AI Tools for Learning and Development: A Small-Company Shortlist

A small-company L&D lead needs four things from AI: a skills matrix to find gaps, a training plan built from it, an onboarding checklist per role, and a way to check learning stuck. Microsoft Copilot, ChatGPT, Claude or Gemini each cover all four if you feed them your own processes.

That is the whole shortlist in one line, and it is deliberately not a ranked table of forty products. A firm of ten to a hundred people rarely needs a dedicated learning platform on day one. It needs someone with limited hours to produce clearer, more consistent training faster, and to see where the gaps are. The sections below go job by job, with prompts, sample outputs and the places the tools go wrong.

Follow me on Instagram@sagnikteaches

Match the tool to the job, not the other way round

Pick the assistant your company already pays for, because the best tool is the one whose privacy terms you have read and whose licence is already approved. For most small firms that is one of four, and the differences matter less than how you use them.

Connect on LinkedInSagnik Bhattacharya

Costs, as list prices to check before you buy: ChatGPT Business is $20 per user per month billed annually or $25 monthly, with a two-seat minimum. Microsoft 365 Copilot Business is listed at $21 per user per month on an annual plan, and the Business Standard plus Copilot bundle at $23.50. Google Workspace Business Standard is $14 per user per month with Gemini in Gmail, Docs, Sheets and Meet. Claude's Team plan is $20 per seat a month billed annually or $25 monthly, for teams of 2 to 150. For one person running L&D, a personal plan at about $20 a month is the cheap way to learn the tools, but company material belongs on a business plan, where content is not used for training by default.

Subscribe on YouTube@sagnikteaches

Because model names change every few months, this tutorial uses the product names only. When a vendor tells you a new model is "better at reasoning", test it on one of your own documents instead of trusting the claim.

Step one: a skills matrix from the job descriptions you already have

A skills matrix lists roles down the side, skills across the top, and a level (none, working, confident, can teach) in each cell. Most small firms do not have one because building it takes a week. AI turns the first draft into an afternoon.

Take an illustrative 14-person accountancy practice: two partners, four seniors, five juniors, two administrators. The L&D lead pastes the four job descriptions (with no names) and asks for a draft.

Here are four job descriptions for an accountancy practice: partner, senior, junior, administrator.
Build a skills matrix as a table. Rows: skills (group them under Technical, Client handling,
Systems, Management). Columns: the four roles. In each cell give the expected level
by the end of year one in the role: 1 aware, 2 working, 3 confident, 4 can teach.
Only use skills that appear in the descriptions or follow directly from them.
Mark any skill you inferred with an asterisk so I can confirm it.

An illustrative extract of what comes back:

Technical | Year-end accounts preparation | Partner 4 | Senior 3 | Junior 2 | Admin 1
Technical | Review of working papers | Partner 4 | Senior 3 | Junior 1 | Admin -
Client handling | Chasing missing records* | Partner 2 | Senior 3 | Junior 3 | Admin 3
Systems | Practice-management software | Partner 2 | Senior 3 | Junior 3 | Admin 4
Management | Delegating and reviewing work* | Partner 4 | Senior 3 | Junior - | Admin -

Two things to fix before anyone sees it. The asterisks are the assistant telling you what it guessed: check them with the partners. And the levels are targets, not facts about people. The matrix only becomes useful when each person's current level is added, which is a conversation, not a prompt. Then paste the filled-in matrix (initials or role codes, not names) and ask: "List the three largest gaps between current and target level across the team, and for each say which of our people could teach it."

That last request is the one most L&D guides skip. In a small firm the best trainer is usually a colleague at level 4, and the matrix shows who. For a similar approach to people and AI skills more generally, see a four-week plan for owners learning AI.

Step two: turn gaps into a training plan with dates and owners

A plan fails when it lists topics without a person, a date or a way to know it worked. Give the assistant the gaps, your constraints (hours a week, busy periods, budget) and ask for a quarter. Say the practice has busy weeks before filing deadlines and wants no more than three hours of training per person per month.

