Capacity planning goes wrong in a specific way. Someone calculates theoretical capacity, compares it to forecast demand, concludes there’s headroom, and then the plant misses dates anyway. The arithmetic was fine. The inputs were fantasy.

Step One: Measure Demonstrated Capacity, Not Theoretical

Theoretical capacity is machine rate multiplied by available hours. Demonstrated capacity is what you have actually produced, historically, under real conditions including changeovers, breakdowns, quality holds, breaks, and the shift that ran short-staffed. The gap between the two is typically wide — and it’s the demonstrated figure that predicts what you can promise.
Pull twelve months of output by work centre and look at the distribution, not just the average. Your median week and your best week tell you different things, and planning to the best week is how commitments get made that only hold if nothing goes wrong.

Step Two: Find the Actual Constraint

Capacity is set by your bottleneck, not by your average utilisation. A plant running at sixty percent overall can be completely blocked if one work centre is at ninety-eight. Adding capacity anywhere other than the constraint produces nothing except more work in progress queued in front of it.
Constraints also move. A bottleneck under one product mix may not be the bottleneck under another, which is why a shift in your order book can cause sudden unexplained lateness in a plant that was coping fine the month before. Identify the constraint under each of your major mix scenarios, not once in general.

Step Three: Plan in Three Horizons

Long range, twelve to twenty-four months, is about whether you need structural change: equipment, space, headcount. It runs off aggregate volume and needs scenario thinking rather than precision, because the forecast at that distance won’t be accurate and pretending otherwise wastes effort.
Medium range, one to six months, is about labour, shifts, and outsourcing decisions. Here product mix matters, because mix determines which constraint binds.
Short range, one to four weeks, is sequencing and load balancing against firm orders. This is execution, and it needs live data rather than a plan.

Step Four: Model Mix, Not Just Volume

This is the step most often skipped. Capacity expressed as units per week is only meaningful if your product mix is stable. Two products with identical volumes can consume wildly different amounts of constraint time. Express capacity in constraint-hours consumed, and your plan survives a mix change.
The practical version: build a table of constraint-minutes per unit for each major product, then multiply through your forecast mix. If the total exceeds available constraint-hours, you have a capacity problem regardless of what your unit-based calculation said. Manufacturers who make this switch often discover they’ve been running far closer to the edge than they believed, particularly during periods when high-content products dominate the order book.

Step Five: Stress-Test It

Before committing, run the plan against plausible bad cases. Demand fifteen percent above forecast. A key supplier two weeks late. The constraint machine down for a week. If the plan only works when everything goes right, it isn’t a plan — it’s a hope with a spreadsheet attached.
ticktick.ai models capacity in constraint-hours across product mix and runs scenario comparisons, so you can see where a plan breaks before you commit to dates.