Training and onboarding materials · AI for HR & Recruiters

The training and onboarding function in a Nepali SME is, almost universally, under-invested. There is a handbook somewhere, possibly from 2078, that no new hire reads in full. There is a half-day induction on day one that covers benefits and the leave form. There is an implicit assumption that the new person will figure out the rest by asking colleagues, and that the colleagues will be patient. Sometimes this works. Often it produces a new hire who is technically employed but practically lost for the first month, asking the same five questions in a slightly different order to a slightly different person each time. AI does not fix the cultural underinvestment, but it removes the most common excuse for it — the time cost of producing decent materials. What used to be a two-week project for an HR generalist is now a focused afternoon, if you know what to ask the model for.

The high-leverage uses

Five concrete uses, in rough order of return per hour invested, cover most of what a Nepali HR function actually needs from AI in this area.

1. Role-specific summaries from dense source material. The 47-page operations manual contains, somewhere, the 8 pages a new accounts assistant actually needs. AI can extract them in a minute. The prompt: “Read this operations manual. For a new accounts assistant joining the firm with no prior context, produce a 6-to-8-page summary covering only what they will need in their first 90 days. Keep the firm’s actual procedures intact; do not generalise to industry norms. Use the firm’s defined terms. List anything you have cut so I can confirm it is not needed.” The output is a working document a new hire can read in an hour rather than a handbook they will never finish.

2. Quick-reference one-pagers. The “how do I file a leave request” question, asked five times by every new hire in their first month, deserves a one-page answer pinned somewhere visible. AI builds these in minutes from the source policy: “Produce a one-page quick-reference card for the leave request process. Plain Nepali-English bilingual. Steps in order, who approves, how long it takes, what to do if rejected, who to contact if the system is broken. Include the form link and the screenshot reference.” Build twenty of these for the twenty most-asked questions and the FAQ inbox empties.

3. Assessment questions for role onboarding. For roles where comprehension actually matters — compliance, customer-facing, anything involving cash or data — produce a short assessment the new hire takes at day 14 to confirm they have absorbed the essentials. AI generates plausible questions from source material in seconds: “From the customer-data-handling policy, produce 10 multiple-choice questions a new hire should be able to answer correctly to pass onboarding. Include the correct answer and a one-line explanation for each.” Review for accuracy, deploy.

4. First-day checklists. The list of accounts to provision, accesses to grant, hardware to issue, forms to sign, introductions to schedule — every firm has one, mostly in someone’s head. AI converts the existing process into a checklist that a manager can hand to an admin and an IT person and trust to be acted on. Bilingual, dated, with owners.

5. FAQ documents. For the recurring questions where a quick-reference one-pager is overkill but a single sentence is too thin, an FAQ keyed to the role serves well. AI drafts the first 30 questions from the source policies; the team adds the next 30 over the first six months based on actual questions asked.

The compounding here is real. A focused afternoon of producing these five artefacts for one role buys back roughly ten hours of question-answering per new hire in that role, and the firm hires that role two or three times a year. The payback is measured in weeks, not quarters.

What AI cannot replace — and why pretending otherwise costs you

The temptation, once the materials are produced, is to treat the materials as the onboarding. They are not. The materials are necessary infrastructure, not the experience itself. Three things AI cannot do, and that the firm absolutely needs the trainer to do:

The live conversations of week one. A new hire’s actual questions in the first week are not the questions the handbook anticipated. They are questions like “this client just asked me for X, my line manager is in Pokhara, what do I do right now?” and “the system shows two different numbers for the same account, which one is real?” and “the colleague next to me said the policy on Y is actually different from what the document says — which is right?” These questions cannot be pre-answered. They require a human who knows the firm, who can read the context, and who can either answer or route to someone who can. A chatbot cannot do this; the chatbot does not know who the colleague is, does not know which version is real, does not know that the Pokhara line manager is in a half-day Lhosar meeting.

The relational sense of belonging. A new hire in week one is asking a question that no policy document addresses: am I welcome here, do these people want me to succeed, will the cost of asking a stupid question be paid in social capital I do not yet have? The answer is built through small interactions — being invited to tea, being introduced by name, being checked on by a manager on day three, being told “everyone asks that, here is what to do”. A bot cannot produce this; the materials cannot produce it; the firm has to. AI buys time for the human work; it does not substitute for it.

The contextual judgement about what this new hire actually needs. Two accounts assistants joining in the same month do not need the same onboarding. One is a fresh graduate from Tribhuvan with strong Excel and no exposure to a working office; the other is a mid-career switcher from a manufacturing firm with operational experience and weak software fluency. The materials are the same; the path through them is not. The human trainer is the one who knows which sections to spend extra time on with each, which to skim, and which to come back to in month two. The model can generate the materials but cannot make this judgement, because it does not know the people.

