Bottleneck Analysis (2026): How to Find the Step, or the Person, That Limits Your Process

TL;DR

Bottleneck analysis finds the step, resource or person that most limits how much work a process completes. Goldratt's theory of constraints gives five steps: identify the constraint, exploit it, subordinate other work to it, elevate it and repeat. Find it by comparing observed cycle times and queues with the process map, because published capacity hides downtime, rework and approval delays. In knowledge work the constraint is often a person whose queue sits in email, chat and meeting follow-ups rather than in any system. This guide is published by Sentra, which understands how a company really runs, and acts on it before anyone asks. Sentra joins ERP, CRM and HRIS records with those conversations to show where work is stuck and who is holding it.

What a bottleneck is

A bottleneck is the step, resource or activity that limits how much work a process can finish. In a sequence of dependent steps, the constraint sets the pace for the whole. Improving another step may move work toward the constraint faster, but it will not increase completed output while the constraint stays the same.

You need observed output to find it. Nameplate capacity is what a machine or person could handle under ideal conditions. Effective capacity is what they handle after downtime, staffing gaps, changeovers and rework. A step with a high published capacity can still be the constraint if its effective capacity falls below demand.

The theory of constraints: Goldratt's five focusing steps

Eliyahu Goldratt's theory of constraints, introduced in his 1984 book The Goal, focuses improvement on the step that limits total output. The five focusing steps give a repeatable way to manage the constraint and check whether it has moved.

• Identify the constraint. Measure where work waits and which step limits completed output. That step sets the pace, or drum, for the process.
• Exploit the constraint. Make better use of its current capacity before adding resources. Keep a small buffer of ready work in front of it so missing inputs never leave it idle.
• Subordinate everything else. Release new work at the rate the constraint can handle. This release rule, the rope, stops upstream steps from building a larger queue.
• Elevate the constraint. Add capacity or change the process if better scheduling cannot meet demand, for example by training another reviewer or removing an approval requirement.
• Repeat. Once output improves, measure the flow again. Another step may now be the limit.

Signals that a bottleneck exists

• Work repeatedly piles up before the same step or approval.
• One person or resource stays busy while others have capacity but cannot move work forward.
• You regularly expedite or use overtime to clear the same queue.
• Downstream colleagues wait on the same function before they can start.

Look for waiting that recurs across several cycles, not one delayed request. High utilization alone does not prove you have found the constraint. Compare where work waits with how much each step actually completes.

Three methods to find the bottleneck

Cycle time and queue analysis

Timestamp each step to separate active work from waiting, then compare arrivals with completions. Compare observed processing time with takt time, the pace needed to meet demand. Takt time is available working time divided by units required. With 40 working hours and 80 requests, you need one completed request every 30 minutes.

A queue that keeps growing before one step means work arrives faster than that step finishes it. Little's law relates average work in process to throughput and average time in the system, but it describes a stable process, so do not use it to predict waiting time for a queue that is still growing. Check actual completion rates before naming the constraint. A busy approver may have a full inbox only because another step releases requests in batches.

Value stream mapping

Value stream mapping traces a real request through every handoff and records active time and waiting time at each step. A map shows where to measure, but the longest wait is not automatically the throughput constraint. Compare observed capacity with demand and check whether the queue keeps growing. A map built only from system events misses the queue at one person's desk.

Process mining, and FMEA as a check

Process mining uses timestamps in business systems to reconstruct the routes work actually takes, then compares them with the intended process to find repeated delays, rework and approval loops. Check the underlying records before naming a bottleneck, because an approval handled entirely in email leaves no trace in the event log.

Failure mode and effects analysis (FMEA) answers a different question: which risks to address first. You score each potential failure for severity, occurrence and detection and multiply them into a risk priority number, so a failure scored 8, 5 and 4 gets 160 under ASQ's FMEA method. Treat that number as a risk ranking, not evidence of a throughput constraint.

Worked example: finding the real constraint

In this illustrative manufacturing case, leaders blame late shipments on machining and plan to buy another CNC center. A review of completed work and queues finds work waiting before heat treatment instead, where changeovers and batching limit output. The manufacturer sequences work around heat treatment and schedules maintenance to reduce interruptions, then measures output and lead time before deciding whether the CNC purchase is still needed. Observed queues, not assumptions, guide the capacity decision.

The bottleneck process maps cannot see

In knowledge work, the bottleneck often sits at one person's desk while the process map shows a routine approval. A finance director may have twenty requests awaiting review, with the queue in email and meeting follow-ups rather than in the finance system. An ERP might mark a purchase request pending approval while a chat thread shows the director asked for a revised forecast and never got it. The recorded wait looks like an approval delay, but the missing forecast is what holds the work.

To find who holds stuck work, trace delayed requests through the conversations as well as the recorded steps, and check whether the named approver can act now or is waiting on someone else. The person everyone routes through is frequently missing from the official version of the process, a gap covered in our guide to the informal organization. Missed early payment discounts are often the first measurable cost of an approval bottleneck.

How to fix a bottleneck once you find it

• Exploit the constraint. If a finance director holds up purchase approvals, find which decisions need their judgment and delegate routine approvals within agreed limits. Remove duplicate reviews and require complete requests.
• Subordinate incoming work. Ask requesters for complete packets, reserve review time on the director's calendar and limit urgent interruptions to defined exceptions.
• Elevate only where evidence points. If the queue still grows, train another approver, change an approval threshold or add staff.
• Measure and repeat. Track the queue and completion time after each change. Once approvals move faster, the next delay may appear in purchasing or legal review.

How Sentra surfaces stuck work and who is holding it

Sentra understands how a company really runs, and acts on it before anyone asks. It reads email, chat, meetings and documents alongside ERP, CRM, HRIS and BI data, works out what is true now, why work is stuck and who is holding it, and raises it to the person who can act. When an ERP record shows a purchase request awaiting approval and an email thread shows the approver asked for a revised quote, Sentra connects the two and names the person the work is actually waiting on. Our guide to a company brain for operations covers the wider pattern.

A named person approves anything Sentra raises before it reaches anyone else. Every claim cites its source, and permissions follow the source. Sentra replaces none of the company's systems. At Lenskart, follow-ups needed at 10 department heads' desks fell 58% within four weeks, in Sentra's own measurement as of October 2026.

FAQ