I Stopped Re-Explaining Onboarding. I Built It Once Instead.

What a Claude skill taught us about the first 90 days

Every hotel has a version of the same story. A new sales manager starts on Monday. Someone digs up last year’s checklist, a few people explain things from memory, and by Friday the new hire has met a lot of people and still can’t tell you who approves a proposal.

That’s the gap we found at the resort I’m working with. HR covers policy and the general overview. Department managers teach the job. Nobody owns the part in between: how departments work together, and who signs off on what.

So my partner there and I built an onboarding skill in Claude. Here’s why a skill, and what’s in it.

A prompt is a conversation. A skill is a system.

A prompt only works if someone remembers to write it, and writes it well, every time. A skill is a packaged set of instructions and reference files that Claude loads whenever the task comes up. You build it once. Anyone who runs it gets the same quality.

What’s inside

Reference files, loaded once and owned by one person: mission and values, guest philosophy, property fact sheets, feeder markets, comp set, commercial priorities, org chart and the metrics each role is measured on.

Intake questions: role, job description, resume, start date, manager, buddy, key people and the systems the hire will use.

From those, the skill produces:

• A resume gap analysis: what they already know and what they still need, focused on the job itself
• A welcome letter and a day-by-day first week
• 30/60/90-day phases: Learn, Contribute, Own
• A systems training schedule and manager check-ins
• A multi-stage assessment, built into the plan

One more detail I’d steal for any hotel: sales and marketing hires experience the product. Overnight stays and time in the facilities go into weeks one and two, because you can’t sell what you haven’t felt.

The assessments are the part I’m proudest of

Most onboarding ends with “how’s it going?” This one measures it, in stages built around Kirkpatrick’s four levels.

Stage 1: a knowledge check before the first day. For the digital marketing role, it tested GA4, UTM parameters and ROAS calculations. It isn’t a pass/fail test. It shows what the new hire already knows and what they don’t, so the plan spends training time on the gaps instead of repeating what they’ve mastered. Retaking it at day 90 shows exactly how far they’ve come.

Stage 2: pulse checks. The new hire tells us how onboarding feels at day 30 and day 90, so we hear about problems while they can still be fixed.

Stage 3: open-book knowledge checks, with a separate answer key for the manager. They start at day 30. Real jobs are open-book. We want people to know where critical information lives, not to memorize it.

Stage 4: real work, scored on a four-level rubric: Not yet, Developing, Proficient, Exemplary. For sellers that can be a mock site inspection, a role-play or their first proposal, reviewed by their manager.

Stage 5: results. Is the new hire ramping against their scorecard targets at 30, 60 and 90 days?

Build it once. Use it for any role.

Here’s what makes this worth building. Once the reference files exist, the skill isn’t about one job. A wedding sales manager, a corporate retreat seller, a revenue manager, a marketer, an operations lead: each gets their own plan from the same skill. The inputs change (job description, resume, key people, systems). The structure stays.

The assessment content is what you swap. GA4 and ROAS for a marketer. Pickup, pace and forecasting for revenue management. Site inspections and proposals for sellers. For ops, the property standards and the systems the team lives in. We designed it to handle a junior coordinator with no hotel experience, a seasoned hire from a competitor and an internal promotion with no resume to analyze.

Which means the next hire costs a manager a short intake conversation and a review, not a blank page.

The guardrails matter as much as the build

• A manager approves every plan before it reaches the new hire
• Anything Claude infers is flagged as an inference
• HR and legal items are marked “HR owns this”
• The sales team doesn’t use Claude, so the manager runs the skill and shares the finished document

The first real test was a digital marketing role. The next role went through with very little extra setup.

What we don’t know yet

It’s new. We’ll judge it on time to first booking, ramp against targets, 90-day retention, the new hire’s own feedback and how long it takes a manager to build a plan. I’ll share what we learn.

The lesson that has nothing to do with AI

The skill took less time than the conversation that came before it: what should a new hire know, do and be measured on in their first 90 days? Most hotels have never written that down. The skill just makes it repeatable.

One coaching tip from the build: describe the outcome and the constraints, then let Claude use its judgment. Over-scripting every step makes a worse skill.

How long would it take your newest hire to explain who approves what? And if the answer lives in one manager’s head, what happens when that manager is out?

I cover builds like this every month in Stay Current AI, where we look at what’s changed in AI and what hotel commercial teams can actually use. If that sounds useful, I’d love to have you.