CALM: Break the Work Into Real Steps

Good workflows are not described — they are mapped, step by step, against what people actually do.

A task analysis makes the invisible structure of work visible: who does what, in what order, under what conditions, and with what information available at each step. Without it, design decisions rest on assumed sequences — flows that look complete in a document but skip the hesitations, lookups, and judgment calls that real users must perform.

Working agreement: before designing or evaluating any workflow, name each discrete user action and the decision or state change it depends on.

The instructions for use (IFU) for a patient monitor once served as the closest thing to a task record. It described what a nurse should do, not what she actually does — so the gap between documented procedure and real use remained invisible until a usability test exposed it. A clinician adjusting an alarm threshold navigates three screens, remembers a unit-specific convention, and performs a silent mental check before confirming — none of which the IFU mentioned.

Today, usability and human factors (HF) teams decompose that workflow into discrete tasks: each action, each prerequisite state, each decision point. That structure becomes the input for design improvements, use-related risk analysis, and test scenario construction. The design team no longer has to guess which step carries the most cognitive load; they can see it.

The same discipline applies anywhere people follow a sequence under pressure: make the work visible before changing the workflow.

Without a task breakdown, teams can improve the interface while weakening the work. A cue may be removed, a step may be reordered, or two decisions may be collapsed before anyone sees the added verification burden.

How teams ship this: Version the task breakdown alongside the design artifact — when a screen changes, the associated task record must be reviewed and updated before the change is approved.

Task Breakdown Record · one entry per user action:

Use this as input for usability test scenario design, risk analysis, and design review. It keeps the task structure traceable when designs change.

Clayton Lewis and John Rieman's Task-Centered User Interface Design remains one of the most direct guides to building task analyses that drive real design decisions rather than documentation artifacts. Freely available online.

Where is your design still making people feel the technology rather than helping the work move?

Choose one small routine you do almost automatically.

Before you start, pause for a few seconds and notice:
What do you do without thinking?
Where do you check, wait, adjust, or repeat?

Do not change the routine yet.
Just make the invisible steps visible.

Artificial intelligence (AI) is beginning to change how teams build task models — but the direction matters more than the speed.

Today, task analysis is largely manual: a researcher observes a workflow, interviews users, reviews documentation, and constructs the breakdown by hand. It is time-consuming and easy to compress or skip, especially under project pressure. AI can reduce that friction by processing session recordings, interview transcripts, and screen captures to extract candidate task steps, flag decision points, and identify where users hesitate, backtrack, or request help. Tools that combine eye-tracking data, interaction logs, or think-aloud transcripts with language models can surface a draft task breakdown faster than any single analyst working alone.

The risk is the same one that appears whenever AI accelerates synthesis: a faster breakdown is not necessarily a more accurate one. Task analysis is not just a list of steps — it is an interpretation of what each step requires cognitively, what information the user relies on, and where failure modes lie. That interpretation requires domain knowledge, risk awareness, and direct observation. An AI model that has never seen an ICU handoff cannot reliably distinguish a routine confirmation step from a high-stakes judgment call that happens to look the same on screen.

The calm AI workflow for task analysis would be: use AI to draft the step list from observation data, then require a human factors engineer or experienced researcher to validate each step, add decision requirements, assign risk levels, and confirm that no hidden steps have been collapsed. AI can compress documentation effort; it cannot replace the judgment that makes task analysis a useful risk input.

In regulated contexts — medical devices, aviation, industrial control — task analysis feeds directly into use-related risk assessments, verification plans, and the design of summative usability studies. At that level of consequence, a task breakdown built entirely on AI output without human review is not just incomplete; it creates traceability gaps that surface during audits, regulatory submissions, and post-market surveillance reviews.

Use AI to see the workflow faster. Use human judgment to understand what it means.

Best,
Andreas Walden

Share this with someone designing for clarity.