AI Prompts for Bottleneck Analysis

Every process has one constraint that governs its throughput, and effort applied anywhere else is largely wasted. Finding it is harder than it sounds: the busiest-looking station is not always the bottleneck, and bottlenecks move as mix changes. The Theory of Constraints gives the method — identify, exploit, subordinate, elevate, repeat — and Little's Law gives the arithmetic that connects WIP, throughput and cycle time.

These prompts apply both. They need data about the process — utilization, queue lengths, cycle times, throughput at each step — and will tell you what to measure if you do not have it. They work for manufacturing lines, warehouses, order-processing flows and service operations alike.

Before you use these

Have these ready to replace the highlighted [variables]:

• Process steps in sequence with capacity, actual throughput and utilization at each
• Queue or WIP levels before each step, and cycle time through each
• Downtime and changeover data by step
• Demand rate on the process

The prompts

• 1. Identify the constraint from data
• 2. Exploit, subordinate and elevate the constraint
• 3. Analyze WIP and flow with Little's Law

1. Identify the constraint from data

Act as an operations analyst identifying the constraint in [process].

Process: [steps in sequence; for each: design capacity, demonstrated throughput, utilization, WIP or queue before it, cycle time through it, downtime %, changeover time, staffing]
Demand: [rate the process must meet]
Observations: [where work waits, where expediting happens, what supervisors say is the problem]

1. Compute effective capacity per step = design capacity × availability × performance × quality (or from demonstrated throughput). Compare to demand — any step below demand is a constraint candidate.
2. Evidence check per candidate: is WIP accumulating in front of it? Is it starved or blocked (which points to the constraint being elsewhere)? Does throughput downstream track its output?
3. Distinguish the capacity-constrained resource from steps that merely look busy because of poor scheduling or batching.
4. Mix dependence: does the constraint move with product mix? Identify the mix scenarios where a different step binds.
5. Policy constraints: rules (batch sizes, shift patterns, approvals) that constrain throughput without any resource being at capacity.
6. Conclusion: the constraint, the confidence level, and the specific measurement to confirm it if the data is ambiguous (e.g. a time study of queue times over a week).

Do not assume the highest-utilization step is the constraint without queue evidence. If the data suggests the constraint is a policy rather than a resource, say so.

2. Exploit, subordinate and elevate the constraint

You are applying the five focusing steps to [constraint] in [process].

Constraint profile: [capacity, current throughput, time lost to downtime, changeovers, quality, starvation, blocking, breaks; scheduling practice]
Non-constraint steps: [their capacities and current scheduling]
Elevation options: [additional equipment, shifts, outsourcing, automation — with cost and lead time]
Throughput value: [contribution margin per unit of throughput]

1. Exploit: list every way to get more from the constraint without capital — eliminate starvation (buffer in front), run through breaks, offload inspection or setup, reduce changeovers on the constraint specifically, quality gate before the constraint so it never processes defects, priority rules. Estimate the throughput gain per action and its value.
2. Subordinate: how non-constraint steps should be scheduled and measured to serve the constraint — buffer sizing before it, release rules (drum-buffer-rope), removing local efficiency targets that overproduce WIP. State the changes to KPIs required.
3. Elevate: only after exploitation, the capital or structural options, their cost per unit of throughput gained, and payback using the throughput value.
4. Predict where the constraint moves after elevation and what to prepare.
5. Sequence: the first 30 days (exploit), the next 90 (subordinate and measure), then the elevation decision with the evidence it needs.

Quantify every action in throughput and value. Do not recommend elevation if exploitation gains have not been captured.

3. Analyze WIP and flow with Little's Law

Act as a flow analyst applying Little's Law to [process].

Data: [for the whole process and per step where available: average WIP (units), throughput (units per period), measured cycle time if known, value-add (touch) time per unit]
Target: [cycle time or lead time the business needs]
Release practice: [how work is released into the process — push to schedule, batch, on demand]

1. Apply Little's Law (cycle time = WIP ÷ throughput) for the whole process and per step. Reconcile with measured cycle time where available and explain discrepancies (measurement, variability, rework loops).
2. Flow efficiency per step and overall = value-add time ÷ cycle time. Identify where units wait longest and why (batching, queue before constraint, handoffs, approvals).
3. WIP required for the target cycle time at current throughput, and the reduction from current. State what would break if WIP were cut abruptly (constraint starvation) and the safe reduction path.
4. Variability: where arrival or process variability inflates queues, and the levers (leveling, smaller batches, pooled resources).
5. Release rule: a WIP cap or CONWIP rule that holds cycle time at target, and how it interacts with the constraint buffer.
6. Metrics to run weekly: WIP, throughput, cycle time, flow efficiency — with the expected trajectory.

Show the arithmetic. Flag any step where WIP data is estimated rather than measured.

Worked example

Related prompts

• Process Improvement
• Production Planning and Scheduling
• Capacity Planning
• Root Cause Analysis
• Warehouse Operations Improvement
• Operational KPI Reviews

Logical next step

After this, most operations teams move on to Process Improvement.

All Operations & Supply Chain prompts · Search the full library