On this page
TL;DR:
- Create an in-house AI trainer role when you have steady training work for at least the next 12 months. Our rule of thumb is two days a week or more.
- In-house is the better choice when your data cannot be shown to outsiders, when your work depends on your own systems, or when you take on new staff all the time.
- An outside provider is usually the better choice for a first rollout, when nobody inside can teach AI yet, or when the need is a one-off.
- Most organisations should start with a part-time role held by someone who already works there. A full-time post is hard to fill with enough work.
- NITA's 2019 guidelines exclude training run by your own staff from levy reimbursement. Check the current position with NITA before you budget.
Who this guide is for
This guide is for HR managers, learning and development leads and team heads who are asking one specific question: should we have our own AI trainer?
It looks at the role itself: the work, the true cost, who to pick and how to keep that person current. If you want the wider comparison of training programme models, read our page on in-house versus external AI training.
We sell outside training, so we have an interest here. We have tried to set out the cases where you should not call anyone like us.
What does an in-house AI trainer actually do?
An in-house AI trainer teaches staff to use AI safely in their daily work, then keeps that skill alive after the first sessions. Delivering sessions is often the smaller part of the job.
| Part of the job | What it involves in practice |
|---|---|
| Finding the need | Sitting with departments and listing the tasks that take too long |
| Preparing material | Building exercises from real work, with personal data removed |
| Delivering sessions | Teaching groups of about 20, with help at each screen |
| Follow-up | Clinics, reviewing real work, answering questions on chat |
| New joiners | Adding AI basics to induction |
| Staying current | Testing new features and updating the material |
| Rules and safety | Working with ICT and your data protection lead on what is allowed |
| Reporting | Showing management what has changed |
If you are looking for a person who only runs workshops, you may not need a post at all. Your HR or training team may be able to take that on. Our guide on AI training for HR teams shows how an HR team can run this training.
When is in-house the better choice?
In-house is the better choice when the need is continuous and closely tied to your own systems. If most of the points below describe you, an outside provider should not be your long-term answer.
Your data cannot be shown to outsiders. Some organisations handle information they will not put in front of an external trainer, even with a confidentiality agreement. An employee already inside your rules can work on real files in your approved tools.
The work depends on your own systems. A generic course cannot teach AI inside a core system only your staff use. Someone who knows those screens and exceptions can.
You take on new staff all the time. Call centres, branch networks and fast-growing firms have new people every month. Booking an outside provider for each small intake is awkward. An internal trainer can fold AI into induction.
You need help at the moment people get stuck. People often get stuck on an ordinary afternoon, weeks after the workshop. A colleague two desks away can answer in five minutes.
You already have the person. Sometimes one careful, respected employee is already teaching everyone. Giving that person time and recognition can be cheaper and better than any contract.
Your staff work in ways outsiders find hard to reach. Night shifts, remote sites and field teams can be easier to serve from inside.
When is an outside provider the better choice?
An outside provider is usually better when you need a strong start quickly, or when nobody inside has the skill yet.
- First rollout. Someone has to teach the first group, and that first group includes your future trainer.
- Nobody inside can teach this yet. Using AI well and teaching it are different skills. Our guide to what makes a great AI trainer sets out the difference.
- A one-off need. A board session or a short series for one department will not fill a job.
- Leaders will not listen to a colleague. Senior teams sometimes accept a hard message more easily from outside.
- No AI rules exist yet. An outsider who has seen other organisations can help you draft a first version faster.
- You want levy reimbursement. That route needs a NITA-registered provider. See the NITA section below.
- You want a test before committing. A short outside programme shows you how much demand exists before you create a post.
Before you sign, compare providers with our guide on how to choose an AI training provider in Kenya.
Score your own situation
Give one point for each statement that is true today. This scoring is our own rule of thumb. It has no research behind it, so use it to start a discussion.
| Statement | Point |
|---|---|
| We can list at least two days a week of AI training work for the next 12 months | 1 |
| We take on new staff most months | 1 |
| Some of our work cannot be shown to an outside trainer | 1 |
| Our staff rely on internal systems an outsider would not know | 1 |
| A named person already teaches colleagues and does it well | 1 |
| That person's manager will truly release their time | 1 |
| A senior sponsor will back this role for at least a year | 1 |
| We already have written AI rules and approved tools | 1 |
| Score | What it suggests |
|---|---|
| 0 to 2 | Use an outside provider. Revisit the question in a year |
| 3 to 5 | Outside provider first, with a part-time internal trainer taking over gradually |
| 6 to 8 | A dedicated internal role is worth building. Consider outside help for the first cohort and for refreshers |
How much work is there really?
