The City Keeps Losing the Workers It Needs

The Philadelphia Fire Department retained 96.4% of its workforce from 2023 to 2024. That number is unremarkable in the sense that most well-managed organizations hit it without trying. In the private sector, the rule of thumb is that 90%+ annual retention is excellent in skilled professional work. Above 85% is acceptable. Below 80% triggers action. Below 75% would generate emergency meetings in any organization with a functioning HR department.

The Mayor's Office retained 50.7% over the same period. The Civil Service Commission retained 66.7%. The Office of the Chief Administrative Officer retained 58.3% — a department of 139 people that lost more than a third of its workforce in a single year and replaced most of them. Numbers that would generate board-level attention, outside consultants, and strategic reviews in any private organization. In city government, they appear in no budget document as a problem.

The variation in retention rates across Philadelphia departments is not a management problem — it's a structural feature of a pay system that makes experienced workers disproportionately expensive to lose. Hiring more people won't fix that.

What the Benchmarks Actually Are

In professional services — law firms, consulting, accounting — the rule of thumb puts 85-90% retention at the floor for organizations serious about maintaining expertise. Above 90% is good. Above 95% is excellent. These are not arbitrary numbers. They reflect the reality that skilled professional work requires months of onboarding, often years before a worker reaches full productivity. Every worker who leaves before that threshold represents a net training investment lost.

The private-sector benchmarks that matter are not arbitrary. They are derived from the arithmetic of onboarding. Skilled professional work requires months of onboarding, often years before a worker reaches full productivity. Every worker who leaves before that threshold represents a net training investment lost — and in municipal work, where the job-specific knowledge is dense (procurement rules, union contracts, regulatory compliance, institutional relationships), that threshold is higher and the loss is larger.

Philadelphia's best departments are in that range. Fire at 95-96%. Police at 84-92% across the two years. The Board of Ethics at 90-100%. These are organizations that have either found a formula that works or are in a labor market where retention is naturally achievable. The variation across departments is not random noise. It is a signal — some departments are meaningfully worse at keeping experienced workers, and the cost of that gap does not appear as "retention spending" in any budget line. It shows up as overtime, as training, as service quality degradation, and as higher per-worker compensation for those who remain.

The Replacement Load: Who Is Actually Replacing Whom

When you decompose total hiring into replacement hires and net new hires, some departments are spending nearly all their hiring budget just to stand still. This is the cost that does not appear in any budget document — not as a recruiting cost, not as a turnover cost, not as a productivity loss. It just shows up as headcount.

The chart below shows the churn cost for the top-20 departments by total hiring volume in 2024-2025. The orange portion of each bar is the number of hires that went purely to replacement — filling the hole left by someone who left. The purple portion is net new hiring — actual headcount growth. In departments like Fire, Prisons, and Aviation, the replacement load is large because the departments are large. But in departments like CAO, Finance, and City Commissioners, the replacement load is nearly as large as the total hiring — meaning most of the hiring energy is going to stand still just to maintain existing headcount.

Retention and Headcount Growth: Not the Relationship You'd Expect

The intuitive assumption is that departments with good retention grow their headcount through net new hiring, while departments with poor retention are shrinking. The data shows the opposite is sometimes true. The scatter plot below plots each department by its average retention rate against its headcount growth. If the relationship were clean, you'd see dots trending up and to the right.

What you actually see is a cloud. Some departments have high retention AND high headcount growth — top right, Fire is here. Others have low retention AND high headcount growth — meaning they are growing their workforce while simultaneously losing experienced workers at alarming rates. A department that can retain its workforce and grow simultaneously is doing something different from one that grows only by replacing departures at scale. The difference matters for cost structure even when the net headcount change looks similar.

The Compounding Cost of the Two-Tier Pay Structure

The cost of losing experienced workers is not proportional to their salary. It is disproportionately large — because city compensation is structured so that overtime access, longevity increases, and schedule seniority all accrue to workers who stay. The chart below shows what this looks like in the compensation data. Citywide, tenured workers earn roughly 138% more than new hires. That is not a productivity differential. It is a structural feature.

When an experienced worker leaves, the department does not simply save money. It pays the new worker AND continues paying the seniority-weighted overtime premiums to those who remain — workers who are now covering more shifts because the department is short-staffed. The new worker cannot immediately fill the overtime-access gap because they are new. The overtime doesn't evaporate when the experienced worker leaves. The cost just gets redistributed to fewer people, making the remaining workers more valuable and harder to replace in turn.

