The Growing-or-Shrinking Occupation Standard, and What Fails It

Before you retrain for an occupation or point your whole search at it, you need to know whether its market is healthy enough to be worth the months you are about to sink in. This guide gives you a fixed pass/hold/fail standard, five measurable bars built from free public labor data, that two people would grade the same way. It is for anyone asking "is my career dying" or "is it worth retraining for a new career" who wants a verdict they can defend, not a listicle.

Most market guides either rank a shortlist of pivots or count live openings in one metro. This one does something different: it hands you a definition of done for a single occupation, so you can grade the exact job you are considering and say why it passed or failed. The hardest part, and the reason ordinary advice gets it wrong, is that a declining occupation still shows plenty of openings. That trap has a section of its own.

What "healthy enough to commit to" actually means

An occupation is healthy enough to commit to when it clears five bars: projected growth at or above the all-occupations rate, real annual openings, replacement demand you understand the source of, a rising or flat real wage, and no numeric automation or offshoring decline. Grade each bar, then aggregate. A single fail on growth or wage sinks the whole occupation.

The five bars, and what each one is for:

Each bar earns its place because each catches a different failure. Growth catches decline. Openings and replacement demand catch the trap where a dying occupation looks busy. Wage trend catches an occupation that is technically growing but paying less every year in real terms. Automation exposure catches the slow single-digit erosion that clickbait either misses or exaggerates.

Bar 1: does it grow at or above the all-occupations rate?

Pass if the occupation's projected percent employment change is at or above the current all-occupations rate. For the 2024-34 projections cycle that rate is 3.1 percent, the pace at which total US employment is projected to grow as the economy adds 5.2 million jobs and reaches 175.2 million. Below the rate is a hold; a projected decline of 1 percent or more is a fail.

The Bureau of Labor Statistics publishes growth adjectives with fixed cut points. I map those directly onto grades:

The adjective bands are BLS categories. The pass/hold/fail lines are my policy, chosen so that "as fast as average" and up passes because it clears the 3.1 percent economy-wide rate, and anything the data calls a decline fails.

One trap here is silent. The all-occupations rate is a moving goalpost. It fell from 4.0 percent in the 2023-33 release to 3.1 percent in the 2024-34 release. An occupation graded "as fast as average" one year can be re-graded the next without its own number changing at all, purely because the comparison line moved. Always cite the release year you graded against.

Bar 2 and 3: openings without the trap

Record the average annual openings, then ask where they come from. This is the single most misread number in labor data. Total openings equal new jobs from growth plus separations, and separations are workers leaving the occupation, either exiting the labor force to retire or transferring to a different occupation. Every occupation, even one projected to decline, generates some openings from separations.

For many occupations the openings from replacing separating workers are much larger than the openings from growth. That means a shrinking occupation can advertise a big annual openings number that comes almost entirely from retirements. Read percent change first, openings second.

O*NET calls 75,000 or more openings over 2024-34 "large numbers of openings," a useful anchor. But large openings on a declining occupation is a replacement signal, not a demand signal. As a worked example, industrial engineers show 11 percent growth over 2024-34 with about 25,200 openings a year, many from replacing workers who transfer or exit. That occupation passes because the growth is positive and the openings are real, not because the openings number is big.

Bar 4: is the real wage rising, flat, or falling?

Grade the wage bar by comparing the OEWS median wage across the last three annual releases, inflation-adjusted. Rising real median passes, flat holds, falling fails. This bar exists because a growing occupation with a shrinking paycheck is not a good place to spend years.

The per-occupation wage line comes from the Occupational Employment and Wage Statistics program, which produces employment and wage estimates annually for roughly 830 occupations. The median wage is the 50th percentile, the boundary between the highest-paid and lowest-paid halves of workers in that occupation. The current national anchor: the May 2025 all-occupations mean annual wage was $69,770 and the median hourly wage was $24.51.

There is no publicly established BLS "wage-decline red flag" threshold, so I will not pretend there is one. OEWS is a point-in-time estimate. A trend has to be built by hand across releases. My defensible rule, which is my choice and not a BLS ruling, is a flat or falling real median across three or more consecutive annual releases. Fewer than three releases is a snapshot, not a trend, and grading a snapshot as a trend is a documented failure mode.

To check this bar you pull the median for the same SOC code from three consecutive annual OEWS releases, deflate each to a common year using a standard inflation index, and look at the direction. If your target is metro-specific, pull the state or MSA table, not just the national figure, because a nationally flat wage can be falling where you live.

Bar 5: automation, offshoring, and e-commerce exposure

Flag automation exposure only when the OOH narrative or a BLS case study names a downward driver attached to a numeric projection. "AI will kill this job" is not evidence. A projected decline of a specific percentage is.