Work out the days before you write the job advert. You may imagine a full-time job and find a part-time one.
Here is a worked example. All the numbers are made up. Replace them with your own.
Assume 240 office staff, groups of 20, two training days per group, a half-day follow-up each, and one day of preparation for each day delivered.
| Work in year one | Delivered days | Preparation days |
|---|---|---|
| 12 groups, two days each | 24 | 24 |
| 12 follow-ups, half a day each | 6 | 6 |
| 4 new joiner sessions, one day each | 4 | 4 |
| Clinics, prompt library and rules, two days a month | 0 | 24 |
| Total: 92 days | 34 | 58 |
If a person has about 220 working days a year after leave and public holidays, 92 days is about 42% of a full-time job. That is roughly two days a week.
So in this example, 240 staff give you a real part-time role and not a full-time one. The second year is often lighter unless you add new content or new departments.
What does an in-house AI trainer really cost?
The salary is only part of the cost. Count the employer's costs, the trainer's own learning, tools and the staff time released to attend.
Using the same example, again with made-up numbers that are not market rates:
| Cost for one year | KES |
|---|---|
| Gross pay at 150,000 a month | 1,800,000 |
| Employer costs and benefits, assumed at 10% | 180,000 |
| The trainer's own courses and development | 120,000 |
| AI tool licences and equipment | 90,000 |
| Total for a full-time post | 2,190,000 |
| Option | Cost in the year | Cost per delivered day (34 days) |
|---|---|---|
| New full-time hire | 2,190,000 | About 64,400 |
| Existing employee, 42% of their time | About 919,800 | About 27,100 |
The second line looks cheap, but somebody has to do the other 58% of that person's job. If nobody does, the training work is usually the first thing to slip.
To compare fairly, ask providers to quote for the same 34 delivered days, including preparation and follow-up. Leave out the time your staff spend attending, because that cost is the same either way. Our guide on how to budget for AI training lists other lines to include, and how to evaluate ROI on AI training explains how to measure what you get back.
Should you hire someone new or grow someone you already have?
For most Kenyan organisations, growing someone you already have is the better first step. They already know the work and the people. Hire from outside when nobody suitable exists or when the workload truly fills a post.
| Factor | Grow an employee | Hire someone new |
|---|---|---|
| Knows your work and systems | Yes, from day one | Takes months |
| Trust from colleagues | Usually high | Has to be earned |
| Teaching skill | May need developing | You can test for it at interview |
| AI knowledge | May need developing | You can test for it at interview |
| Risk | The person's old job gets neglected | The role may not have enough work |
| Speed to start | Fast | Slower, because of recruitment and notice periods |
Who to look for
Choose a person who explains things patiently and checks their own work. The most enthusiastic AI user in the building may be a poor choice if they skip checking and make others feel slow.
Look for these signs:
- Colleagues already ask this person for help.
- They can explain a hard thing in plain words, without jargon.
- They are careful with data and admit when they do not know.
- They are curious enough to keep testing new tools.
- They do not push one favourite tool.
A simple selection test
Ask each candidate to teach for 15 minutes. Give them a real but harmless task, such as turning rough notes into a clear memo with an AI tool. Put two or three staff who know little about AI in the room.
Watch whether the learners do the work themselves. Then ask the candidate to show you how they would check the output for errors. Mistakes in that part matter more than a polished delivery.
How do NITA rules affect this decision?
NITA's published guidelines exclude "training conducted by employer's own resource persons" from levy reimbursement. So sessions run by your internal trainer cannot be expected to bring any levy money back.
