Building a More Intelligent Workforce Compliance Program

Most workforce compliance programs are built around filings. Data gets assembled once a year, submitted, and set aside until the next cycle. That model worked when there was one federal filing and an agency that investigated before anyone sued.

It does not work now. This guide covers what to build instead — practically, in the order the work actually has to happen.

Stage 1: Establish what applies to you

Start here, because employers are routinely wrong about it. Thresholds differ, and several are ambiguous.

Obligation	Threshold	Timing
Federal EEO-1 Component 1	100+ employees; federal contractors 50+	Annual — rescission proposed, not final
California pay data (Gov. Code §12999)	100+ nationwide, 1+ in California	Second Wednesday of May
California labor contractor report	100+ labor contractor workers	Same deadline, separate report
Illinois EPRC	100+ Illinois employees	Every two years
Massachusetts workforce data	100+ and EEO-1 covered	February 1 annually
Colorado periodic report	100+ workers doing business in CO	From July 1, 2027, rolling anniversary

Two traps. California counts employees nationwide — a company with 400 employees and one person in California files. And Colorado's threshold does not specify whether it means Colorado workers or the whole workforce, so treat it conservatively until the state clarifies.

Stage 2: Fix the data before you analyze it

Nearly every failed analysis fails here. Before any statistics are meaningful, you need four things.

Complete demographic data

Self-identification is voluntary, and unknowns cluster. If 18% of your workforce has no race or ethnicity recorded and the gaps concentrate in one business unit, every subsequent analysis of that unit is compromised. Measure your unknown rate first, and know where it sits.

Decisions, not just headcount

Adverse impact analysis examines selection decisions — applicants and hires, candidates considered and promoted, employees evaluated and separated. A headcount snapshot cannot produce a selection rate. Many HRIS configurations retain the outcome but discard the pool, which is the number you need.

Stable job architecture

If job titles have drifted, been created ad hoc, or mean different things in different divisions, category-level analysis produces noise. This work is also the prerequisite for the California SOC remap, so it pays twice.

Separated storage, in California

Since January 1, 2026, SB 464 requires demographic information collected for pay data reporting to be stored separately from personnel records. If yours lives in the employee file, that is a current violation, independent of filing.

Stage 3: Run both tests, not one

Two calculations matter, and using either alone produces misleading results.

The impact ratio — the four-fifths rule

Divide each group's selection rate by the highest group's selection rate. A ratio below 0.80 is what the EEOC's Uniform Guidelines (29 CFR 1607) treat as evidence of adverse impact. It remains the operative benchmark in litigation regardless of federal enforcement posture.

Its weakness is small samples. In a pool of twelve candidates, one person changing outcome can swing the ratio below 0.80. Treated alone, it generates false alarms in small groups and misses real problems in large ones.

Statistical significance — the two-proportion Z-test

This asks whether the difference is larger than chance would explain. Courts have used standard-deviation analysis since Castaneda v. Partida and Hazelwood School District v. United States, generally treating two to three standard deviations as the threshold where a disparity becomes hard to attribute to chance.

Used together, these answer different questions: the impact ratio asks whether a gap is large, significance testing asks whether it is real. A finding that is both is what warrants investigation.

Correct for multiple comparisons

Testing many groups across many decision points guarantees some will cross the threshold by chance. Run twenty independent tests at a 5% threshold and one false positive is the expected result. A false discovery rate correction such as Benjamini-Hochberg controls for this. Without it you will chase noise and, worse, generate documents describing disparities that were never real.

Stage 4: Decide what is privileged — before you analyze

This is the step most often skipped, and it cannot be fixed retroactively.

An adverse impact analysis creates a document describing statistical disparities in your employment decisions. That document is discoverable unless it is protected. Analysis conducted at the direction of counsel, for the purpose of obtaining legal advice, may fall under attorney-client privilege or work-product protection — but the structure has to be established before the work begins, not after a finding appears.

Have the conversation with your employment counsel now: which analyses run under privilege, who receives them, how findings move into remediation without waiving protection, and what gets documented in the ordinary course. There is no universally correct answer, but there is a clearly wrong one, which is not deciding.

Stage 5: Set a cadence and document the response

Annual review is too slow. A disparity that appears in Q1 and is found the following January has had a year to compound, and a year of decisions were made on the same process that produced it.

Quarterly works for most organizations. What matters more than frequency is that each cycle produces a record:

• What was tested, over what period, using which method
• What crossed threshold, and by how much
• Who reviewed it and when
• What explanation was investigated — and what the investigation found
• What was decided, what changed, and what was deliberately left alone with reasons

That last point matters. A documented decision not to act, with reasoning, is far stronger than silence. Silence looks like you never looked.

Stage 6: Run the SOC remap as a project

For California filers this is the largest single piece of work between now and May 2027, and it deserves its own plan.

1. Export every active job title with incumbent counts and current EEO-1 category.
2. Map to the 23 SOC categories. Expect genuine ambiguity. Management splits into chief executives and everything else; roles grouped today will separate.
3. Flag every judgment call and record the reasoning. When CRD or counsel asks why a title landed where it did, you want the answer written down.
4. Have someone who knows the jobs review it. A compensation analyst catches what a systems export cannot.
5. Run a dry-run filing against the new categories well before the window opens.

Penalties for a failed California filing are now mandatory and calculated per employee, and good faith does not mitigate. The cost of starting early is a few weeks of analyst time. The cost of starting late is measured in tens of thousands of dollars.

Five mistakes worth avoiding

• Treating filing as the program. Submitting a report is an output. Knowing what it says about you is the point.
• Analyzing only what is easy to measure. Hiring data is usually cleanest, so hiring gets analyzed and promotion, pay, and separation do not. Separation patterns frequently carry the largest exposure.
• Single-axis analysis only. Testing race and sex separately can show both within tolerance while a specific combination sits well outside it.
• Finding something and stopping. An unexplained flagged disparity in an undocumented file is worse than no analysis at all.
• Waiting for federal clarity. State obligations do not depend on it, and litigation risk never did.

Where to start this month

If you do only one thing: measure your demographic completeness rate and confirm whether you retain applicant and candidate pools. Those two facts determine whether everything else on this list is even possible, and most organizations do not know the answer.

EquityGuard Intersect™ provides compliance analytics, not legal advice. Regulatory positions described here reflect publicly available information as of September 2026 and are changing quickly. Confirm current obligations with qualified employment counsel before acting.