High utilisation, growing work in progress and customer waiting.
All notes

Supply chain

Why busy machines make customers wait: Little's Law and takt time on the floor

Keeping every resource busy feels productive and often makes lead times worse. Two simple formulas explain why, and how launch times fell from five months to two.

When I ran the supply chain side of new product launches at British American Tobacco, launch lead time came down from about five months to two. People sometimes assume that meant everyone worked faster. Very little of it did.

Most of the gain came from deciding what not to start. Readiness gates, clear checkpoints and one cross-functional governance routine meant work only entered the system when it could actually flow through it. The calculations below connect those release decisions to work in process, throughput and the pace set by demand.

Little's Law, in one line

For any stable system, over a consistent period (Little, 2011):

Work in process = Throughput × Flow time
Average orders inside the line Average completions per day Mean days inside the line

Divide average work in process by average throughput to get mean flow time. On a saturated line, releasing more work can lengthen the queue while completions stay flat.

40 orders in process2 daysaverage flow time80 orders in process4 daysaverage flow timeBoth lines finish 20 orders a day. Each row is one day of output. The extra rows are pure queue.
At the same throughput, twice the average work in process means twice the average time inside the line.

Measure average work in process and average daily completions across the same line over a stable period. At 20 completions a day, an average of 40 orders inside the line gives a mean flow time of two days. If average work in process rises to 80 and throughput stays at 20, mean flow time rises to four days. Vrolen’s worked Little’s Law example shows how to set the measurement boundary.

Vrolen’s Little’s Law video works through the same calculation with two support desks.

Faster data can make it worse

Woschank and colleagues ran 5,400 simulations of a make-to-order electronics line. More frequent updates raised machine utilisation across the strategies they tested, but lead time moved in different directions. It rose under CONWIP, which controls work release with a limit across the line, and fell under Kanban, which uses local replenishment signals. Test how often you update the release rule as well as what the rule allows. Woschank et al., 2024.

Takt time: the pace the customer sets

Little's Law tells you what happens to work inside the system. Takt tells you the pace you need to hit. It's available production time divided by customer demand (Lean Enterprise Institute).

An illustrative shift:

Shift Value
Shift length 8 hours
Breaks and planned stops 60 min
Available time 420 min (25,200 s)
Customer demand 210 units
Takt time 25,200 ÷ 210 = 120 s per unit
Observed cycle time 144 s

A 144-second cycle exceeds the 120-second takt by 20%. Translate that difference into the number of units the team can finish in the shift:

Demand 210 units Capacity at a 144 s cycle 175 units 35 short 420 available minutes, takt 120 s, observed cycle 144 s. Before any downtime, changeover or quality loss.
Twenty percent slower than takt is 35 units short per shift, before a single breakdown.

Thirty-five units short, before any unplanned downtime. That's a far more useful sentence to put in front of a team than "cycle time is 20% over", because it tells everyone what the current process design can and can't deliver.

What to watch instead of utilisation

Instead of Watch Because
Utilisation of every machine Throughput of the constraint Only the constraint sets the pace of the whole line
Orders released WIP against a limit WIP above what the line can finish is pure waiting
Average cycle time Cycle time against takt, in units per shift "35 short" drives action; "20% slow" doesn't
Starts Flow time and on-time delivery That's what the customer actually experiences

How this showed up in product launches

A new product launch is a flow problem wearing a project plan. Materials, artwork, trials, quality sign-offs and line time all queue behind each other, and every half-ready launch that gets "started early" joins the queue.

What worked for us at BAT was treating readiness as a gate every launch had to pass:

  • Clear checkpoints with owners across sourcing, planning, manufacturing and quality, so problems surfaced while there was still time to fix them.
  • One governance routine across functions, instead of a dozen side conversations.
  • MRP-based planning across nine production lines and five warehouses, so material readiness was visible before a launch was released, not after.

Put simply: fewer things in flight, each one moving. That's Little's Law applied to a project portfolio.

A few questions for your own operation

  1. What decides when work is released: a due date, material availability, a WIP limit or the constraint's capacity?
  2. Do you know your takt time, and do your teams see the gap in units rather than percentages?
  3. If demand went up 10% tomorrow, would you release more work, or protect flow at the constraint?
Why a 5-Minute Task Takes 2 Days A separate office example: one five-minute approval waits two days in a busy team. It shows the same split between working time and waiting time.

Sources

Line and shift figures are illustrative. The launch figures are from my own work.

NextDid the improvement hold?

Discuss an operations problem

I work with operations and transformation teams on operational excellence, digital transformation programs, supply chains and industrial AI. Send me the problem you’re working on and the result you need.

Abolfazl ShirkavandImage preview unavailable