An AI HR advisor,
available right now.

Almost every employment problem that becomes expensive was manageable earlier and undocumented. The concern that was never written down, the conversation nobody recorded, the policy that existed in one manager's head.

The HR Copilot pushes toward writing things down, in specific and observable terms, at the point where it is still routine rather than defensive. That is unglamorous and it is what actually protects a company - and the employee.

Federal law sets a floor. States add substantially on top - final paycheque timing, notice, non-compete enforceability, leave entitlements - and cities add more. Several federal obligations only attach above headcount thresholds of 15, 20 or 50.

Which means the honest answer to a lot of HR questions starts with which state, and any tool that answers confidently without asking is giving you something worse than no answer.

Harassment, discrimination, retaliation, termination and wage-hour classification. Those five are where an informal approach creates liability rather than resolving anything, and where the cost of getting counsel early is trivial against the alternative.

This will help you understand the landscape and arrive prepared. It will not tell you it is safe to proceed, because it cannot know that.

Unstructured interviews are among the weakest predictors of job performance in the research literature, and they remain the default almost everywhere. The reason is that they feel informative - a conversation produces a strong impression, and strong impressions are convincing regardless of what they are tracking.

What improves prediction is structure. The same questions, asked in the same order, to every candidate, scored against criteria agreed before anyone was interviewed. It is less enjoyable and considerably more accurate, largely because it makes candidates comparable rather than memorable.

Work samples predict better than questions. A short exercise resembling the actual job tells you more than an hour of discussion about how someone would approach it, and it is fairer to candidates who interview badly and work well.

Defining criteria in advance is the step that does most of the work. Without it, panels rationalise a preference after the fact, and the language used to justify a decision - culture fit, not quite the right level - is precisely the language that hides bias from the people expressing it.

None of this requires a hiring function. It requires writing the questions down before the first interview and scoring against them afterwards, which is an hour of preparation that changes the quality of every hire that follows.

Early attrition is expensive and largely preventable, and it is decided in the first fortnight far more than in the first year.

The failures are mundane. Access that does not work on day one. No clear owner for questions. A first task that is either trivially small or impossible. A month of ambiguity about what success looks like. Each is individually minor and collectively communicates that nobody was expecting this person.

What works is unglamorous and cheap. Equipment and access ready before the start date. A named person responsible for the first two weeks who is not the manager. A first-week task that ships something small and real. Written expectations for the first thirty, sixty and ninety days.

The written expectations matter most for the manager. A new hire who does not know what good looks like will invent a standard, and it will not be the one they are measured against - which is how a performance conversation happens in month four about a misunderstanding from week one.

Ask at thirty days what was confusing. It is the cheapest process improvement available, the answers are specific, and the next hire benefits from every one of them.

Worker classification is one of the most expensive mistakes a small company can make, and it is made casually because the practical difference feels administrative.

It is not. Misclassification exposes an employer to back taxes, unpaid overtime, penalties and interest, and both federal and state agencies pursue it. Several states apply a stricter test than the federal one, which means a worker can be a contractor for one purpose and an employee for another.

The tests turn on control rather than on paperwork. Who decides when and how the work is done, who provides the tools, whether the worker offers services to others, whether the relationship is open-ended, and whether the work is central to the business. A signed contract calling someone a contractor carries very little weight against those factors.

The common pattern is a long-term contractor who works set hours, uses company equipment, has no other clients, and does core work. That arrangement is an employment relationship being described differently, and it is the exact fact pattern agencies look for.

The related trap is exempt status. Paying a salary does not make someone exempt from overtime; exemption depends on duties and a salary threshold, and misapplied exemptions generate wage claims with substantial penalties. Get advice before assuming either.

The annual review is widely disliked and the reason is structural: it compresses a year of feedback into one conversation, which guarantees that some of it is late enough to be useless and some is a surprise.

The alternative is not a better form. It is more frequent and smaller conversations, which changes the review from a verdict into a summary of things already said. A manager who has raised something in the month it happened is not delivering news in December.

Recency bias is the second structural problem. A review written in December over-weights November and under-weights March, and it does so invisibly. The correction is notes taken through the year rather than a better memory, and it takes a few minutes a month.

Specificity is the third. Feedback that describes a trait cannot be acted on, because a person cannot become differently-natured by request. Feedback that describes a behaviour, a situation and an effect can, and the difference between those two forms is most of whether a review changes anything.

Ratings deserve scepticism. Forced distributions and numerical scales create a great deal of internal politics for a signal that predicts less than their prominence suggests, and they turn a development conversation into a negotiation about a number.

Whatever the format, write it down and share it. An undocumented conversation protects nobody, and the employee is entitled to know where they stand in a form they can refer back to.