Workforce · 작업자 관리

Warehouse Workforce Management
Why labor cost leaks even when everyone hits their numbers

Plenty of centers pull the same report every month: a UPH leaderboard by worker. The top of the list gets recognition or an incentive, the bottom gets a conversation. The ranking is crisp, and it definitely feels like management happened.

And yet there's one question that report never answers. Why did labor cost go up this month?

Labor cost rising while everyone hits standard is common. So is the reverse — a month where the bottom of the list didn't move and labor cost fell anyway. If it feels strange that the leaderboard and the labor bill move independently, that's because they were answering two different questions all along.

This piece starts by separating those two. In "Understanding Standard Costing in Logistics" we established that more than half of warehouse cost is labor, and that labor always decomposes into time × rate. This time we're on the other side of that equation — the people spending the time.

The two questions are not the same

There are broadly two questions you can ask in workforce management.

QuestionMeasuresControls
A. Did this person hit standard?The individualIndividual efficiency, training needs
B. Were the right number of people in the right places to hit today's throughput?The buildingTotal labor cost

Most centers only look at A. But the one actually holding the cost structure is B.

Check it with arithmetic. If a process that needs 12 people gets 16, then all 16 can hit 100% of standard and four people's worth of labor still walks out the door. Nothing shows up on the leaderboard — everyone rates excellent. Reverse it: put 14 people in a window that needs 20, and the leftover volume rolls into overtime at a rate above the day shift. And here the leaderboard actually improves — short staffing means less waiting, so individual UPH climbs.

Individual performance metrics are designed to stay silent about staffing errors. The denominator is the individual's hours.

One more line item hides here. Industry analysis puts roughly 25% of total paid hours into non-productive activity: travel, shift handoffs, waiting, manual check-ins. None of that time lands in anyone's UPH, and all of it lands on payroll. Which is why raising labor utilization (productive hours ÷ paid hours) by just 5% is estimated to save a mid-size DC 400,000–700,000 USD a year.

The metric to watch isn't UPH — it's labor cost per unit — UPH measures a person's speed; labor cost per unit measures the building's efficiency. The first is a coaching tool, the second is a management number. Centers where every individual UPH rises while cost per unit holds flat are not rare — the extra speed simply converted into more waiting.

Make the standard a scorecard and the floor will game it

Wire performance-against-standard directly into individual evaluation and people optimize remarkably fast. It's just that what they optimize is the metric, not the work.

The patterns repeat almost everywhere.

That metrics drive behavior isn't the bad news. The bad news is that badly designed metrics drive bad behavior.

And a safety dimension sits on top. U.S. warehousing logged an injury rate of 4.8 cases per 100 FTEs in 2024, and musculoskeletal disorders from overexertion and repetitive motion run around 31.8 cases per 10,000 full-time workers economy-wide, with transportation and warehousing among the highest. It's why the relationship between production quotas and injuries has come under regulatory review in the U.S. Use a standard as a pressure device and people don't get faster — they start reaching past their limits.

If you're going to use it for evaluation anyway — you need three corrections at minimum. ① Normalize task difficulty — recompute standard time weighted by line count, UOM, and travel distance. ② Separate tenure bands — covered below. ③ Separate indirect time — capture housekeeping, recovery, and training under their own codes and pull them out of the UPH denominator. Run a leaderboard without those three and the winner is simply whoever got assigned the easiest work.

The primary purpose of standard costing is attributing cost to activities; individual evaluation is a by-product. Reverse the order and the first thing to break is ledger data quality — because people start managing their records.

New hires can't hit standard. That's normal.

The learning curve is closer to physics than to something you manage. Floor practice applies standard attainment in steps.

TenureExpected attainmentWhat to do in this band
Weeks 1–370%Accuracy first, no speed demands
Weeks 4–885%Complete proficiency in one process
Week 9 onward100%Begin cross-training on a second process

Without this split, an efficiency dip in a month heavy with new hires gets misdiagnosed as a bad standard or a lazy floor. The actual cause is workforce composition.

So how much does turnover really cost in effective headcount? Let's compute it for a 100-person center.

ScenarioPeople still rampingWeighted attainmentEffective loss
36% annual turnover · even inflow2 in weeks 1–3 · 3 in weeks 4–898.9%1.1%
60% annual turnover · even inflow3 in weeks 1–3 · 6 in weeks 4–898.1%1.9%
20 hired at once before peak20 in weeks 1–394.0%6.0%

The conclusion is counterintuitive. When turnover is spread evenly, the learning-curve loss is only 1–2%. Warehouse turnover gets cited anywhere from 20% to 60% annually depending on how it's counted, averaging around 36% — and spread evenly, the effective loss is surprisingly small.

The problem is that turnover and hiring never happen evenly. You put 20 people in at once ahead of peak, and they leave at once after. And the month you dropped in those 20 is precisely the month you have to push 30% more volume.

