The Competitor Pay-Band Teardown: Posted Ranges to a Graded Band Table

This guide shows you how to tell comp and hiring leads what one named competitor pays, by level and location, for the roles you both hire for. It uses nothing except that competitor's own job postings. It's written for talent-intelligence analysts, strategy and research teams, and operators sizing a market. The output is a band table in which every cell has a minimum, an implied midpoint, a likely-offer zone, a posting count and a confidence grade. Every number traces back to a dated posting.

What a competitor pay-band table contains

A competitor pay-band table has one row per level and location tier. Each row has five numbers and a stated adjustment that explains how the posted range became a likely offer. It answers the question the comp lead is really asking, "what would they offer this person, here?", instead of a blended market median that hides the rival's strategy.

Terms used throughout:

• Range width: (max minus min) divided by midpoint. This is the measure used in the Minneapolis Fed and NBER research, and every width in this guide uses it.
• Hire range: a range for one specific opening, usually flagged with words like "targeted" or "expected starting".
• Whole-grade range: the full span, minimum to maximum, of the pay grade the role sits in. New hires rarely land in the top half of it.
• Multi-level range: one posted range that covers two or more levels, often because a job board merged them.
• Multi-zone range: a posting with separate ranges for each geographic pay zone, or one range adjusted by a zone multiplier the employer doesn't publish.
• Implied midpoint: the arithmetic middle of a decoded range, (min plus max) divided by two.
• Likely-offer zone: where offers most probably land. It's the published targeted range if the competitor gives one. If not, it runs from the minimum to the implied midpoint.
• Confidence grade: A, B or C, set by the rubric in the procedure section, so the reader can see how much weight each cell can bear.

Competitor:
Role families:
Window: first capture date to last capture date
Currency and basis: annual base salary, hourly x 2,080

Level | Location tier | Min | Implied mid | Likely-offer zone | Postings | Range type | Adjustment used | Confidence | Source URLs + capture dates
L? | Tier ? | | | | | hire / whole-grade / multi-level / multi-zone | targeted range / min-to-mid / split at sub-range | A / B / C |

How accurate are posted salary ranges?

Posted ranges are honest about the envelope and optimistic about the offer. In Glassdoor's pay-range accuracy study, over 60% of salaries fell below the midpoint of the posted range. For 22% of listings, the pay reported by people in those jobs fell below the range entirely.

That study has a caveat. It compares postings with what current employees report, not with what new hires were offered, and it covers a single six-month window of listings. It's still the best public evidence on direction, and the direction is consistent: the top half of a posted range is where tenure and raises sit, not where offers start. That's why I default to a likely-offer zone that runs from the minimum to the midpoint.

Four more facts affect how far you can trust a competitor's postings:

• Most postings carry no pay at all. In the NBER working paper, only 13.5% of all job postings include any type of wage information. More than 57% of the ones that do give a range rather than a single figure.
• The rivals you care about disclose least. The same research finds that postings for high-wage occupations and at high-wage firms are less likely to give any wage information. When they do, they're more likely to use ranges, and the ranges are wider. So the adjustment from posted range to offer matters most for exactly the roles you're pricing.
• The legal meaning of a range can change under you. California's SB 642 redefined "pay scale" as "a good faith estimate of the salary or hourly wage range that the employer reasonably expects to pay for the position upon hire", effective January 1, 2026. Before that, the definition covered the range the employer expected to pay for the position generally. A California range posted after that date is legally closer to a hire range. One posted before it could legitimately be the whole grade.
• Ranges are base pay only. Postings exclude RSUs and sign-on bonuses, so the band table is a base-salary table and should be labelled that way.

Where will a competitor's postings carry a range?

A competitor's postings reliably carry a range only where local law requires the range in the advert itself. Everywhere else, disclosure is voluntary and patchy. Before you harvest anything, tag every location the competitor hires in as posting-law, on-request or no-law. That way, thin data reads as a coverage gap rather than as no hiring.

US jurisdictions

Posting-law (the range must appear in the advert):

• California, Colorado, Hawaii, Illinois, Maine, Maryland, Massachusetts, Minnesota, New Jersey, New York, Vermont, Virginia and Washington
• The District of Columbia and New York City

On-request (the range is disclosed if a candidate asks):

• Connecticut, Nevada and Rhode Island

Sources disagree on Connecticut: one dates a rule requiring the range in the advert to October 1, 2026. Treat Connecticut as unresolved and check the statute before you tag it. Colorado was the first US jurisdiction to require ranges in adverts, in January 2021, so Colorado postings give the longest history to work from.

