The Experience Cliff

What succession planning has to do differently when the knowledge has already walked out the door

Every federal workforce planner has run the standard succession exercise. Identify the mission-critical roles. Project retirement eligibility three and five years out. Build a bench behind each position and develop it.

The exercise assumes time. More precisely, it assumes that the people holding the institutional knowledge are still at their desks while you build the pipeline behind them.

For a growing number of agencies, that assumption no longer describes reality. The departures already happened. The planning question has changed from how to prepare for a loss to how to reconstruct what is already gone.

The Numbers, and Exactly What They Cover

Start with the case that should concern federal HR directors most, because it happened at the government’s own human capital agency.

Between December 31, 2024 and March 31, 2026, the Office of Personnel Management’s headcount fell by 1,052 people, a decline of 35 percent. The Government Accountability Office published that analysis on July 20, 2026, in GAO-26-108916, drawing on OPM’s own Federal Workforce Data.

Of the 1,235 employees who separated from OPM during that 15-month window, 57 percent had 11 or more years of service and 18 percent had at least 31 years. GAO’s own characterization was direct: this represents “a significant loss of institutional knowledge.”

One precision note, because this audience will check it. Those percentages describe separations from OPM as an employer. They are not government-wide separation figures, and they should not be cited as such. The reason to lead with them is not scale. It is that OPM is the agency responsible for federal human capital policy, and it absorbed a 35 percent reduction weighted toward its most experienced staff.

The composition of the loss matters more than the headline. Fifty-nine percent of those separations came through the deferred resignation program and 10 percent through a reduction in force. Employees aged 60 and older declined by 49 percent. At the same time, 41 percent of OPM staff under age 30 separated. The agency lost the top of the experience curve and the bottom of the pipeline in the same period, which is the specific pattern that makes recovery slow.

For government-wide scale, OPM reported roughly 317,000 departures from the federal workforce during calendar year 2025 against approximately 68,000 hires.

As for the current shape of the workforce, OPM’s Federal Workforce Data platform reports a median federal employee age of 47 and 13.5 percent of federal employees currently eligible to retire, using data as of July 2026.

Two notes on the source, because both affect how the number should be read. Federal Workforce Data replaced FedScope in January 2026 and is now OPM’s authoritative public workforce data platform; a figure still cited to FedScope in 2026 is being drawn from a retired system. And OPM currently carries a caveat on the platform itself: “A change in Department of War (DOW) data processing prevented some components from submitting June 2026 and July 2026 data.” Non-DOW data is unaffected, but any DOW-inclusive cut of the July 2026 release should be treated as provisional until the outstanding submissions land.

Succession Planning Assumes a Future Loss. This One Already Happened.

Conventional succession planning is an anticipatory discipline. Its entire toolkit assumes an overlap period: a named successor who shadows an incumbent, a developmental assignment that runs 18 months, a phased retirement that keeps a senior specialist available part-time. Each of these requires the departing expert to still be reachable.

Remove the overlap and most of the toolkit stops working. A bench plan for a position whose last three qualified incumbents left in the same quarter is not a succession plan. It is a recruitment requisition with a chart attached.

GAO has made this point for years. Its 2019 work on federal talent pipelines warned that when turnover is not strategically managed and succession plans are not in place, gaps develop in an agency’s institutional knowledge and leadership. Federal skills gaps have been on GAO’s High-Risk List since 2001. At OPM specifically, GAO’s February 2023 recommendation to build an action plan for identified skills gaps remained open at the time of the July 2026 report, with the agency citing competing workforce priorities. The recommendation predates the departures. The departures arrived first.

What Changes When the People Are Already Gone

The shift is from pipeline construction to knowledge reconstruction, and the two are not the same activity.

Pipeline work develops people toward a known standard. Reconstruction work has to first establish what the standard was. When a 31-year employee leaves without a transition, what departs is rarely documented procedure. It is the judgment layer: which exceptions are real exceptions, which stakeholder to call before a decision goes final, why a workaround exists, which edge cases the written guidance never anticipated. That layer is tacit, and it is not in the shared drive.

Three practical consequences follow for HR directors.

