The Adjacent-Title Map, One Held Title Cut to a Ranked Target Set

This guide is for anyone whose current job title has become too narrow or is shrinking, and who needs to know which neighboring titles they could credibly apply to next week, and which of those are actually hiring. It takes one real held title, Technical Recruiter, through public occupation data to a ranked set of five to eight adjacent titles. Each title ends up with a skill-overlap score, a level match, a same-day posting count, and a named gap. I've left the wrong turns in, because that's where most homemade lists go bad.

What is an adjacent-title map?

An adjacent-title map is a ranked list of five to eight titles you could apply to next week. Every row records four facts: skill overlap with your current occupation, a level match, a same-day demand count, and the single biggest gap. Any title that fails on overlap, level, or demand gets cut. It doesn't get parked.

Most advice on "what other jobs can I do with my experience" stops at typing your title into ONET and writing down two or three ideas. ONET is the US government's occupational database. It describes each occupation by its tasks, skills, knowledge, and training level. Using it is the right first move, but stopping there leaves you with a brainstorm. That brainstorm has no overlap cutoff, no level check, and no demand step, so you can't defend it to yourself or to anyone else.

Some terms used throughout:

• SOC code: the Standard Occupational Classification code that O*NET attaches to each occupation, such as 13-1071 for Human Resources Specialists.
• Reported job titles: the sample of real-world titles O*NET lists under each code.
• Job Zone: O*NET's measure of how much preparation an occupation needs.

Each of the four fields proves something specific, and each one can mislead you in a specific way:

Where does the worked example start?

The worked example starts by mapping the held title to one O*NET-SOC code and reading the reported job titles under that code. Technical Recruiter maps to 13-1071, Human Resources Specialists. That code also lists Recruiter and Corporate Recruiter, and it sits in Job Zone 4.

I chose this title because it's a common case for a career pivot and because Refolk's index has a large pool of people holding it.

Reading the related-occupations list

O*NET gives every occupation exactly 20 related occupations. It builds them from three inputs: what people in the occupations do, what they know, and what they are called. Experts then review the links and sort them into tiers:

• Primary-Short: the 5 most strongly related codes.
• Primary-Long: the 6th to 10th.
• Supplemental: the 11th to 20th.

This file replaced the older Career Changers and Career Starters matrices. If you find advice that still refers to those matrices, it was written against a retired file.

I'm not reproducing the 20 codes for 13-1071 here. Read them on O*NET OnLine or My Next Move on the day you do this, because the file changes between releases. The one cross-code candidate I carry with full public demand figures is Training and Development Specialists, 13-1151. For the rest of your list, the method is identical and only the numbers change.

Wrong turn: HR Generalist

The first title most recruiters write down is HR Generalist. It's a real option. Refolk's index holds 8,706 people titled HR Generalist in the US. But it's a reported title under the same code, 13-1071, so the related list will never show it. If you score it, you're comparing the 13-1071 skill profile with itself, and the overlap comes out perfect, which tells you nothing.

So the decision at this fork is to keep HR Generalist but move it into a lateral lane: titles inside your own occupation that need different framing rather than different skills. The adjacent-title map proper covers titles under other codes.

Wrong turn: trusting tier order

The next temptation is to take the Primary-Short five as your answer. One published analysis looked at 4,615 Primary-Short links:

• 68.3% stay in the same Job Zone.
• 15.5% point to a higher zone.
• 35.1% cross into a different SOC major group.

Tier order is an expert judgment about how related two occupations are. It isn't a measured overlap score, and it doesn't check level. Pull all 20 codes and score them yourself.

How do you cut candidates on level and skill overlap?

Cut on level first by comparing Job Zones, because that takes minutes and removes candidates before the expensive step. Then compute importance-weighted cosine similarity from O*NET's Transferable and Essential Skills files, and record one overlap number and one named gap per candidate.

Level: comparing Job Zones

O*NET now uses four Job Zones, since zones 1 and 2 were merged. The ones that matter for a Zone 4 starting point:

• Zone 3: one or two years of training.
• Zone 4: a minimum of two to four years of related experience, and usually a bachelor's degree. Human Resources Specialists sit here.
• Zone 5: many occupations need more than five years.

No published standard says how many zones apart two occupations can be before a move counts as a step up or a step down. That means you have to set the rule and write it down, or you'll bend it for whichever title you like most.

For the worked example, every candidate is compared against Zone 4. A Zone 3 candidate gets marked "down" with a note on why you would accept that. A Zone 5 candidate gets marked "up" and has to show the experience years, not the hope.