This is the spiral that makes the retention problem compounding rather than self-correcting. Departments that cannot retain cannot build experience depth. The workers who remain pick up more overtime. Higher overtime pay makes those workers more expensive and simultaneously more valuable to competitors. The retention problem gets worse. The cost per unit of output goes up.

The Payroll Fingerprint of Experience Premiums

In some departments, total payroll is growing faster than headcount can account for. Workers are getting more expensive per person even as headcount grows. This is what compounding experience premiums look like in aggregate — existing workers getting more expensive through seniority accruals and overtime exposure, while new hires do not offset the cost because they cannot access the same compensation channels.

The Sanitation Department: A Case Study in Low-Pay-Driven Turnover

The case that makes the selection problem most concrete is Sanitation.

The retention data for Sanitation is incomplete — the department was categorized differently across years, which makes the year-over-year comparison impossible. What the data does show is that in 2024-2025, the department hired 2,060 workers and retained essentially none from the prior year cohort. This is itself a data problem: when a city's own statistics cannot reliably track a department of 2,000 workers across years, the operational oversight is not there.

In July 2025, nearly 10,000 Philadelphia city workers went on strike — primarily sanitation workers, 911 dispatchers, and water department employees. The first major city strike since 1986. Residents named the trash piles after the mayor. The settlement was a $1,500 bonus and 3% annual raises for three years — well below what the union sought. At the wages Sanitation pays, the workers who stick around are those who cannot find better options. The retained workforce is selected for the wrong reasons. High turnover at the bottom of the pay scale is not just a morale problem. It is a selection problem.

The Departments That Break the Pattern

Some departments have good retention AND growing headcount. Fire has retained 95-96% of its workforce while adding staff. Streets has relatively stable headcount and lower overtime than Fire or Police. These departments exist — and what they have figured out is not visible in the data. But their existence rules out the structural determinism objection. The problem is not inevitable. Some departments have found a different way.

It does not mean these departments have solved the overtime problem. Fire has the highest overtime per employee of any tracked department — $30,794 per worker in 2025, more than double the citywide average of $15,048. Tenured Fire workers earn nearly 8 times what new hires earn in overtime alone. The overtime-concentration pattern exists in Fire just as it does everywhere else. What Fire has that some other departments lack is a retention model that holds — and a headcount trajectory that grows even as the overtime problem compounds. That is not a solution to the structural problem. It is a milder version of the same problem, sustained at lower cost.

The departments running the standard model — where seniority-weighted overtime flows to the same workers, where new hires cannot access the compensation channels tenured workers use — are paying a cost that does not appear anywhere in the budget. The budget documents show headcount and salary lines. They do not show the cost of a workforce that is structurally selected for seniority, where the workers who stay are the ones who couldn't leave, and where the cost of replacing the workers who could have left compounds with every year the problem goes unaddressed.

The overtime data, the payroll growth data, and the retention data are three views of the same underlying problem. This piece covers the retention angle. The overtime-concentration piece covers the distribution angle — how a small senior cohort captures a wildly disproportionate share of overtime dollars. The hiring piece covers why adding headcount does not fix it. These are not independent data posts. They are three cuts at the same structural problem — and the three together make the case that the problem is not a line item in any budget, which is precisely why it has not been addressed.

Notes, Sources, and Methodology

Figures are from the City of Philadelphia Employee Earnings dataset, published via OpenDataPhilly and accessed through the CARTO SQL API (source table). The data covers 2023–2025 and contains quarterly pay records for every city employee, broken out by category (base, overtime, longevity, post-separation, miscellaneous). Retention and tenure comparisons use an anonymized employee ID that is consistent across years.

The compensation analysis is restricted to Salaried employees; part-time and seasonal workers are excluded because their quarterly records don't represent a full-year salary. The retention analysis is not — it counts every employee who received pay in at least one quarter of each year.

One caveat. Sanitation is categorized inconsistently across years in the City's data, so its year-over-year retention rate is unreliable; the article uses a headcount-cohort frame for that department instead.

https://nickjain.com/blog/the-city-keeps-losing-the-workers-it-needs