Europe and elsewhere

Directive (EU) 2023/970 lets employers give the range either in the posting or before the job interview. A fully compliant EU employer can therefore post adverts with no range at all. Only 4 of 27 EU countries met the transposition deadline, so check each country's status before assuming any obligation applies. The Directive also preserves "bargaining power to negotiate a salary even outside the indicated range", which makes EU ranges weaker evidence of offers.

Voluntary coverage varies a lot by country. In Indeed Hiring Lab's tracker, the UK leads Europe, and Germany sits last of the eight countries covered at around 12 to 14% of postings. Japan is near the top at around 97%. Canada reached 57%, with Ontario at 71%. These readings move from month to month, so treat them as the mechanism and re-check the current values before citing any of them.

Why German reconstruction is thin twice over

Germany has few postings with pay, and few people who could help you corroborate them. In Refolk's index of professional profiles, 5,018 people in the US hold Compensation Analyst, Senior Compensation Analyst, Total Rewards Analyst or Compensation Partner titles, against 14 in Germany.

The German count matches English titles only, so it's an undercount. Two conclusions still hold. In Germany, plan on on-request channels and local-language searching rather than posted ranges, and grade German cells C unless you have several postings. The talent-intelligence row says something about your audience too. In the index, comp specialists outnumber talent-intelligence analysts by about 201 to 1, so the person asking "how do you know they pay that?" usually knows bands better than you do. Build the table for that reader.

How to read range width and text cues

Width tells you how much of a pay structure a range spans. Text cues tell you what kind of range it is. Never classify on width alone. Calibrate width on postings where the competitor has labelled the range itself, then apply that calibration to the unlabelled ones.

The clearest evidence comes from employers that publish their whole structure in one posting. The Federal Reserve Bank of Minneapolis does this. Its Cash Operations Specialist posting reads "Salary range is $42,700-$53,405-$64,100 (For Level II: $46,900- $58,565- $70,300). (Targeted Range: $45,000-$55,000)". Its Assistant General Auditor posting gives a grade of $158,500 to $286,550 and an expected starting range of $175,000 to $200,000.

All widths and ratios in that table are derived from the two postings. They teach three things:

• Single-level grades can be 40% wide. A 40% width doesn't mean a range covers several levels. At this employer, it's simply how wide a grade is.
• The hire zone sits below the grade midpoint, and further below as seniority rises. The targeted midpoint is 93.6% of the grade midpoint for the operations role and 84.3% for the senior auditor role. Senior grades are wider, which leaves more room above the hiring point.
• Job boards collapse levels. Syndicated, the Cash Ops posting shows as one range from 42,700 to 70,300 at 48.8% width. That's two levels shown as one band.

Look for postings like these first. One fully disclosed posting can calibrate dozens of unlabelled ones. The cues to look for:

• Three numbers in a row (min, mid, max): a whole-grade range with the midpoint published. You don't need to infer it.
• "Level I", "Level II", "For Level II:": a multi-level posting with sub-ranges you can split at.
• "Targeted range", "expected starting": a hire range. Use it as the likely-offer zone as it stands.
• Zone names or lists of metros: a multi-zone posting. Decode it before you assign levels.

Width benchmarks and their limits

Nobody has published width thresholds that separate single-level, whole-grade and multi-level ranges. The only threshold source I know of is a practitioner heuristic: 20 to 30% is a one-level band, and anything over 40% is multi-level or padded. It calls itself "guidance from observed posting patterns, not a formal study", and the Fed postings above show where it misfires. So treat width as evidence of how far a range spans, never as proof of what kind of range it is. Width misleads in two ways. Some employers just run wide grades, and syndication merges levels into one envelope.

A posted range describes the grade, not the offer, until the posting itself tells you otherwise.

How competitors encode location in posted ranges

Three location formats are documented: numbered zones published in a separate document, per-zone ranges inside each posting, and zone multipliers the employer never publishes. Work out which one the competitor uses before you assign any posting to a location tier.