The important insight is that automation shows up as single-digit erosion, not collapse. The headline "dying jobs" are extreme outliers. The economically large AI-exposed roles decline modestly, and a fixed -1 percent fail line catches them while clickbait either misses the modest ones or invents catastrophes.

Notice the range. Typists at -38 percent are the clickbait headline. Customer service representatives at -5.0 percent and network administrators at -2.6 percent are the quiet ones, and they employ far more people. All six fail the growth bar, which is the point: the numeric line does the work that the narrative alone cannot.

Sector context matters too. Retail trade is projected to lose the most jobs of any sector as automation, consolidation, and e-commerce reshape retail sales occupations, while transportation and warehousing grow about 3.0 percent as online purchases move volume. The gasoline stations industry is projected to decline 10.0 percent, a loss of 98,400 jobs. If your target occupation concentrates in a declining sector, that is a driver worth naming even when the occupation-level number is milder.

The numeric line does the work the narrative cannot: -5.0 percent is a fail whether or not a headline noticed.

The six-step grading procedure

Run these steps in order. The whole grade takes under an hour once you have the SOC code. Each step has a clear "done" so you know when to move on.

A note on order. BLS's own guidance says employment size is a useful starting point, because large occupations usually have more openings than small ones regardless of growth, so some practitioners rank size first. This standard puts growth first and uses size only to interpret openings. Direction is the decision; size is context.

Occupation and SOC code: ____________
Release year graded against: ______ (all-occ rate: ____%)

Bar 1  Growth vs rate:     ___% change  ->  PASS / HOLD / FAIL
Bar 2  Annual openings:    _______ /yr
Bar 3  Openings source:    GROWTH-led / SEPARATION-led
Bar 4  Real wage (3 rel.): rising / flat / falling  ->  PASS / HOLD / FAIL
Bar 5  Automation driver:  ________________  ->  exposed YES / NO

Aggregate rule: any FAIL on Bar 1 or Bar 4 = FAIL; two HOLDs = HOLD.
VERDICT: PASS / HOLD / FAIL
One-line defense: ______________________________

How this goes wrong: failure modes and false positives

Most bad verdicts come from one of seven predictable errors. This is the most valuable part of the standard, because a grade you cannot trust is worse than no grade. Each entry names the trap, what the false result looks like, and the check that catches it.

Two of these deserve extra weight.

The openings trap is arithmetic, not opinion. Because total openings equal growth plus separations, and separations dominate in many occupations, a declining occupation can honestly report large annual openings. This is not a judgment call. Read the percent change, and if it is negative, the openings are people leaving a sinking occupation.

Supply stock is not demand. This is where headcount data misleads. In Refolk's index of professional profiles, there are 73,458 US bookkeepers against 4,872 in the UK, a 15.1x gap far larger than the population difference. That tells you competition, not future hiring. Stock size is the crowd you would compete against, never the demand you would enter.

The stock-versus-flow point holds even for growing occupations. In Refolk's index, growing health roles (nurse practitioners plus physician assistants) number 130,291 profiles against 32,324 in declining data-entry and typist titles, a ratio of only 4.0x. Even a fast-growing occupation does not dwarf a shrinking one in present headcount. That is exactly why you grade on projected direction, not on how many people hold the title today.

Reading the verdict and deciding your move

A pass means the occupation is worth the months. A hold means stay if you are in it, but do not retrain toward it. A fail means find an escape route out, and grade the destination before you jump. The verdict is not just direction; it points to different actions.

Verdicts expire. Employment projections are published every year covering a rolling ten-year period, and OEWS updates annually. Recent projections releases have landed in late August. Any grade card you keep should carry the release year, and any occupation you are serious about should be regraded when the new release drops, because both the occupation's number and the all-occupations pass line can move.

Once you have a passing occupation, the next question is whether people are actually being hired into it near you, and whether workers are successfully moving into it from where you stand now. That is a demand-and-path question, not a projections question. Refolk can surface both from public profile histories: who has entered the occupation recently, and who transferred in from a role like yours.

If you are the one deciding where to aim, Refolk can also read your own history and tell you which passing occupations your existing skills already reach, which shortens the retraining you would otherwise sink months into. And when you apply, Refolk tailors the resume and scores your fit against each posting, so the direction you chose from this standard turns into applications that land.

Verify before you call it graded

Run this checklist before you trust a verdict enough to act on it. If any item is unchecked, the grade is not defensible yet.

The discipline that makes this a standard rather than an opinion is that every bar has a source, a cut line, and a check for when it lies. Two people who follow the grade card and the aggregation rule will reach the same label on the same occupation. That is the whole point: a verdict you can hand to a partner, a mentor, or a lender and defend line by line, before you spend a single month retraining for the wrong thing.

Keep the grade card. Regrade every August when the new projections land. An occupation that passed on the 4.0 percent line may only hold on the 3.1 percent line, and that shift, invisible to anyone reading a stale page, is exactly the kind of thing this standard is built to catch.