Start with the payoff question
A first-pass capacity estimate helps teams decide whether a bottleneck change deserves deeper investigation before a detailed simulation project. A bottleneck fix is only commercially interesting if it changes the output of the whole workflow. Before building a detailed model, write down the current demand, the main steps and the capacity of each step in the same period. The lowest required step capacity is the first place to test.
Quick estimate or full simulation?
A practical example
Imagine a service team needs to complete 55 cases each week. It is considering three possible improvements:
Triage
64 cases/weekProposed fix: Add more admin time
No extra output while specialist review still caps the flow.
Specialist review
42 cases/weekProposed fix: Reduce review time by 15%
Likely increases weekly output until approval becomes tight.
Approval
58 cases/weekProposed fix: Automate final notes
Useful locally, but not the first payoff lever.
The rough estimate does not prove the final design. It does show which improvement deserves the next hour, workshop or modelling budget.
When a rough model is enough
Use a quick estimate when the decision is still early: should we investigate this constraint, collect better data, prepare a client scoping report, test an automation idea or challenge a hiring request? If the estimate shows no system-level output gain, pause before funding the local improvement.
When deeper simulation is better
Move to full simulation when the workflow has high variability, complex routing, shared resources, batch rules, queues, timing effects or expensive implementation risk. Improve First is the front-end decision aid; it helps decide what deserves that deeper work.
For an example of what this adds, MathWorks explains how queues and servers model waiting and processing. Improve First does not calculate those waiting times.
Compare bottleneck tool choicesHow can I start when the numbers are uncertain?
Begin with one clear type of completed work and one shared period. For example, count finished orders per working day, not emails per hour at one step and orders per week at another. List the main activities in order, estimate the active working time per item, and enter how much working time and how many people or machines are actually available. Keep waiting for another person's reply separate from time spent doing the task.
You do not need to pretend a rough estimate is exact. Write down who supplied it and what it excludes. Use a faster and a slower plausible case, then see whether the same step remains the priority. If the conclusion changes easily, the next useful action may be observing that activity more closely, not buying equipment. A model is most useful when it exposes which uncertain number could change the decision.
What makes a fair before-and-after comparison?
Keep demand, working time and the meaning of a completed item unchanged while testing one improvement. Compare expected completed output and the remaining gap against demand. Check where the limiting step moves after the change. If another step becomes the limit, a second improvement at the original step may add little or no finished output. If demand is already met, more capacity does not by itself generate sales.
A reduction in minutes per item and an increase in capacity are different assumptions; use the field that matches the proposed change. Do not describe a percentage capacity increase as the same percentage time saving. Review the resulting figures before sharing a recommendation. Compare costs, quality, staff workload, safety and implementation risk outside the output calculation rather than presenting extra capacity as guaranteed profit.
When should I investigate more deeply?
Before using a result in a proposal, check for double-counted resources. A person who divides a day between checking and packing does not provide a full day to both. Enter only the time available to each activity. Where the same machine or specialist switches between many queues, a more detailed resource model may be needed. A tidy diagram alone does not remove that conflict.
Repeat work matters too. If items often return for correction, count that extra effort rather than assuming every item passes once. If only some items visit a specialist, check the visit assumption. Keep the meaning of each number visible so another person can follow the calculation. This is especially important when comparing two estimates made by different teams: they may be using different finish lines or counting different work.
Investigate further when work follows different routes, the same person serves several steps, or delivery promises depend on irregular arrivals and long queues. The simple capacity model does not predict those interactions or average waiting times. Different types of work may use each step differently, so a changing mix can change the real limiting activity even when the staffing stays the same.
Take the starting assumptions and the proposed change into that investigation. Ask the team whether the change can actually be achieved and what could prevent it. Use anonymised labels when discussing client work. The useful output is a clear, checkable next decision, not a claim that a first-pass model has diagnosed every cause of delay or proved the outcome of an investment.
Where can I read about the methods?
MIT OpenCourseWare’s operations-management case questions cover process maps, step capacity and bottlenecks. MathWorks’ queue-and-server documentation explains the timing and queue behaviour that a simple capacity estimate does not predict. These sources explain the methods and limits; they do not validate Improve First or the fictional figures on this page.