• Numbered zones in a separate document. Benchling publishes a US salary zones document. Its Zone 1 is San Francisco, Boston, New York City and Seattle, with a 100-mile radius rule. To decode it, build the zone table from the document and map every posting's location to a zone, applying the radius rule.
• Per-zone ranges inside the posting. Users report that Square/Block lists "salary ranges for each 'zone' (locations)", and that Databricks posts a Zone 3 range of $139,700 to $247,300. That works out to 55.6% wide, and it comes from an unverified forum report. To decode, create one row per zone per posting, and never average zones together.
• Undisclosed zone multipliers. For GitHub, one user reports that "each zone seems to be a 10% either increase or decrease". That's anecdotal. To decode, compare postings for the same title in different metros posted close together, and estimate the step. Grade the result C until at least two pairs agree.

Some limits apply. Nobody has published how common each format is. Two other patterns are plausible, but I can't point to public examples of either. One is a highest-city range, where the most expensive metro's range is shown everywhere. The other is location exclusion, where posting-law states are left off a listing to avoid disclosure. If you suspect either, compare the same requisition across its listed locations and note the suspicion in the adjustment column rather than correcting for it silently.

Remote postings get mapped to whatever remote rule the competitor states, or to the zone the posting names. Done means a location-tier table that every posting maps to.

If you don't yet know which peers in your market run numbered zone structures, the people who build them are easy to find. Their posting habits are often the fastest way to spot the format before you harvest.

The procedure: posting history to graded band table

The method has nine steps. For one competitor and a handful of role families, it takes about six to seven working days. Order matters: classify the range type and decode location before you assign levels, because inferring a level depends on knowing what kind of range you're looking at.

Harvest and normalise

Capture each posting at the moment you find it, because pages hosted on the competitor's applicant tracking system (ATS) often disappear once a role closes. For each posting, record the URL, capture date, title, location, the range text verbatim, the full description, and where you found it (careers page, ATS feed, job board or archive). Whenever both exist, prefer the original posting text over the job-board copy. The board copy is where levels get collapsed.

During normalisation, treat a repost as the same title, location and range text appearing again. Keep the earliest capture date and record the repost count separately. A role reposted five times is one opening, not five pieces of evidence.

Classify, decode and assign

Classification and level assignment happen together, and the competitor's own labelled postings are your calibration set. If it publishes even a few postings with level names or targeted ranges, measure their widths and use those widths, not a generic heuristic, to read the unlabelled ones. List the calibration postings in the source sheet so a reviewer can check them.

For levels, the competitor's title ladder matters as much as its pay data. Its current staff's profiles show which titles really exist: whether there's a "Staff" rung, whether "Senior II" is a real level. In Refolk, you can ask for people at the competitor by role family and read the ladder off the titles that come back, without paging through profiles one at a time. Use years-of-experience requirements as a second signal.

Split a multi-level range only where the posting publishes sub-ranges, as the Fed's "For Level II" text does. If there are no sub-ranges, don't divide the range arbitrarily. Keep it as one row spanning the levels it covers, and grade it C.

Compute and grade

For each level and location tier:

• Min is the lowest decoded minimum among the cell's postings.
• Implied midpoint comes from the decoded range, not from the syndicated envelope.
• Likely-offer zone is the published targeted or expected-start range where one exists. Otherwise it's minimum to midpoint.
• Adjustment used names which of those rules applied, plus any split or zone mapping.

The 93.6% and 84.3% hire-positioning figures come from one employer. Use them to explain why the offer zone sits low, not as a constant to multiply by. If the competitor publishes targeted ranges for some roles, compute its own ratio and state it in the adjustment column.

Grade A
- 3 or more distinct postings (reposts removed) in the cell
- Original posting text captured, not only syndicated board text
- Range type set by an explicit text cue: level name, "targeted", "expected starting"
- No contradiction from matched LCA offered wages, where any exist

Grade B
- 2 distinct postings, or
- Level assigned from width calibrated on the competitor's own labelled postings, or
- Whole-grade range converted to a min-to-midpoint offer zone

Grade C
- 1 posting, or syndicated text only
- Multi-level range left unsplit, or zone step estimated from metro pairs
- Location tagged on-request or no-law

Downgrade one grade if a California series crosses January 1, 2026 without being split.

