AI for spotting bottlenecks in your process without a consultant

6 days ago
3 min read

Updated: 4 days ago

Introduction

Ask where the delay in a process is and you will be told about the step that takes longest. That is almost never the answer. In most business processes, the working time is a small fraction of the elapsed time, and the rest is spent in queues: waiting for approval, for information, for the person who does that, for the next batch to be assembled.

This distinction is the whole insight. A process taking eleven days may contain six hours of work. Making those six hours faster changes almost nothing; removing three of the queues halves the elapsed time. Finding the queues requires timestamps rather than opinions, and that is why the analysis is worth doing properly.

1. AI for spotting bottlenecks in your process means measuring the waiting

Elapsed time, not effort.

For each step, record when it became available and when it was completed. The gap between those two, summed across a process, is your waiting time, and it is normally the overwhelming majority of the total.

2. Use the timestamps you already have

They are more available than people expect.

Order dates, status changes, email times, invoice dates, job card entries, system logs. Reconstructing an elapsed-time picture from existing records is usually possible without instrumenting anything new.

3. Look for work in progress piling up

The visible symptom.

Whichever step has a queue in front of it is the constraint. In a physical operation you can see this; in an office process you have to count items sitting at each stage, which nobody does.

4. Expect the bottleneck to be an approval or a person

Rather than a machine.

One person who signs off, prices, checks or authorises is the most common constraint in small businesses. The queue forms at their desk and is invisible because they are always busy, which reads as diligence.

5. Distinguish a bottleneck from a temporary backlog

Different problems.

A step that is always behind is a capacity constraint. A step that is behind this month because of a spike needs catching up, not redesigning. Confusing the two leads to hiring for a problem that was temporary.

6. Fix by removing the queue, not by working faster

The order of remedies.

Reduce batch sizes, remove an unnecessary approval, delegate authority, run two steps in parallel, or set a rule that handles the routine cases without review. All of these attack elapsed time; working faster does not.

7. Beware of optimising a non-constraint

The classic error.

Improving a step that already has spare capacity produces no gain at all and frequently increases work in progress in front of the real constraint. Every improvement should be tested against whether it touches the bottleneck.

8. Watch the bottleneck move

It always does.

Relieve one constraint and another appears immediately downstream. This is expected and it means the analysis is a repeated exercise, not a project with an end.

9. Report total elapsed time as the headline

The number that reflects the customer's experience.

Days from enquiry to quote, from order to delivery, from job complete to invoice. These are what customers feel and what cash flow depends on, and they are the right measures to improve.

Be careful about removing approvals that exist for a reason. Some checks prevent expensive errors or are required by regulation, and the correct fix there is to make the check faster rather than to delete it.

Conclusion

Measure the waiting between steps, because that is where nearly all elapsed time sits.

Reconstruct the picture from timestamps you already hold, look for work piling up in front of a step, expect the constraint to be an approval or a specific person rather than equipment, distinguish a permanent capacity constraint from a temporary backlog, remove queues by reducing batch sizes and delegating authority rather than by working faster, check every proposed improvement against whether it touches the real bottleneck, expect the constraint to relocate once relieved, and report total elapsed time as the headline measure.