drill the backups cold, in the system, with the primary locked out.
- Anything that can’t run without one person gets a documented risk acceptance with a named owner and an expiry date — not a line in a succession deck.
There’s a specific shape to this failure and it is almost never a resignation. It’s a Tuesday. The transaction monitoring tuning memo is due, the analyst who writes it is on a plane to a vendor conference, and the person who opens the shared drive finds a folder called “TM tuning” containing one 2024 file and a readme that says “see Priya.” The control ran. The evidence didn’t. The examiner sampling Q3 finds fourteen controls with a named owner and two with a working one.
Nobody in that scenario was negligent. The control had an owner. The backup was assigned in last year’s RCSA refresh. It’s just that the backup’s name was in a cell and the knowledge was in Priya’s head, and no process in the building compared those two facts.
The OCC has a version of this insight that it wrote into an enforcement order. Citibank’s October 7, 2020 consent order, AA-EC-2020-64, doesn’t stop at “hire more people.” Article X, Staffing and Technology Resource Assessment, requires the bank to assess the aggregate skills and expertise needed across front-line risk management functions, independent risk management, and internal audit, “and identify any gaps in those aggregate skills and/or expertise within the Bank’s current staff, including the appropriateness of where the staff has more than one role across functions or business lines (‘dual roles’) or where staff has supervisors in more than one function or business (‘matrix reporting’)” (AA-EC-2020-64, Art. X §1(a)). Paragraph (1)(c) requires annual board review of every dual role and matrix reporting arrangement with documented approval or disapproval. That requirement is still live — the OCC terminated only the July 10, 2024 amendment in a December 11, 2025 termination order.
Dual roles and matrix reporting are the supervisory vocabulary for key person dependency. Not a staffing number. A structural condition.
Applicability, before the framework
12 CFR Part 30, Appendix D covers banks with average total consolidated assets of $50 billion or more, plus banks whose parent controls a covered bank, and the OCC can apply it to smaller institutions it deems highly complex (App. D, §I.A). Appendix D, §II.L goes further than most people expect — it requires the board or a board committee to “review and approve a written talent management program that provides for development, recruitment, and succession planning” regarding the CEO, the chief audit executive, the chief risk executive, their direct reports, and other potential successors.
Below $50 billion, the hooks are different but real. The Comptroller’s Handbook on corporate and risk governance says the succession policy “should identify critical positions that would fall in the scope of a succession plan,” and notes plainly that at smaller banks “the depth of talent available for key management positions may be limited” (Corporate and Risk Governance, V2.0, p. 20). The FFIEC’s BCM booklet puts “defining BCM roles, responsibilities, and succession plans” on management (FFIEC BCM, §II.A).
And if you buy compliance tooling from a vendor, the interagency third-party guidance expects you to evaluate that vendor’s own exposure: §III.C.1.e asks whether the third party has “succession and redundancy planning for key personnel,” and the contract provisions section expects notification of “key personnel changes” (SR 23-4, p. 7 and p. 13). Your vendor’s Priya problem is your outage.
Note what every one of those citations has in common: they’re all role-level or relationship-level. None of them tell you how to find out which of your 340 controls actually depends on one human. That’s the gap this work fills, and it’s why a headcount ask doesn’t satisfy it.
Build a control leverage map, not a succession deck
The unit of analysis is the control, scored against the artifact. Four tiers, applied to controls already in the RCSA — don’t build a parallel universe of controls for this exercise, or you’ll finish with two inventories nobody reconciles.
| Tier | Condition | Typical control | Required treatment |
|---|---|---|---|
| 1 — Single point | One person can execute and evidence it; query logic or judgment undocumented | Manual monitoring tuning, bespoke reconciliation, model overrides, judgmental EDD escalation | Written procedure with the actual query, second person trained to execution, drill within 60 days, owner risk acceptance until T2+ |
| 2 — Thin | Procedure exists but dated or partial; backup has read it, never run it | Sampling and testing routines, most manual QA | Procedure rewrite to current state, one cold execution by backup, drill evidence filed |
| 3 — Covered | Backup has executed unaided in the last two quarters, procedure current | Established review and challenge routines | Annual drill, procedure review on system change |
| 4 — Durable | Runs correctly without a specific person; tooling enforces the steps | Automated reconciliations, system-enforced approvals, report-driven evidence | Change control only |
Score with two people in the room — the control owner and someone from a different team who has actually pulled the evidence. Owners self-assess optimistically; they score their own substitution. A control the owner executes alone and scores T3 is the most common error, and the independent second opinion is what surfaces it.