• Your process documentation is now a hypothesis, not a record. Written procedures were maintained by people who knew where the documentation diverged from practice. Treat every procedure inherited from a vacated role as unverified until someone executes it end to end.
• Risk concentrates in the handful of tasks nobody has run yet. The damage from an experience cliff does not appear evenly. It appears at the first annual, quarterly, or exception-driven task that the departed employee used to handle alone. Those tasks are often invisible on an org chart and absent from a position description.
• Remaining staff become the reconstruction instrument and the retention risk simultaneously. The same people who can help rebuild the knowledge are the ones absorbing the workload of the vacancies. Load them without relief and the second wave of departures is self-inflicted.

The Compounding Factor: The Backlog Shrank. The Wait Did Not.

The retirement processing picture deserves care, because the widely repeated version of it is now out of date.

OPM’s retirement claims inventory did hit a modern record of 65,237 pending claims in February 2026. It has fallen every month since: 55,681 in March, 49,888 in April, 38,547 in May, 33,851 in June, and 24,784 in July 2026. That is a decline of roughly 62 percent from the February peak, and July 2026 is the most recent month published as of this writing. Any article claiming a currently record-high backlog is describing February, not now.

The compounding problem is that clearing inventory has not restored speed. Claims completed in July 2026 took an average of 109 days, essentially unchanged from 108 days in June. OPM’s own targets are 45 days for all cases and 15 days for digital submissions. Applications filed through the Online Retirement Application averaged 98 days; paper claims averaged 156.

Two structural reasons explain the gap, and both matter for planning. First, OPM worked the simple cases first, so the remaining inventory is weighted toward claims requiring military service deposits, survivor elections, court orders, or corrections to incomplete files. Second, OPM’s Retirement Services office itself shrank 16 percent in full-time equivalents between fiscal years 2024 and 2026, even as it absorbed a historic retirement surge.

For a workforce planner, the operational translation is simple. An employee who separates is off your rolls immediately and unavailable for knowledge transfer immediately, but may spend three to five months on interim payments before their annuity is final. That gap is a real, and largely unused, window. A recent retiree in month two of interim payments is often willing to take a call, sit for a recorded interview, or accept a short reemployed-annuitant engagement. Agencies that plan for that window recover knowledge. Agencies that do not, lose the contact entirely.

What Actually Transfers Tacit Knowledge Under These Conditions

The methods that work after the fact are narrower than the ones that work with an overlap period.

What does not work:

• Documentation sprints assigned to people who never did the job. They produce volume, not judgment, and they generate false confidence because a binder now exists.
• Exit interviews as a knowledge instrument. They are designed to capture why someone left, not how they worked, and they happen at the least cooperative moment in the relationship.
• Generic mentoring programs. Mentoring requires a mentor. This is the one input the experience cliff removed.
• Competency frameworks written in the abstract. They describe the role as designed, not as it was actually performed by the person who left.

What does work:

• Structured retrospective interviews with recent separations, conducted while they are still on interim annuity payments and still reachable. Ask for decisions and exceptions, not process steps.
• Reemployed annuitants and short-term rehire authorities used surgically — scoped to a specific knowledge target and a defined end date, not open-ended backfill.
• Task-execution audits. Have a remaining employee run the actual procedure while a second person records every point where the written guidance failed. This surfaces the judgment layer faster than any interview.
• Peer-agency sourcing. The same role exists at other agencies. Where the knowledge is regulatory or procedural rather than agency-specific, borrowing a practitioner is faster than rediscovery.
• Cohort-based reconstruction. Put three or four remaining staff on the same recovery target together. Collective recall reliably outperforms individual recall, and it distributes the resulting knowledge across more than one person, which is the failure that created the cliff.

A Realistic First 90 Days

For an agency that has already taken the loss, the following sequence is achievable with existing staff and existing authorities.

Days 1–30: Establish what actually left

1. Inventory departures by length of service, not by headcount or grade. An 18 percent loss weighted toward 20-year employees is a different problem than the same percentage weighted toward new hires.
2. Identify orphaned tasks, not orphaned positions. Ask each supervisor a single question: what did the person who left do that nobody here has done yet?
3. Build a contact list of recent separations while contact information is still current and the relationship is still warm. This has a short shelf life and costs almost nothing.

Days 31–60: Recover what is recoverable

1. Run structured retrospective interviews against the orphaned-task list. Prioritize tasks with a statutory or regulatory deadline in the next two quarters.
2. Execute task-execution audits on the highest-risk procedures. Document divergence between written guidance and actual practice.
3. Scope any reemployed-annuitant or peer-agency request now, so the approval clock runs in parallel rather than after.