Carry the arithmetic. Volume up 30% means 130 people's worth of work at standard. But with 20 of them new (70%), effective capacity falls short — so you actually have to staff 136 people to get 130 people's output. Six people's worth of labor cost vanishes into the learning curve. And not one bit of it is any worker's performance problem.

The cost of turnover hides in the average. The cost of hiring detonates at peak.

A cash cost sits alongside that. Replacing one warehouse worker is estimated at 25–150% of annual salary depending on the source. The range is wide because sources include different things, but the common items are recruiting, training time, lost productivity during ramp-up, overtime covering the gap, and the higher accident risk that comes with more inexperienced hands.

So half of peak hiring is timing — drop 20 people in two weeks before peak and all 20 meet the peak inside the 70% band. Stagger them starting eight weeks out and the same 20 meet it in the 85–100% bands. The labor cost difference shrinks from six people's worth to one or two. You aren't hiring fewer people — you're pulling the hiring curve ahead of the volume curve. That's a saving you get from planning alone.

Korea's labor shortage isn't only about failing to hire

The numbers are clear enough. In the Korean Ministry of Employment and Labor's second-half 2025 Occupational Labor Force Survey (as of October 1, 2025), the economy-wide shortfall was 449,000 workers and the Q3 unfilled rate was 8.4%. The shortfall itself fell by 78,000 (−14.8%) year over year, but roughly one in every dozen openings still goes unfilled.

The reasons are the interesting part. The top cause of unsuccessful hiring was no applicants with the required experience (26.9%), and second was pay and working conditions not matching what job-seekers expect (20.5%). In other words, a large share of the "shortage" is not an absolute lack of labor but a mismatch in terms.

Corporate responses point elsewhere. In the same survey, the leading effort to resolve shortages was increasing recruiting spend or diversifying hiring channels (62.6%), with raising wages and improving conditions (32.1%) at half that rate.

Twice the money on inflow, half as much on retention. Pouring more water into a leaking bucket. In a high-turnover center, adding recruiting channels treats the symptom; removing the reasons people quit inside 90 days treats the cause.

And wages alone won't differentiate you. Korea's 2026 minimum wage is 10,320 KRW per hour (+2.9% year over year), or 2,156,880 KRW a month at 40 hours a week. Warehouse floor roles mostly compete above that line already, so a few hundred won an hour doesn't distinguish you from the center next door.

What's left are the levers that aren't money.

Under subcontracting, task assignment becomes a legal question

A large share of Korean warehouse floor labor runs not on direct employment but on subcontracting or daily hire. And from here, workforce management overlaps with system design.

The crux is the line between subcontracting (도급) and worker dispatch (근로자파견). Ministry guidance and court reasoning look not at what the contract says but at the substance of operations. The factors weighed together are broadly these.

  1. Does the principal issue binding work instructions?
  2. Are the contractor's workers substantively integrated into the principal's operation?
  3. Does the contractor exercise staffing authority independently?
  4. Is the work specific and specialized?
  5. Does the contractor hold an independent organization and equipment?

Where illegal dispatch is found, the principal bears an obligation to directly employ those workers under Article 6-2(1) of the Dispatched Workers Protection Act.

Here's the point warehouses running a WMS easily miss. If the principal's system assigns tasks directly to individual workers employed by the contractor, scores them individually, and pushes their next task in real time — those logs themselves can become evidence for ① and ②. Instructions that used to be given on paper leave a far sharper trace once they move into software.

So the design used in practice is to change the unit of assignment. Instead of throwing tasks at individuals, hand volume to a team or to the contractor and let the contractor distribute it internally. The system supplies volume, deadlines, and quality criteria; who does what is the contractor's call.

This section is not legal advice — dispatch-versus-subcontracting determinations weigh contract form together with operational substance and turn on the specific case. The point here is one thing: it's worth checking once whether your warehouse's task assignment is direct-to-individual or delegated-to-team. If your workforce runs mainly on subcontracting, get a labor-law review alongside any change to those system settings.

One more thing: subcontracting also blurs cost visibility. When billing arrives as headcount × days or as a lump volume figure, actual time per activity never lands anywhere — and standard-cost variance analysis becomes impossible. Writing activity-level reporting into the subcontract is a precondition for cost management.

Multi-skilling isn't a buzzword, it's a capacity device

Volume swings by hour, weekday, and season; people can't be added and removed by the hour. The device that absorbs that mismatch is multi-skilling.

In a center staffed only with single-skilled people, two strange things happen at once. One process waits while another is the bottleneck. Pickers idle through a morning heavy with inbound; inbound staff sit around in an afternoon closing on cutoff. The two groups cannot help each other — they learned different things.

The management tool is a skills matrix: a grid of people × processes with proficiency marked in each cell. It moves the question from the individual to coverage.

How you read itQuestionFailure signal
Process columnHow many people can staff this process?1 = the process stops when they're off
Person rowHow many processes can this person run?All at 1 = idle and bottleneck coexist
Whole gridHow many can be redeployed at peak hour?0 = peak always means overtime

The goal isn't to make everyone multi-skilled in everything. Switching has a cost, and someone who does a bit of everything hits standard in nothing. A workable baseline is at least two people per process, and three or more on the processes where peaks land.