Overlap: importance-weighted cosine similarity

The Transferable Skills file rates 25 skill elements on two scales: Importance (1 to 5) and Level (0 to 7). Both are commonly rescaled to 0 to 100. Under the analyst procedure, a Level rating is given only when Importance is 2 or higher, so treat a missing Level as zero.

For each occupation, build a 25-element vector where each element is the rescaled Importance multiplied by the rescaled Level. A skill that barely matters to the job then contributes almost nothing. Cosine similarity measures how closely two such vectors point the same way. On these non-negative vectors it runs from 0 to 1.

Check the Date column on every row before you trust a rating. I did not establish when the 13-1071 ratings were last updated, so read that date yourself from the file.

These are the published benchmarks:

None of these is an O*NET-endorsed cutoff, and the scales don't line up exactly. Nesta's thresholds were set for a four-measure average, not for cosine alone. In Nesta's own worked example, a hotel concierge has 51 viable transitions, of which 18 count as safe and desirable. That's a reminder that clearing an overlap bar is the start of the filtering, not the end.

Is skill similarity worth the hour it takes? The evidence says it carries real signal. One skills-similarity recommender predicted real job transitions with 76% accuracy. The Egypt graph study shows what happens without weighting: counting at least three shared activities produced 11,388 pathways, but the average skill transfer across them was only 26.7%. That study had to add a proportional threshold to fix it. Weighting by importance is what keeps a list of titles with merely nice-sounding overlap off your map.

The gap: one skill, named

IRPP gives a usable rule. A gap exists where a skill goes from below-average importance in your current occupation to above-average importance in the target. Find that element in the vectors you already built, then read the target's task statements to confirm it shows up in actual daily work. Write it as one specific phrase. "Instructional design" is a gap. "Upskilling" is not.

How do you measure demand for each candidate title?

Demand is three numbers recorded together. The first is a same-day posting count, pulled narrow and widened on two boards. The second is BLS annual openings from one projection cycle. The third is how many people already hold the target title. No candidate gets ranked without all three.

Same-day posting counts

I did not pull live counts for this guide, and I won't put a number in front of you that you'd be comparing against a different day. You pull these yourself on one calendar day. For each candidate, run a narrow search on the exact title, then a wide search that adds the alternate titles O*NET lists under that code. Do both on the same two boards.

Date pulled: ____
Board A, narrow: "Training and Development Specialist"
Board A, wide: "Training and Development Specialist" OR "Learning and Development Specialist" OR "Training Specialist"
Board B, narrow: "Training and Development Specialist"
Board B, wide: "Training and Development Specialist" OR "Learning and Development Specialist" OR "Training Specialist"
Counts: A-narrow ____ | A-wide ____ | B-narrow ____ | B-wide ____

The gap between the narrow and wide counts is useful in itself. A big jump means employers use many names for this role, so your resume headline has to match the name each posting uses.

Wrong turn: borrowing a snapshot. One state career portal, New Jersey's, lists 581 open HR Specialist jobs and 1,694 Training and Development jobs. It's tempting to use that, because it seems to say training roles are three times as easy to find. But the snapshot is undated and covers one state, so it can't be compared with anything you pull today. I threw it out.

BLS openings, from one cycle

BLS Employment Projections appear on both O*NET and CareerOneStop. For the two codes I have full figures for, here is the current ten-year cycle as CareerOneStop shows it:

HR Specialists are projected to reach 1,002,700 jobs and Training and Development Specialists 501,000. Training and Development ranks 58th of 832 occupations by growth.

Wrong turn: ranking on growth. A 10.8% growth rate looks better than 6%, so the instinct is to rank Training and Development higher. But HR Specialists produce almost twice the annual openings, 81,800 against 43,900, because openings count replacement hiring as well as new jobs. Openings per existing job are close (8.7% against 9.7%). Rank on openings volume, and use growth only to break ties.

Wrong turn: mixing cycles. O*NET OnLine has already moved 13-1151 to the next projection cycle, showing 472,500 employed and 46,000 openings. CareerOneStop still shows 43,900 for the earlier cycle. If you take one candidate from each source, you're comparing different time windows. Choose one source and one cycle for every row.

The holder pool next to the posting count

A posting count tells you how many seats exist. A holder count tells you roughly how many people could compete for them. In Refolk's index:

Take two things from this. First, each market has its own denominator. A count of 100 postings means something very different against 21,285 holders than against 709. Second, the exact-title shares come from only 25 profiles each, so treat them as directional. HR Generalist's lower share suggests the title is written in more ways, which is another reason to run the widened query for it. Employers are spread out in the sample too. No employer accounts for more than three of the 25 sampled Technical Recruiter profiles in either market.