Days 61–90: Stabilize and prevent the second wave

1. Rebalance workload for the staff absorbing vacancy coverage, and say so explicitly. Silent reassignment is how the next departures start.
2. Set a minimum redundancy standard: no single-point-of-knowledge task without a named second person, with a date.
3. Reset the succession plan against the workforce you have now, not the one the last plan assumed. Pull current retirement-eligibility data from OPM’s Federal Workforce Data platform and date it to the month you pulled it.

The Reframe

Succession planning built for anticipated loss is a scheduling problem. Succession planning after an experience cliff is a forensics problem, and it rewards a different set of moves: speed in reaching people who just left, honesty about which documentation is fiction, and a hard bias toward recovering judgment rather than cataloging process.

The agencies that recover fastest will be the ones that stop treating the last 20 months as a staffing shortfall to be backfilled and start treating it as institutional knowledge to be reconstructed. Those are not the same project, and only one of them can be solved by hiring.

Sources

1. U.S. Government Accountability Office, “GAOverview: Recent Federal Workforce Changes at OPM,” GAO-26-108916, published and publicly released July 20, 2026. (OPM headcount −35% / −1,052 Dec 31 2024–Mar 31 2026; 57% of separations with 11+ years and 18% with 31+ years, n=1,235; 59% deferred resignation, 10% RIF; employees 60+ down 49%; 41% of under-30 staff separated; Retirement Services FTE −16% FY2024–FY2026; GAO analysis of OPM Federal Workforce Data obtained June 2026.) https://www.gao.gov/products/gao-26-108916
2. U.S. Government Accountability Office, “A Framework for Managing Fraud Risks / federal talent pipeline and succession findings,” GAO-19-181. (Cited within GAO-26-108916 for the finding that unmanaged turnover without succession plans produces institutional knowledge gaps.) https://www.gao.gov/products/gao-19-181
3. U.S. Government Accountability Office, GAO-23-105528, February 2023 recommendation that OPM establish an action plan to address identified skills gaps — reported as still open in GAO-26-108916. https://www.gao.gov/products/GAO-23-105528
4. U.S. Office of Personnel Management, Federal Workforce Data (FWD), “Demographics.” Data as of July 2026; site last updated September 2, 2026. (Median federal employee age 47; 13.5% of federal employees currently eligible to retire; 553,863 veterans in the federal civilian workforce.) FWD replaced FedScope in January 2026. https://data.opm.gov/explore-data/analytics/demographics
5. U.S. Office of Personnel Management, Federal Workforce Data (FWD), “Workforce Changes.” https://data.opm.gov/explore-data/analytics/workforce-changes
6. U.S. Office of Personnel Management, Retirement Processing Status Report — monthly inventory and average processing times. (February 2026 peak 65,237; March 55,681; April 49,888; May 38,547; June 33,851; July 24,784. July 2026: 10,538 new claims, 17,841 processed, 109-day average processing time; ORA digital 98 days vs. paper 156 days.) https://www.opm.gov/retirement-center/retirement-statistics/retirement-processing-status.pdf
7. U.S. Office of Personnel Management, “Retirement Processing Times” — published service standards of 45 days for all cases and 15 days for digital applications. https://www.opm.gov/retirement-center/apply/retirement-processing-times/
8. My Federal Retirement, “OPM Retirement Backlog Falls Below 25,000 in July 2026,” August 5, 2026 — secondary compilation of the OPM Retirement Processing Status Report monthly series used to cross-check the figures above. https://www.myfederalretirement.com/opm-retirement-backlog/
9. U.S. Office of Personnel Management, “OPM Marks Last Day of Paper, Ending Paper Retirement Processing After More Than 65 Years,” effective July 1, 2026. https://www.opm.gov/news/news-releases/opm-marks-last-day-of-paper-ending-paper-retirement-processing-after-more-than-65-years/
10. FEDmanager, “OPM Reports 317,000 Fed Departures in 2025, DoD Largest Drop by Number.” (Government-wide CY2025 departures ~317,000 against ~68,000 hires.) https://www.fedmanager.com/news/opm-reports-317000-fed-departures-in-2025-dod-largest-drop-by-number
11. U.S. Office of Personnel Management, “OPM Launches New Workforce Data Site to Modernize Access and Increase Transparency” — FedScope-to-FWD transition, January 2026. https://www.opm.gov/news/news-releases/opm-launches-new-workforce-data-site/