Gaps (from our matrix): 1) juniors weak on review-note writing; 2) seniors weak on delegating;
3) two juniors new to the practice-management software.
Constraints: max 3 hours of training per person per month; no sessions in the two weeks
before our main filing deadline; one partner (Partner A) can teach delegating; budget $600 for the quarter.
Draft a 12-week plan as a table: week, topic, format (peer session, self-study, external),
who leads, time per person, how we will check it worked, cost.
Prefer internal sessions. Do not name external courses or providers.

The final line is deliberate. Assistants will happily invent plausible courses, providers and "accredited" certificates. Ask for the structure and find real external options yourself on the provider's own site.

A sensible result gives each gap one internal session, one practice task and one check. For review notes, for example: week 2, a 45-minute peer session led by a senior using three anonymised old review notes; week 3, each junior rewrites one note; week 4, a senior marks them against a four-line checklist. The "how we check" column is what turns a calendar into a plan. If the assistant writes "feedback survey" for every row, push back and ask for an observable behaviour in the work.

Step three: onboarding by role, built from your real process

Onboarding is where small companies lose the most time and where AI helps the most, because the content already exists in people's heads and scattered documents. The trick is to build the guide from your own material, not from a generic template. Pull together the welcome email, the software logins list (without passwords), the "how we handle a new client" notes and the org chart, and ask for a first-week plan.

You are helping onboard a new junior at a 14-person accountancy practice.
Using only the attached documents, write a first-week plan, day by day, with:
- what they should be able to do by the end of each day
- who they spend time with
- which document or system to read, named exactly as in the files
- one small real task on days 3 to 5 with a clear finish
Flag anything the documents do not cover as "NEEDS AN ANSWER" and list those questions at the end.

The "NEEDS AN ANSWER" list is where the value shows up. Say it returns: who orders the new starter's laptop, which clients a junior may email directly, and when the first review meeting is. These are the gaps that make the first week feel chaotic, and finding them takes a minute instead of a bad first Monday. Fill the gaps, then keep the finished guide as a shared document or as the instructions in a shared Project in ChatGPT or Claude, so a new starter can ask "where do I find the template for a client engagement letter?" and get an answer grounded in your files. Whoever builds that Project should check what it can see: only upload documents you would be happy for every new starter to read.

An illustrative extract of the day-3 entry shows what to check: "Day 3. By the end of today: log in to the practice-management software and find a client's open jobs. With: a senior, 10:00 to 11:00. Read: 'New client checklist' and 'Engagement letter template'. Task: draft the missing-records request for one small-business client from the template; a senior reviews it before it is sent." Two fixes are typical. The assistant may name a document slightly differently from the file ("New client checklist" when the file is called "Client take-on checklist"), so match the names exactly or the new starter will not find it. And it may set a task that touches a real client before anyone has agreed whether juniors email clients, which is one of the open questions; use a dummy client until that is settled.

For a day-by-day structure and a first-two-weeks view, see AI employee onboarding: plan a new starter's first two weeks, and for teaching new people your AI rules in particular, how to onboard new hires onto your AI tools and rules.

Step four: lessons and quizzes from your own procedures

Staff learn from their own firm's examples. Take one real procedure, say the process for handling a late client record, strip client names, and ask the assistant to produce a short lesson.

Below is our written procedure for handling late client records (client names removed).
Create:
1. A 300-word lesson for a new junior in plain language.
2. Two realistic scenarios with a right and a wrong response, based only on the procedure.
3. Five quiz questions, three multiple-choice and two short-answer, with answers.
Do not add steps that are not in the procedure. If the procedure is ambiguous, list the ambiguity.

Run it, then read the "ambiguity" list. That list often exposes the real problem: the procedure says "chase twice" but not when, so two seniors do it differently. Fixing the procedure is worth more than the quiz. When a quiz question is wrong, as one will be, fix the question and the procedure it came from. For ways to make quizzes reveal what actually stuck, see how to use AI in business training: quizzes that show what stuck.

Handling training requests without a spreadsheet of chaos

In a small firm, training requests arrive as hallway comments and emails: "Could I do a pivot-table course?", "Can someone show me the new invoicing screen?" The L&D lead ends up as a switchboard. Paste a month of these requests (role codes, not names) and ask the assistant to group them.