A realistic mapping output for a 340-control RCSA in a mid-size institution tends to land somewhere near 25–40 controls at T1, most of them concentrated in a handful of functions. Calibrate against your own inventory; the concentration will cluster around whatever team is thinnest, and that clustering is itself the finding.
Where the map gets uncomfortable
Two places. First, judgment controls score T1 far more than people expect, because the “procedure” is a paragraph that says “review for reasonableness.” That’s a description, not a procedure. A useful T1 rewrite names the fields examined, the comparison source, the threshold, and what happens at the threshold.
Second, the map exposes the owner who is also the backup for four other controls. That shows up as one name in nine Tier 1 cells and it looks fine on paper, which is precisely the point — it’s the same illusion as dual roles in a staffing chart. Cap any single individual at three T1 controls; past that, you’ve documented a queue rather than coverage.
The cold-execution drill
Warm walkthroughs are theater. The primary is present, the backup watches, everything works. Drill it cold or don’t count it.
Protocol, per control:
- Primary stays out of the session. No chat, no screen share, no “it’s under the other tab.”
- Backup gets the control ID, the policy reference, and nothing else.
- Backup produces the evidence artifact from scratch — population, sample, workpaper, sign-off — with the clock running.
- Record: minutes to first correct step, missing access, procedure gaps hit, defect count, whether the artifact would survive a challenge unchanged.
- Primary reviews within 48 hours, then the procedure gets amended for every gap the drill hit.
Escalation defaults, to be recalibrated after two quarters of drill history: a T1 control whose backup can’t produce a clean artifact in two attempts moves to the issue log with an owner and a target date. Any drill where the backup’s artifact has three or more defects implies a procedure rewrite, not coaching. A drill where the backup can’t get system access at all retires the “backup” designation, because access is the cheap half of coverage.
Anti-gaming pairs, because drills fail in two predictable ways. The first is the backup who escalates to the primary in a panic — count it, and log the moment it happened, since that timestamp is your real coverage number. The second is a drill pack rewritten to be drill-friendly: sampling sizes inflated, steps reordered to match the drill sheet rather than the work. Catch it by having the tester run one drill from the policy text alone, not the drill pack.
The 30/60/90, with the decisions in it
Days 1–30. Scope the map to second-line controls in the highest-change areas — monitoring, testing, regulatory reporting, issue validation, vendor oversight. Score every control. Bring Tier 1 findings to the risk committee with named owners and a proposed treatment per control: proceduralize, cross-train, or accept.
Days 31–60. Rewrite the T1 procedures, with the actual query text, report path, fields examined, and threshold logic. Provision backup access. Run the first cold drills on the ten controls with the highest regulatory salience. Update the RCSA entries — the T1 score belongs in the control’s inherent risk assessment, not in a separate spreadsheet, or it never gets re-tested.
Days 61–90. Board or committee memo summarizing concentration, especially any individual holding more than three T1 controls, any function above 15% of its controls at T1, and every accepted risk with the owner and expiry. Then set the annual cadence and put drills on the internal audit universe so they don’t lapse when attention moves.
The 30/60/90 shape isn’t new. What has to be non-negotiable is the decision content: every control lands on proceduralize, cross-train, or accept, and every accept has a signer and an expiry. An accepted T1 control with a documented owner decision reads as managed. A T1 control with no decision reads as unmanaged, which is how it will be written up.
Evidence the file should hold
The committee memo with owners and treatments. The leverage map with scores and the independent second opinion. Cold drill records with timestamps and defects. Amended procedures, versioned, with the change rationale. Access provisioning confirmations for backups. Board or committee minutes for the annual review of dual roles and matrix reporting, which is the artifact App. D §II.L(2) and the Citi order Article X(1)(c) are actually after.
If continuity of critical activity is the question rather than control execution, the FFIEC personnel continuity and succession coverage piece handles the six tests for alternates on critical activities. This is the narrower, duller cousin: whether a control can run and be evidenced on an ordinary non-event day. The mechanism for the map itself sits in the RCSA scoping and coverage matrix walkthrough, and when a T1 control fails its drill and needs tracking with an owner and a date, the issue management framework is where it goes — not the drill log.
So What?
Pull your last quarter’s sample of tested controls and pick the ten with the highest regulatory salience. For each, answer one question: name the person who could produce that evidence next Tuesday if the owner were unreachable. Not the role. The person. If you can’t, you’ve found your Tier 1s without building anything, and you can start this week on a sample of ten instead of waiting for a 340-control mapping project to be funded.
Then test one of them cold, this month, with the primary locked out. Whatever breaks is cheaper now than in a targeted exam.