How do you rank the surviving titles?

Rank the survivors on overlap first and demand second, within the same level band, and stop at eight titles. A title with high overlap and no demand is one you can't get paid for. A title with high demand and low overlap is a retraining project, not something you can apply to next week.

Within the top-right quadrant, break ties in this order:

1. Same zone beats one zone down, which beats one zone up.
2. A higher overlap score beats a lower one.
3. More annual openings from the shared cycle beats fewer.
4. Fewer holders per posting beats more.
5. A gap you can show evidence for in your record beats one you can only promise to close.

Record every row in the same format so the list stays comparable as you add titles.

Title | SOC code | Job Zone (mine / target) | Level mark (same / up / down)
Overlap score (cosine) | Cutoff used | Named gap
Posting counts, date ____: A-narrow / A-wide / B-narrow / B-wide
BLS openings a year (source, cycle) | Holders of this title (market)
Decision: rank # / lateral lane / retraining list / cut | Reason:

Once five to eight titles survive, each one needs its own version of your history. A recruiter applying as a Training and Development Specialist has to lead with different lines than one applying as an HR Generalist. Refolk writes the resume from your own record, tailors it to each posting, and scores how well you fit. That gives you an independent check on whether the gap you named is the one the posting actually weighs.

The adjacent-title procedure, step by step

This is the full run in order, with the time each step takes. Plan on about a day of focused work, and keep step 6 on a single calendar day.

A note on order. Nesta computes similarity first and then applies desirability and safety filters. IRPP starts by identifying who and what is at risk, then measures similarity. I put the level check ahead of overlap because it takes 15 minutes and removes candidates before the one-to-two-hour scoring step. That doesn't change the results, only the time you spend getting them.

How does an adjacent-title map go wrong?

Most bad maps fail in one of eight predictable ways, and each one makes a title look closer, easier, or more in demand than it is. Check every candidate against this list before it goes on your final list.

1. Treating a same-code title as adjacent. HR Generalist and Recruiter are both reported titles under 13-1071. The related list will never show the move, and an overlap score comes out inflated because you're comparing the occupation with itself. Check: compare SOC codes before scoring.
2. Trusting tier order as a measure of overlap. The tiers come from expert review, not a published score, and 35.1% of Primary-Short links cross major groups. Check: compute cosine similarity yourself for every candidate, including the top five.
3. Counting shared skills instead of weighting them. Counting alone can show a "viable" pathway that carries only 26.7% skill transfer. Check: use importance-weighted vectors and a proportional cutoff.
4. Missing a level jump. 15.5% of the closest links point to a higher Job Zone. Check: write down both zones and the gap in experience years for every row.
5. Mixing projection cycles. O*NET OnLine shows next-cycle figures for 13-1151 (46,000 openings) while CareerOneStop shows the earlier cycle (43,900). Check: take every candidate's figures from one source and one cycle.
6. Comparing narrow and widened posting counts across days. Counts change daily, so a Monday narrow count and a Thursday wide count measure different markets. Check: pull both versions on the same day, on the same boards, and write down the date.
7. Over-reading small-sample title shares. The exact-title shares in Refolk's index come from 25 profiles each. Check: present them as directional and never rank on them alone.
8. Overlap without demand. Nesta itself notes that similarity alone ignores education, salary, and location. Check: require a demand figure before any title gets a rank.

A title with perfect overlap and no postings this week is a fact about your skills, not a target.

The pattern behind all eight is the same. Each failure replaces something you measured with something that merely looks plausible: a tier for a score, a count for a weight, a snapshot for a pull. When a row on your map has an empty field, the honest entry is "not established", followed by the date you'll fill it in.

How do you keep the map current?

Re-pull posting counts every week on the same weekday and the same boards. Re-read the O*NET files whenever a new release appears. Re-score any candidate whose skill ratings show a newer date than the one you recorded.

The overlap scores and Job Zones are slow-moving, because they change only when O*NET re-rates an occupation. Each release updates only some occupations, and the Date column on every Transferable Skills row tells you whether yours was touched. Posting counts are fast-moving and are the part of the map that goes stale first. If a ranked title's narrow count drops sharply across two weekly pulls, move it to "close but quiet" and promote the next survivor.

When a title moves onto the active list, the work becomes per-posting: a resume variant that leads with the lines the posting weighs, and a cover letter that names the gap honestly. Refolk drafts both from your own history for each posting, so adding a sixth title doesn't mean rewriting your record by hand.