Industry: Manufacturing
Product: ModelRisk
Application: Workforce Planning under Seasonal Demand

Why the $180k Cross-Training Programme Loses to a Temp Pool: A Twelve-Month Workforce Simulation

An automotive supplier with 72 FTE on the line was running its 95% on-time-delivery commit with no tail margin under fixed-headcount staffing — a mean OTD of 97.6% that looks comfortable until you notice the P10 sits right on the 95% line, so roughly one year in twelve dips below the SLA, almost always during the October–November demand peak when seasonal volume rises 25% above mean. HR's plan was a $180k/year cross-training programme to absorb absenteeism shocks. Operations preferred a flex pool of temp workers at a 21% wage premium. The plant manager wanted a number, not an argument. The team built a 12-month Monte Carlo workforce model in ModelRisk and ran the four candidate policies on the same 5,000 simulated years. The verdict: the temp-pool strategy lifts annual on-time delivery from 97.6% to 100% at +$60k of cost, while the cross-training programme lifts OTD to only 98.9% at +$180k. Cross-training as a standalone intervention is dominated. Combined with a temp pool, it adds $130k of redundant insurance and does not improve P10 OTD further.

The four annual-cost curves separate cleanly: fixed is cheapest but leaves a tail below the SLA, flex sits just $60k to its right while erasing that tail, and the two cross-training policies cost the most while doing no better than flex on delivery. The cheapest policy that actually clears the commit every year is the temp pool.

Three sources of variability, twelve months at a time

The line ships into a market with a stable seasonal pattern — mean 22,000 units/month, multiplied by a known seasonal index (low in winter, peak 1.25× in October). Three stochastic processes ride on top of the seasonal mean:

• Monthly demand: Normal with CV = 12% around the seasonal mean. Includes both customer order-book noise and the occasional pull-forward request.
• Per-unit productivity: LogNormal hours-per-unit around a base of 0.50 hr, σ_log = 0.10. LogNormal because hours-per-unit is right-skewed — most months hit the mode, but a labour-relations week or a quality-rework run lifts the tail.
• Absenteeism rate: Beta(2, 38) per month, mean ≈ 5%, with the right tail reaching 15–20% during a bad flu month or vacation cluster.

Fixed-headcount capacity at 72 FTE × 165 hr/month is 11,880 hours, or 23,760 units at base productivity. The seasonal demand fan tells the story:

October and November median demand sits above base capacity, and the P90 reaches 30,000+ units — well clear of the 23,760-unit ceiling. Either overtime, temp pool, or cross-trained backfill is required to bridge the gap. The deterministic plan that uses mean demand without the absenteeism overlay shows November as just-covered; the simulation says it isn't.

Four candidate policies

• Fixed headcount: 72 FTE, mandatory overtime capped at 20% of available hours. Anything beyond that is a late-delivery penalty at $14/unit.
• Flex (temp pool): 72 FTE + temp workers called in on demand. Temp wage $34/hr (vs $28 regular), 25% productivity penalty (~longer ramp). Overtime cap reduced to 10%.
• Cross-training: 72 FTE with $180k/year programme that effectively absorbs 60% of monthly absenteeism (a cross-trained worker can step into the absentee's role). Overtime cap raised to 25%.
• Flex + cross-training: both, with overtime cap at 15% and a 10% temp productivity penalty (the cross-trained pool helps temps ramp).

What the deterministic comparison missed

A deterministic capacity-vs-demand spreadsheet using mean demand and zero absenteeism reports all four strategies as equivalent — each meets the seasonal-mean demand within the wage envelope. The Monte Carlo reports something different.

Annual cost across 5,000 simulated years:

Flex costs $60k/year more than fixed and buys complete OTD coverage. Cross-training costs $180k/year more and still misses the 100% mark. The flex+cross combination adds another $130k on top of flex and does not improve the OTD distribution further — pure redundant insurance.

On-time-delivery is the right success metric

Plotting the annual on-time-delivery rate as a CDF clarifies why fixed headcount is commercially fragile even though its mean OTD looks comfortable at 97.6%: the red curve crosses the 95% SLA line at about the 8th percentile, so roughly one year in twelve breaches the commit — and its P10 sits right on 95.3%, leaving no tail margin at all. A mean does not tell you whether you make the SLA in a bad year — the distribution does. Flex and flex+cross both put the entire annual-OTD mass at or above 99%. Cross-training alone tightens the distribution but still leaves a thin left tail below 95%.

What actually moves labour cost

Sensitivity analysis on the fixed-headcount baseline ranks the drivers of annual labour-cost variability:

The seasonal-peak multiplier (Oct/Nov demand index) and per-unit productivity sigma are the two biggest movers — each worth ±$240–360k on annual cost across plausible ranges. Absenteeism comes third. Late-delivery penalty rate sits fourth, which is exactly why the OTD-vs-cost trade-off is dollar-balanced rather than one-sided: shaving penalty exposure costs roughly as much as it saves, until the right policy is found.

From insight to action

• Temp-pool agreement signed with a regional staffing agency, callable up to 6,000 hours/year, with priority during Sept–Dec. Annual incremental cost +$60k vs fixed; OTD lifts from 97.6% → 100% mean and from 95.3% → 100% at P10. Payback in customer-SLA penalty avoidance under 6 months.
• Cross-training programme deferred as a standalone option. The simulation showed it is dominated by the temp pool; HR has retained the cross-training curriculum design for future use if the temp market tightens.
• Heaviest-month commits revised: October and November commit dates extended by one shift on the customer schedule, reducing late-delivery penalty exposure in the tail months the simulation flagged.
• Workforce dashboard live: the simulation workbook now refreshes monthly with the latest demand, productivity, and absenteeism distributions, and the temp-call decision uses the workbook's mid-month forecast rather than a heuristic.

ModelRisk Functionality Used

• Per-month stochastic demand, productivity, and absenteeism — Normal CV 12% on demand, LogNormal σ_log 0.10 on productivity, Beta(2, 38) on absenteeism — composed into a 12-month annual simulation, replacing the deterministic mean-demand capacity planner.
• Strategy switch as a single workbook parameter so the same simulation engine evaluated all four policies on identical demand and absenteeism traces, eliminating run-to-run noise from the comparison.
• OTD CDF outputs that exposed the fixed-headcount strategy's thin tail margin — a P10 sitting right on the 95% commit and roughly one breach year in twelve — a fragility invisible in the deterministic plan's mean OTD of 97.6%.
• Tornado on labour-cost drivers that ranked seasonal-peak demand and productivity σ_log above absenteeism, redirecting the HR conversation from absenteeism management to demand-pull-forward conversations with the customer.
• Dominance check across all four strategies showing flex+cross to be redundant — the workbook explicitly computes "is policy B Pareto-dominated by A" and flagged the cross-training-only and combination options before the budget meeting.

The Monte Carlo lesson for workforce planning is that average headcount is the wrong question. The right question is the distribution of the gap between supply and demand across the year, and which lever — overtime, temp pool, cross-training — closes the worst months for the fewest dollars. Once the distribution is visible, the dominated options eliminate themselves.