Half of federal employment law does not reach a 20 person company. The half that does starts at your first employee. The state rules start over every time you hire somewhere new. We build the systems that tell you which rules apply to you, and keep your records where they belong.
Show Me What This Looks LikeAlmost everything written about HR assumes federal law applies to you. At your size, most of it does not apply yet. What does apply depends on your headcount and on where your people sit.
| What It Covers | Starts At |
|---|---|
| Work eligibility records, wage records, background check rules | 1 employee |
| Some state discrimination rules, including new AI notice duties | 1 employee |
| Other state discrimination rules | 5 employees |
| Core federal discrimination protections | 15 employees |
| Age discrimination, health coverage continuation | 20 employees |
| Federal family and medical leave | 50 employees |
Two firms the same size can live under completely different rules, depending on where their people sit. Neither founder was told.
The line also keeps moving. In a single day this year, fifteen states changed their employment rules. One state's paid sick leave requirement dropped from 25 employees down to 11. Companies got pulled into coverage without hiring anyone.
You do not have a paperwork problem. You have a nobody-ever-wrote-down-which-rules-apply-to-you problem. That is a filtering job, and filtering is what these systems are actually good at.
State law change count from Ogletree Deakins' 2026 employer compliance watchlist.
One page per state you employ in, every line linked to the agency that wrote it.
Most of the risk at your size is not that you broke a rule. It is that nobody told you the rule existed.
We start with a simple matrix: each person, their work state, hire date, and status. From that we generate a brief per state. Final paycheck deadlines. Whether accrued time off has to be paid out. Sick leave accrual. Paid leave contributions. Pay transparency posting. Which continuation coverage applies at your size. Every line carries a citation and the date it was last checked.
Then a monthly job watches the public compliance trackers and flags only the states you are actually in. The filter is the whole point. A general alert about employment law is noise. An alert about the three states you hire in is a five minute read.
The same twelve questions, answered from your actual documents, with the source attached.
How much time off do I have. Can I carry it over. What happens with parental leave. Do we cover the home office. Someone asks, and the founder answers from memory.
We build an assistant over your handbook and benefits documents that lives in Slack, where people already are. Every answer quotes the passage it came from and links to the source. Nobody has to take it on trust.
Three things it refuses to answer by design. Leave entitlement. Accommodation or discipline. Anything about another named person. Those route to a human every time. That refusal list is the deliverable, more than the assistant is.
The second output is the one clients end up caring about. It logs every question asked, which tells you exactly which twelve things your handbook should have said in the first place.
The best predictor of job performance is also the one small companies skip.
Three people interview a candidate. Everyone asks whatever comes to mind. The decision gets made in a room where the strongest opinion wins. A year later nobody can reconstruct why the other person was passed over.
We turn a job description and a twenty minute conversation into a working kit. Four to six job-related competencies. Two behavioural questions each. A rating scale written as things you can actually observe. Each interviewer gets assigned competencies, so the same ground gets covered twice and not four times. Scorecards go in before the debrief.
Where you record interviews with consent, the system pulls what the candidate actually said. It lines each quote up against a competency. Then it flags where nobody got any evidence at all. That last flag is what round two is for.
One rule holds the whole thing together. The system produces evidence. The person assigns the rating.
Structured interviews carry a corrected operational validity of .42 for predicting job performance, ahead of job knowledge tests at .40 and work samples at .33. Sackett, Zhang, Berry & Lievens, Journal of Applied Psychology, 2022, peer-reviewed. The validity comes from the structure. AI just makes the structure cheap enough to bother with.
If a rejected candidate's lawyer asked you to describe your process, could you?
Most companies this size already have AI in their hiring and have never written it down. The applicant tracking system has scoring built in. Somebody pastes resumes into a chatbot. A note-taker sits in on interviews.
We inventory every tool that touches an applicant. What it does, whether it produces a score or a ranking, and which states it reaches. Then we classify each one against the definitions that govern it. Those definitions are not the same in every state. Out of that comes your applicant notice, your record retention rule, and five questions to send every vendor.
Some states now require you to tell applicants when AI is involved in a hiring decision. One of them starts at your very first employee. They also hold you responsible for what the tool does, whether or not you built it.
Medical, immigration and background check records each have to be kept apart. At your size they are usually in one folder.
Nobody set out to do this wrong. There is a Drive folder per person and everything went in it. The accommodation conversation stayed in the founder's inbox where it happened.
We sort what you already have into the categories the law treats differently. You get a move list and a set of permission changes. We calculate the retention clock per person, so you know what has to be kept and for how long. Then we set the intake rule, so the mess does not rebuild itself.
One design rule we hold hard. Wherever the sorting can be done from structure rather than contents, it is. The contents are the material that is supposed to stay restricted.
Most of what we build is specific to how you run.
None of the above may be your bottleneck. That is normal. Every engagement starts by mapping where the week disappears, then building against the two or three places that keep costing you.
Where your HR time actually goes, and which rules actually reach you. No tooling decisions yet.
The highest-leverage two from the map, working in your stack, inside the engagement.
Documentation, templates, and the written rules for what the system is allowed to do. Your team runs it without us.
Async by default. Written updates, recorded walkthroughs, working systems you can inspect. We take a small number of clients at a time, which is why the work goes deep rather than wide.
A week recovered, per recruiter. A six-step automated recruiting intake workflow built for a high-volume intake operation.
A week of data entry eliminated. CRM auto-enrichment and contact scoring from records that fill themselves from public signals.
Both are intake and records work, the same shape as most of this page. We do not yet have a published result from a compliance or policy build. When we do, it goes here with a real number.
Tell us which states your people work in and roughly how many of you there are. We will send back a short breakdown of which rules reach a company your size, what a coverage map would take to build, and what it would not fix.
Send Me The Breakdown