Cross-check against LCA filings

The Department of Labor's LCA disclosure files include employer name, FEIN, SOC code and title, prevailing wage source and level (I to IV), offered wage and its unit, and worksite location. One public aggregator holds 9,558,753 records covering 167,760 employers and 783 SOC occupations, so coverage of employers that sponsor visas is broad. For each band:

• Match filings on employer FEIN, SOC code, worksite and a date window around your postings.
• Compare the offered wage amount with the band's minimum and likely-offer zone.
• Read the raw DOL files, not an aggregator's display, so you see both the from and to wage fields.
• Record the share of bands that had at least one matched filing.

The typical match rate between LCA filings and posting-derived bands isn't established publicly, and neither is the typical offset. Your own match rate is therefore part of the deliverable, not a footnote. A filed wage well below a band's minimum is worth flagging. A filed wage inside the offer zone is support, not confirmation.

How competitor pay-band reconstruction goes wrong

Most errors come from reading the wrong layer of a range: an envelope read as a band, a grade read as an offer, or a filed floor read as pay. Each failure below has a check you can run before the table leaves your hands.

1. Syndication collapses levels. Boards show the outer envelope, so the Cash Ops posting appears as 42,700 to 70,300 at 48.8% width. It reads as one band, but it's two levels. Check: open the original posting text and look for sub-ranges.
2. Whole-grade ranges read as offers. The auditor grade midpoint is $222,525, but the expected-start midpoint is $187,500. Report the first as "what they pay" and you've overstated the offer for a senior hire. Check: look for "expected starting" or "targeted" language, and treat widths over 40% as probable whole-grade ranges until the text says otherwise.
3. Width alone misclassifies. The Fed's single-level grades are 40% wide, which the common heuristic would flag as multi-level. Check: calibrate thresholds on the competitor's own labelled postings and list them.
4. California ranges compared across January 1, 2026. Before SB 642 took effect, a California range could cover pay for the position generally. After it, the range is a good-faith estimate of pay on hire. A band that seems to narrow or drop across that date may just reflect the new definition. Check: split the series at that date.
5. The LCA wage level read as the internal level. Some workers "are classified at Level I or Level II, even when the salaries are six-figure offers." The prevailing wage level is a filing category, not the competitor's job ladder. Check: match on the wage amount, never on the wage level.
6. The LCA filed wage treated as actual pay. Aggregators use the lower bound of filed ranges, and the filing is a wage floor. Check: download the raw DOL files and read both the from and to wage fields.
7. Thin EU coverage taken as no hiring. An employer that complies with the Directive can disclose the range before interview instead of in the advert. Check: look up the country's transposition status, and grade the cells C rather than leaving the location off the table.

One limit belongs on the face of the deliverable. This method reconstructs the competitor's posted pay architecture and estimates where offers land within it. It doesn't observe offers directly. Say so in the cover note, and point the reader to the confidence column.

Keeping the band table current

Bands move, and a band table is only as good as its most recent dated bracket. Re-harvest on a fixed schedule and date every move as a window between two snapshots, never as a single date.

To date a change, find the last snapshot showing the old range and the first showing the new one, and report the move as "between capture date X and capture date Y". The Internet Archive's CDX server lists the captures it holds for a URL, so you can find the bracketing snapshots without opening each one. Check that it actually covers the competitor's careers domain before you rely on it; I haven't verified its coverage for any particular site. Your own harvest captures are the more reliable record, which is another reason to capture everything at the moment you find it.

Update rules make live postings useful evidence. University of Washington guidance on Washington law says that "if the pay rate or range changes after publishing a job posting, the posting must be updated." So in posting-law locations, a range that changes on a live posting is a signal worth logging. One secondary source lists state update cadences: Washington 180 days, Illinois quarterly, New York City 30 days for major shifts, California annually, Colorado 30 days after a change. Those haven't been checked against the statutes, so don't build anything on them.

I re-harvest monthly for active rivals in posting-law locations and quarterly elsewhere. I re-run the LCA match whenever the DOL publishes a new disclosure file. Each time, re-grade any cell whose posting count changed, and keep the old version of the table so a comp lead can see how the band moved.

When the checklist passes, send the table together with its source sheet. The source sheet is what lets you answer "how do you know they pay that?" in one click rather than one meeting.