Bilingual onboarding materials

Most Nepali firms employ staff with mixed language preferences. The young engineer from Kathmandu is most comfortable in English; the accounts assistant from Birgunj is most comfortable in Nepali, with English as a second register; the warehouse supervisor in Bhairahawa works almost entirely in Nepali. Onboarding materials in only one language quietly say to half the staff: we built this for the other half.

The discipline is to produce every material in both languages, in the same document, paired. AI is genuinely useful here — drafting the English version, then producing the Nepali version that says exactly the same thing, is consistent and fast. The risks are familiar by now: the model will sometimes soften commitments in translation, sometimes mis-render numbers (mixing Nepali digits and English digits within the same line), sometimes drift into a corporate Nepali register that no one in the office uses. A fluent Nepali reader on the team checks the translated version before it ships; this is non-negotiable.

The deeper point: for many staff, the Nepali version is the version they will actually use. The English version is for the new hires who prefer it and for the international colleagues who need it. If the Nepali version is treated as the secondary translation — produced in a hurry, less carefully reviewed — the firm has implicitly told half its staff that their language is the lower-priority one. Treat the two as equal and they will be used as equal.

The first-day usefulness test

Every onboarding document the firm produces should pass one test: it answers at least one question a new hire will actually have on day one or in week one. If a document does not pass this test, cut it. The handbook section on the company’s strategic vision for 2087 fails. The “welcome to our journey” letter from the CEO fails. The seventeen-page history of the firm fails. Cut them all. They are not malicious — they are simply not what the new hire needs in the moment, and including them buries the documents that are what the new hire needs.

The test cuts both ways. It identifies the documents that should not exist; it also identifies the documents that should exist but do not. The “where do I get a sim card for the company line” question, the “what is the dress code on Friday” question, the “is there a Nepali keyboard layout installed on the company laptops” question, the “how do I expense the taxi from the airport on the way back from the Pokhara training” question — these are real day-one questions that often have no documented answer. Each one is a small piece of friction. Together, they decide whether the new hire feels the firm has prepared for them or not.

A useful audit: in the new hire’s first three weeks, log every question they ask anyone. At the end of the three weeks, look at the list and ask, for each question: was the answer in our materials, was it findable, was it correct? The questions that fail any of those tests are your next batch of FAQ updates. Do this for the next three hires and the materials converge on something that actually works.

Continuous improvement — the questions you are not anticipating

The single largest improvement to a firm’s onboarding materials is to systematically capture what new hires actually ask in their first month and feed those questions back into the materials. Every new hire will ask between five and ten questions not covered in the materials, no matter how thorough the materials were at the start. The questions are not failures of the materials; they are the materials’ best feedback loop.

The mechanism is light: ask each new hire, at the end of week four, to send you (or to record in a shared doc) the five questions they had to ask a colleague that they wish had been answered in the onboarding materials. Compile across hires. After three or four hires, patterns emerge: there are usually three or four questions that every new hire asks. Add those to the materials. After ten hires the materials cover roughly 80% of week-one questions. After that the marginal returns flatten — but the firm has converged on materials that are actually useful, which most firms never do.

AI accelerates this loop. The questions a new hire compiled into a list can be turned into FAQ entries in minutes with a structured prompt: “For each of these questions, produce a clear bilingual FAQ entry following our existing format. Use only the firm’s actual procedures provided in this document. Flag any question I have not provided enough information to answer fully.” The HR generalist reviews, accepts, ships. The next new hire arrives to materials that answer their predecessors’ questions before they think to ask them.

Check your understanding

Quick check

—

Your firm has hired a new accounts assistant who starts on Shrawan 1. You have one afternoon to prepare onboarding materials and AI tools available. Which combination of uses gives the highest leverage while keeping onboarding actually effective?

Use AI to produce role-specific summaries from the operations manual, quick-reference one-pagers for the most-asked questions, a first-day checklist, and a short day-14 assessment — and plan for the trainer to spend dedicated time on live conversations in week one, week-three check-ins, and the contextual judgement about which areas this specific hire needs more time on.
Use AI to produce a comprehensive 90-page onboarding handbook covering everything the new hire could possibly need, and rely on the handbook to answer their questions in week one.
Skip the materials and rely on a chatbot installed on the new hire’s laptop to answer their questions as they arise.
Do not produce any AI materials — keep onboarding entirely human-led so the new hire builds relationships from day one.

What comes next

You have now seen the four chapters of AI-assisted HR work that this course covers in depth — where AI fits, JDs and sourcing, screening and bias, interviews, and now policies and internal communications. The remaining question, and the one this course closes on, is the hardest: how does the HR function hold AI responsibly across all of these workflows, when the data the firm handles is some of the most sensitive in the company and when the model’s behaviour is changing month by month? Chapter 6 is about privacy, ethics, and the future of the HR function — what to log, what never to log, how to set the firm’s red lines so they hold when the model gets dramatically better, and what the role of an HR professional looks like five years from now.