Here is what we confirmed on NITA's and KRA's own documents on 10 November 2026. NITA's guidelines date from February 2019. That is still the version on NITA's website, and we found no newer one. Check anything that affects your budget with NITA in writing.
| Point | What the source says |
|---|---|
| Levy rate | NITA says KES 50 per employee per month. That is KES 600 per employee per year |
| How it is paid | KRA says that for periods from May 2023 the levy is calculated within the PAYE return |
| Who can train | An employer "may only train with a Training Provider who is registered and/or renewed" by NITA |
| When to apply | At least 7 days before local training |
| In-house course size | 8 to 25 participants per session, unless the Director General makes an exemption. There is no limit for in-house team-building and sensitisation courses |
| Rate for private providers | Set per participant each year. We did not find the current rate on the NITA pages we opened |
| Claim deadline | Within 60 days of the training ending. The Director General can accept a late claim on written request |
| Own staff as trainers | Not eligible for reimbursement |
NITA announced that, from 1 November 2025, employers would apply for training approval through an online system.
Three practical points follow:
- Do not let the levy drive this decision. A company with 240 staff pays KES 144,000 a year in levy. Compare that with the cost of a post. Reimbursement may help, but it is too small and too uncertain to justify either route.
- Your future trainer's own course may be claimable. If you send that person to an approved course with a registered provider, the normal rules should apply. Confirm with NITA before you enrol them.
- NITA also offers its own trainers. The guidelines mention using the levy fund for services from NITA, including NITA trainers. Ask whether this could help with general skills such as teaching methods.
FuKazee does not yet hold NITA accreditation. If levy reimbursement matters for your budget, plan on that basis and ask any provider, including us, to show their current status.
How do you keep an in-house trainer current?
Give the trainer protected time every week, or their material will go out of date within months. AI tools change too often for occasional reading.
Microsoft's Copilot release notes show how fast the ground moves. Between 23 September and 4 November 2026, Microsoft posted four sets of updates. Other vendors change what they offer too, as our guide on what to do about custom GPTs shows.
We suggest these habits. They are our advice, and we have no study to prove the exact amounts:
- Keep half a day a week for testing tools and updating material.
- Follow the official release notes of the tools your organisation uses, and ignore most online hype.
- Keep a dated list of what each session teaches, so that old content is easy to find.
- Test the free version and the paid version of each feature your staff might meet.
- Join at least one outside community of trainers, so the person is not working alone.
Does training the trainer work?
It can work, though the evidence comes from another field. A 2012 systematic review by Pearce and others looked at 18 train-the-trainer studies in health and social care. Thirteen reported gains in knowledge, behaviour or patient outcomes. One found no effect, three showed possible effects, and one found that a CD-ROM course taught knowledge better than train-the-trainer. The authors rated the studies at moderate risk of bias and suggested a blended approach, while saying more research is needed.
We found no equivalent review on AI training. So treat a train-the-trainer plan as a reasonable bet, and watch the trainer teach before you hand over the programme.
How do you avoid depending on one person?
Write everything down and keep it where others can reach it. If all the knowledge sits with one trainer, the programme stops when that person leaves.
- Keep slides, exercises and prompts in a shared folder owned by the organisation.
- Name a deputy who sits in on sessions and can run the basics.
- Record a short version of the core session, with no real data in it.
- Add the role to a job description, so that it survives a change of manager.
- If you plan to pay for expensive outside training and want a commitment in return, take advice from an employment lawyer first.
A 90-day plan for your first in-house trainer
Days 1 to 30: set up
- Confirm the person, the time they are released and who covers their other work, in writing.
- Agree the AI usage policy or write a first version.
- Have the trainer attend a good external programme as a learner.
- Run our AI readiness assessment to see where you stand.
Days 31 to 60: teach with support
- Choose one friendly department and three real tasks.
- The trainer delivers one session while someone experienced observes.
- Collect what confused people, and fix the material.
- Start the shared prompt library.
Days 61 to 90: review
- Run a second group without the observer.
- Hold one clinic and review real work from at least three people.
- Compare time taken before and after, including time spent checking.
- Decide whether to continue, expand or stop. Stopping is a fair result.
Three prompts to help you decide
These use the four Cs we teach: Clarity, Context, Constraints and Customisation. Remove names and salaries before pasting anything into a tool.
1. Estimate the workload
Help me estimate the workload for an internal AI trainer. Context: we have [number] staff in these departments: [list]. Groups are 20 people or fewer. Constraints: show a table with delivered days and preparation days for year one, using two training days per group, a half-day follow-up and one day of preparation per delivered day. State every assumption, and do not add any activity I have not listed. Customisation: the reader is our HR director, who wants to know if this is a full-time or part-time role.