Two side effects follow. One is turnover resilience — with multi-skilled cover, one departure doesn't stop a process. The other is retention: the "skills path" from the previous section becomes something that actually exists. Positions with something to learn hold people longer than positions without.

Connecting back to the costing series — in accounting terms, multi-skilling is a device that shrinks volume variance, because staff on a process that lost volume can be shifted to another, reducing unabsorbed fixed cost. "Understanding Standard Costing in Logistics" assigned volume variance to sales, but the capacity to absorb it is built by operations.

Where workforce management is shifting in 2026

① From hiring to growing.

In MHI's annual industry report, 52% of companies rated hiring and retaining workers extremely challenging and 45% cited a talent shortage. The more telling move is in the response: the share of organizations planning to increase re-skilling and retraining efforts rose from 25% in 2024 to 38% in 2025, up 13 percentage points. MHI again named workforce and the talent gap among the most impactful issues for 2026.

② Automation doesn't remove people, it changes them.

Industry estimates put process improvement at 15–30% labor savings with almost no capital outlay, while automation delivers 30–50% but demands an 18–36 month payback. What changes after automation is the character of the remaining work. Equipment takes the repetitive tasks; what's left for people is exception handling, judgment on anomalies, and supervising equipment. Required skill goes up — and training quietly becomes part of automation's cost.

③ The burden of maintaining standards itself grew.

E-commerce order complexity has pushed the number of distinct DC task types up 3–4× since 2018, with omnichannel centers going from roughly a dozen task types to 40–50. More tasks means a wider span demanded of multi-skilled staff and more standards to keep current. Half the reason best-practice recalibration is 18–24 months while reality is 3–5 years lives right here — there are simply too many standards to maintain.

④ A realistic expectation for labor software.

Deployments of dedicated labor management systems (LMS) report labor cost reductions of 10–34%, but a realistic payback is 12–18 months, not the six months vendors promise. And there's an important caveat: software cannot fix a process that is already broken. Bolt a leaderboard onto a center with no assignment discipline and unmanaged chaos simply becomes measured chaos.

⑤ Gamification is a tool, not a strategy.

Live leaderboards and incentives genuinely work. But section two applies unchanged. Show individual rankings alone and metric-gaming accelerates; include team metrics and indirect work and you get a safety rail. The question to ask before hanging a scoreboard isn't "what should we display" but "if this ranking goes up, does labor cost per unit go down?"

So what should you measure

Switching metrics is the cheapest improvement available. The right-hand column below actually moves cost.

What's usually measuredWhat to measure insteadWhy
UPH per workerLabor cost per unitIt's a throughput problem, not a speed problem
Total hours workedUtilization (productive ÷ paid)The non-productive 25% surfaces here
Average standard attainmentAttainment by tenure bandSeparates composition shift from efficiency decline
Monthly absence rateSame-day no-show rate + time to backfillAbsence is expensive because it wrecks staffing
(usually nothing)Planned vs. actual staffing gapMeasures question B directly
(usually nothing)Available headcount per processMulti-skill coverage = capacity resilience

Six things are enough for the first month.

  1. Create indirect work codes. Housekeeping, recovery, training, waiting, each under its own code. That alone makes utilization visible for the first time.
  2. Put tenure bands in the ledger. With hire dates alone, the 70/85/100 bands compute themselves.
  3. Record planned and actual headcount side by side, by time window. Four windows a day is a fine start.
  4. Build the skills matrix on one page. A spreadsheet is fine. Find the cells where a process has exactly one available person first.
  5. Apply difficulty weighting to the individual leaderboard — or take it down for now. An unadjusted leaderboard eats your ledger quality.
  6. Pull the hiring curve 6–8 weeks ahead of the volume curve. You can apply this to the very next peak plan.

All six share one premise. Who spent how many minutes on which activity has to land in the ledger. Without activity-level records, utilization, attainment by tenure, and staffing gap are all uncomputable. The starting point of workforce management isn't an HR policy — it's a work record.

You aren't managing people, you're managing work

Compress the argument into one line: the name "workforce management" creates the misunderstanding. What actually needs managing isn't people; it's the placement of work.

There are centers where labor cost moves 10% with the same people. Nobody got faster — the right number were simply in the right places. And there are centers where tightening the leaderboard leaves labor cost immovable. Not because people are lazy, but because the problem was never individual speed.

None of this makes individual performance management pointless. It remains necessary as a basis for coaching, and nothing replaces it for finding workers whose accuracy is slipping. But try to control cost with it and you get stuck in the state where the leaderboard is perfect and the labor bill hasn't budged.

The money you save from people moving faster is smaller than the money you save from people standing in the right place.

Docktre records the work, not the person

Who spent how many minutes on which activity lands in the work ledger, alongside tenure bands, indirect time, and team-level assignment. If you'd like to see it in your operation, get in touch.

Contact

Related reading