Skip to content
Stephen Kovach
Stephen Kovach

Explore Stephen Kovach

Automate the Bottleneck, Not Just the Easiest Task

Faster software does not automatically create a faster operation. Measure the entire workflow and find where unfinished work accumulates.

Automation projects often begin with the task that is easiest to demonstrate. A document is classified instantly, a report is generated in seconds, or a form is populated without typing. Those improvements may be useful, but they do not establish that the overall service is faster.

A process is finished when the required outcome reaches the person who needs it. Every queue, review, correction, and handoff between the initial request and that outcome belongs in the measurement.

Follow one item from start to finish

Consider a hypothetical intake process with three stages: entry, review, and completion. Entry handles 30 cases per hour, review handles 40, and completion handles 50. If demand is sufficient, entry initially limits the flow.

Automating entry to 90 cases per hour does not make the complete operation process 90. Review can still handle only 40. If 60 cases now arrive each hour, and those rates remain constant, unfinished review work grows by 20 cases per hour.

The calculation is simple, but it changes the investment discussion. The first automation has removed one constraint and exposed another. The next useful question concerns review capacity and quality, rather than another improvement to an already faster entry stage.

Measure waiting as well as working

Amazon’s engineering account explains how a queue can preserve work during interruptions while allowing a backlog to grow faster than it can be processed. Work accepted successfully may still arrive too late to be useful. [1]

For a business workflow, that suggests tracking elapsed time from receipt to completion, not just the seconds spent by the automated component. A five-second task followed by a two-day wait is still a two-day service experience.

Use several views: the number of items waiting, the age of the oldest relevant items, completion time for typical cases, and completion time for difficult cases. An average can look acceptable while a small group of customers waits much longer.

Reduce rework before increasing volume

Suppose review finds that many cases lack a required document. A better intake check might improve the operation more than a faster review screen. Preventing an incomplete submission reduces repeated contact, reassignment, and another pass through the queue.

This is an analytical hypothesis to test. Measure how many cases arrive complete, how many return for correction, and whether the change reduces end-to-end time without creating an unreasonable burden at entry.

A process map should therefore include loops. A diagram that sends every case neatly from left to right can hide the repeated work that consumes much of the available capacity.

Plan for demand that exceeds capacity

Google’s reliability guidance discusses limiting load and handling overload deliberately, rather than allowing excess demand to destabilize a service. The exact software techniques vary, but the business implication is straightforward: accepting more work is not the same as being able to finish it. [2]

An operation may need realistic appointment limits, clear priorities, a controlled backlog, or additional capacity. The appropriate response depends on what the work means and which delays cause the most harm. An urgent case and a routine background task should not automatically receive identical treatment.

Before funding an automation project, state the expected improvement in completed outcomes, identify the next likely constraint, and define the measurement period. Then compare results with the original expectation.

The best automation removes friction from the whole process. A faster isolated task is only the beginning of that assessment.

Sources

  1. Amazon Builders’ Library: Avoiding insurmountable queue backlogs
  2. Google SRE: Handling overload

Research, reflection, and a wider perspective.

Continue exploring

LETTERS FROM STEPHEN

A note from Stephen.

New essays, research, and occasional notes. Choose the subjects that interest you.

Choose your interests
Send me writing about

Leave all unchecked to receive all topics.

Confirm your address by email. Privacy policy

inj[jm