Group these 18 training requests into themes. For each theme give: number of requests,
roles involved, whether it looks like (a) a one-off question, (b) a skill gap needing a session,
or (c) a process problem that training will not fix. Suggest the cheapest sensible response for each.
Requests: [paste]

An illustrative result: seven requests about the invoicing screen (a process problem, because the screen changed and nobody updated the guide: fix the guide), five about spreadsheet formulas (a skill gap for juniors: one internal session plus the follow-up on using ChatGPT with Excel as self-study), four one-off questions (answer them), and two about leadership (a development need for the seniors, to go into the quarter's plan). Category (c) is the one that saves money. In this illustration seven of the eighteen requests pointed at a broken process, and no course fixes that.

A compliance and refresher calendar

Every company has recurring training it must repeat: safety briefings, data-protection reminders, client-confidentiality rules, anything a regulator or insurer asks you to prove. List each item, who needs it, how often, and the evidence you keep. Then ask the assistant to build a twelve-month calendar that spreads the load so that no month has more than two sessions, and to draft the reminder email for each.

Two cautions. The assistant must not decide what the legal or contractual requirement is: you take that from your adviser, insurer or contract and paste it in as a fixed list. And it must not draft attendance records. A register is signed by the person who attended. An AI can produce a blank register template and a reminder to complete it; it should never produce a filled one.

A short filled-in row of the calendar looks like this, illustratively: "Client-confidentiality refresher | all staff | every 12 months | March | 30 minutes led by a partner | evidence: signed register plus three-question check | reminder email sent two weeks before". Adding the evidence column during planning, not after an inspection, is the real gain.

A coaching guide for line managers

The weakest link in small-company training is often the line manager who is expected to coach but was never shown how. A one-page guide per role can fix more than a course. Give the assistant the matrix level descriptions and ask for a guide a senior can use in a 20-minute one-to-one with a junior: three questions to ask, what a level 2 versus a level 3 answer sounds like, and one small task to set. Test it with one senior before rolling it out. Ask them whether the guide matched how they actually talk, because a guide that sounds like an HR policy will not be used.

Using the tools for a manager's own learning, too

The person running L&D usually has no budget for their own development. A chat assistant works as a study partner for the subject itself. Ask it to explain a concept, such as how adults retain new procedures, then ask it to quiz you, then ask where its own explanation is weakest. Treat what it says as a lead to follow up, not a citation. Anything you plan to put in front of the team, such as a statistic about retention or a named learning model, should be checked on a reputable source first, because assistants state half-remembered research with confidence.

A practical routine: thirty minutes a week, one topic, one output. Week one, "write the three questions I should ask a new starter on day five". Week two, "draft a short briefing for line managers on giving feedback". Each output goes into the source folder after a review. In a quarter you have twelve usable pieces and a clearer sense of which tool suits you.

Finally, agree who owns the assistant's instructions. If your firm uses a shared Project, one person edits the instructions, keeps a changelog of what changed and why, and tests two or three standard questions after each change ("Where do I find the engagement letter template?", "Who approves a training request?"). That small discipline stops the shared assistant drifting as documents change.

Where tools go wrong in training work

Invented sources. Ask any assistant to "recommend a course" and it may produce a title, a provider and sometimes a price that do not exist. Rule: the assistant proposes skills and structure; you find the courses on the provider's site.

Generic content that sounds like every other firm. A lesson built without your process reads like a textbook, and staff stop reading it. A realistic sign is that a new starter asks the same questions the lesson "covered". The fix is to attach your real documents and ask for flagged gaps.

People data in prompts. Appraisal notes, grievances, health information and pay do not belong in a chat tool. If you need to analyse team gaps, use role codes and skill levels. A typical way it happens: a manager pastes an appraisal to "help phrase feedback", and the document now sits in a tool the company has not assessed.

Over-trusting a quiz score. A junior scoring 5 out of 5 on a recall quiz may still miss the procedure at the desk. Use quizzes as a prompt for a conversation and check the behaviour on the job.