The RCSA (Risk & Control Self-Assessment) template includes the control attributes field that carries a leverage score alongside inherent risk, so coverage re-tests on the normal refresh cycle instead of living in a deck.
FAQ
What is key person dependency risk in a second-line compliance function?
It’s the condition where execution of a control and production of its evidence depend on one person’s undocumented knowledge. The control is nominally owned, the procedure nominally exists, and the evidence trail reconstructs cleanly — right up until that person is on leave, resigns, or gets pulled onto a remediation project. Then the control either stops running or runs without a record, and both outcomes are findings.
Does the OCC actually require staffing assessments, or is that exam narrative?
In Citibank’s October 7, 2020 consent order, AA-EC-2020-64, Article X is titled Staffing and Technology Resource Assessment and requires a board-submitted assessment of the skills needed across front-line risk management, independent risk management, and internal audit. Paragraph (1)(a) specifically requires identifying gaps in aggregate skills and expertise, including where staff hold more than one role across functions, or where staff have supervisors in more than one function. Board review of every dual role and matrix reporting is annual and documented. The 2020 order remains in force; OCC terminated only the July 10, 2024 amendment on December 11, 2025.
What is a control leverage map and how is it different from a succession plan?
A succession plan names a replacement for a role. A control leverage map scores each control on how much of its execution lives in one head, using tiers from single-point-of-failure through fully proceduralized. It renders a role-level plan into a control-level one, which is the unit examiners sample against.
What is a cold-execution drill?
The named backup runs the control start to finish with no help from the primary and produces the evidence artifact from scratch. Time to completion, missing access, procedure gaps, and defects are recorded. It’s the only test that distinguishes a name in a cell from coverage, because the primary being available quietly masks everything.
How do we keep coverage mapping from becoming a headcount ask?
Tie it to control inventory rather than to roles. Score only controls already in the RCSA, and let the tiers carry the argument: a control that can’t run without one person is a documented risk accepted by a named owner, not a staffing request. The decision memo is a risk acceptance, and the budget conversation follows from it rather than leading it.
◆ Need the working template?
Start with the source guide.
These answer-first guides summarize the required fields, evidence, and implementation steps behind the templates practitioners search for.
◆ Related template
RCSA (Risk & Control Self-Assessment)
141 fintech risks with mapped controls, a 97-question self-assessment, control testing plan, challenge log and a one-page Board Summary.
◆ Immaterial Findings · Weekly
Sharp risk & compliance insights. No fluff.
◆ FAQ
Frequently asked questions.
What is key person dependency risk in a second-line compliance function?
Does the OCC actually require staffing assessments, or is that just exam narrative?
What is a control leverage map and how is it different from a succession plan?
What is a cold-execution drill?
How do we keep coverage mapping from becoming a headcount ask?
Author
Rebecca Leung
Rebecca Leung has 8+ years of risk and compliance experience across first and second line roles at commercial banks, asset managers, and fintechs. Former management consultant advising financial institutions on risk strategy. Founder of RiskTemplates.
◆ Related framework
RCSA (Risk & Control Self-Assessment)
141 fintech risks with mapped controls, a 97-question self-assessment, control testing plan, challenge log and a one-page Board Summary.
◆ Keep reading
Related posts.
Operational Risk
The OCC Just Hit American Express With a $350 Million AML Fine. $13 Billion in Suspected Trade-Based Money Laundering. Here's What Every Bank's BSA Officer Needs to See.
On October 8, 2026, the OCC assessed a $350 million civil money penalty against American Express National Bank for BSA/AML program failures tied to $13 billion in suspected trade-based money laundering activity spanning 2014 to 2025. The Federal Reserve issued a parallel consent cease-and-desist order against the holding company. Here's what the enforcement action reveals about AML program adequacy.
Oct 9, 2026
Operational Risk
The OCC and FDIC Just Redefined What Gets You an MRA. Here's What Your Bank Needs to Do Before November 2.
The OCC and FDIC's final rule redefining 'unsafe or unsound practice' and raising the bar for MRA issuance takes effect November 2, 2026. Here's what changes — and what you should be doing now.
Oct 6, 2026
Operational Risk
40 States Just Made Credit Acceptance Corporation Pay $700 Million for Loans It Knew Borrowers Couldn't Repay. Here's What Consumer Lenders Need to Lock In Now.
On September 17, 2026, a 40-state coalition secured a $700 million settlement against Credit Acceptance Corporation for predatory subprime auto loans—including $634M in debt relief for 55,000 consumers. The smoking gun was CAC's own internal predictive model showing expected defaults. Here's what every consumer lender needs to do before November.
Oct 2, 2026