2. Draft the role description
Draft a one-page role description for an in-house AI trainer at a [type of organisation] in Kenya. Context: this person will teach staff who have never used AI, and will work with ICT and our data protection lead. Constraints: plain English, no more than 300 words, include what the role does and does not do, and do not invent a salary. Customisation: this will be read by current staff who may apply, so make it clear no technical degree is needed.
3. Check what a provider should hand over
I am comparing outside AI training providers. Context: after the programme, our own staff will continue the training. Constraints: give me 10 questions to ask each provider about what they will hand over, who owns the material, how it can be updated and what support continues afterwards. Customisation: put them in a table with a column for what a good answer sounds like.
What goes wrong, and how to avoid it
| What goes wrong | Why it happens | What to do |
|---|---|---|
| The role is added on top of a full job | Nobody released the time | Put the released days in writing, signed by the line manager |
| Attendance falls after a good start | Managers see it as optional | A senior sponsor sets expectations and joins a session |
| The trainer teaches one favourite tool | Personal habit | Teach methods first, and use the tools the organisation has approved |
| Material gets out of date | No time to keep up | Protect weekly time, and date every session |
| Staff treat the trainer as IT support | The role was never explained | Say what the trainer does and does not do |
| Real data appears in exercises | Short preparation time | Prepare masked sample files before every session |
| The trainer leaves and everything stops | Material lives on one laptop | Use a shared folder and a named deputy |
| The post has too little work | Numbers were never done | Calculate the days before advertising |
Questions people ask
How many staff justify a full-time AI trainer?
There is no reliable figure, because it depends on how much training and follow-up each person needs. In our made-up example, 240 office staff gave about two days of work a week. Work out your own days before you decide on a post.
Does an in-house AI trainer need a technical background?
For teaching everyday use of AI at work, teaching ability and care with data matter more than technical training. Someone who builds AI systems is a different hire. Keep ICT involved so that the trainer knows which tools are approved.
Can in-house AI training be claimed from NITA?
NITA's guidelines exclude training run by the employer's own resource persons, so do not budget for reimbursement on those sessions. "In-house course" in those guidelines can also mean a registered provider teaching only your staff, and that can be a different case. Ask NITA about your specific plan.
Can the role be part-time?
Yes, and for many organisations it should be. The part-time arrangement only works when the person's other duties are reduced in writing and the time is protected.
What if our trainer leaves?
Plan for it from the first day. Keep all the material in a shared folder, name a deputy and record the core session without real data. That way a departure slows you down without ending the programme.
Should the trainer sit in HR or ICT?
Either can work. HR or learning and development usually fits better, because the job is mostly about people and habits. The trainer still needs a close working link with ICT and your data protection lead.
Sources
- NITA, Guidelines for Training and Reimbursement, NITA/IBTA/G/03, February 2019, last checked 10 November 2026
- NITA, Levy Inspectorate, last checked 10 November 2026
- NITA, "NITA Unveils Online Skills Upgrading Training Approval System", last checked 10 November 2026
- KRA, NITA Levy Compliance by Employers (notice, 2024), last checked 10 November 2026
- Microsoft, Release notes for Microsoft 365 Copilot, last checked 10 November 2026
- Pearce and others, "The most effective way of delivering a train-the-trainers program: a systematic review", Journal of Continuing Education in the Health Professions, 2012 (Epistemonikos record), last checked 10 November 2026
Where FuKazee fits in
FuKazee is an outside provider, so here is where we can and cannot help. We can teach your first groups, and we can work alongside the person you want to grow into your internal trainer. We can leave your team with exercises built from your own tasks.
We assume nobody knows AI, we avoid jargon, and we say plainly that nobody becomes an expert in a few days. We show the free version of every feature, we never promote one tool, and Microsoft 365 and Copilot are part of what we teach. Since 2022 we have trained more than 3,000 people.
If your numbers point to an in-house role and you do not need us, that is a good result. You can see how our AI training in Kenya works if you want to compare.
To talk through whether your organisation needs an in-house trainer, email us at lead@fukazee.com
Want guided support instead?
Plan AI training around your team's real work. Answer a few quick questions about your sector, team size and goals, and we will recommend the right programme and format. Sessions are built on your workflows and include a usage policy.
Rodney has run FuKazee's training since 2022 and still teaches inside every programme line: corporate sessions, the postgraduate cohort and the KCSE examiner webinars. First Class Honours in Journalism and Mass Communication, University of Nairobi.