A quick sum: what a year of this costs

Say the L&D lead is one person for the 14-person practice, spending about six hours a week on training and onboarding, and has a business chat plan. Five hours a quarter on the skills matrix refresh, plus one new-starter guide per hire, is where the saving lies. Suppose the practice hires three people a year and each guide used to take about 12 hours to write and now takes about 5 with AI drafting and a real review: that is 21 hours saved. Add four hours a quarter on the matrix and plan saved, and the total is about 37 hours a year. Against that, one business seat at $20 to $25 a month is $240 to $300, but the two-seat minimum on some plans doubles it. At a modest internal cost per hour the saving still clears the subscription comfortably, and the more important gain is consistency: every new starter gets the same quality of week one.

Treat those figures as an illustration. Count your own hours for a month before and after, because the value depends entirely on how much of this you currently do by hand.

Five habits that keep a small L&D function honest

1. Keep a source folder. One folder of procedures, templates and examples that the assistant draws from, owned by a named person and reviewed twice a year.
2. Date every guide. Put "Reviewed: month and year, by: role" at the top of each onboarding guide so staff know when it was last true.
3. Ask for gaps, not completeness. Prompts that end with "list anything the documents do not cover" find missing procedures faster than any template.
4. Pilot with one person. Give the AI-built first-week plan to one new starter, ask what was missing, and revise before it becomes standard.
5. Measure behaviour two to four weeks later. Choose one observable result per programme: a task done unaided, fewer repeat questions, a shorter time to first independent job.

If you prefer to work through your own matrix and onboarding process with someone, that is what my AI implementation consultation covers; you can get in touch here. Otherwise, start with the skills matrix. It takes an afternoon, and every other training job gets easier once it exists.

Small-company learning and development: common questions

Do I need a learning management system to use AI for training?

Not at the start. A shared folder, a spreadsheet skills matrix and a chat assistant cover training plans, onboarding checklists and quizzes for a team under about fifty. Consider an LMS when you must prove completion to a client or auditor, or when the same course goes to many new starters every month.

Can AI write our training content from scratch?

It can draft outlines, scenarios and quizzes quickly, but your own procedures and examples should be the source material. Paste the real process in and ask for a lesson built from it. Content invented from general knowledge tends to sound right and miss how your firm actually works.

Is it safe to paste staff performance details into a chatbot?

Avoid it. Use roles and skill levels instead of names, and keep appraisals, grievances and pay out of prompts. If you do need to discuss an individual, use a business plan with training on your content switched off by default, and check your own policies first.

How do I know the training worked?

Decide the evidence before you start: a task done correctly without help, fewer repeat questions, a shorter time to first independent job. A quiz score alone shows recall, not behaviour. Check on the job two to four weeks later.

Further reads

• How Much Time and Money Does AI Staff Training Take? — Budget the time and money for teaching a team to use AI.
• AI Training for Construction Companies: Workshop or Self-Taught? — Workshop or self-taught: a decision for a trades firm.
• How to Stop AI Tools Training on Your Business Data — Switch off training on your content before staff paste anything.
• How to Build a Kitchen Training Manual With AI — Build a procedure manual from what staff already do.
• How to Test Job Candidates' AI Skills in an Interview — Check AI skills before hiring, not after.
• AI Training for Dietitians: A Practice Owner's Team Plan — A practice owner's training plan in another small-team setting.
• Apprentice Off-the-Job Training Log: What to Record and How AI Helps — An employer-side training log: the fields, a filled-in week, sign-offs, a month-end check against the provider's records, and where AI helps.
• AI in Employee Training: Turn Your SOPs Into Short Modules and Quizzes — A practical SOP-to-lesson workflow, including quiz checks, manager approval and a small-team completion sheet.
• How to Start a Training Company: Accreditation, Insurance and AI — First-year costs, how to pick an accreditation route, which insurance to price, and the proposal, booking and certificate admin AI can handle.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ChatGPT Business pricing page, Microsoft 365 Copilot Business pricing page, Google Workspace pricing page, Claude pricing page (all checked 4 November 2026).