On this page
TL;DR:
- Mombasa businesses get the fastest return from AI on office work around cargo and guests: client updates, document checks, claims letters, safety notes and guest replies.
- AI can read, compare and draft. It should never choose a tariff code, set a customs value or decide who pays a claim. A named person does that.
- The legal risk is real. Kenyan court decisions summarised by KRA say a customs agent is personally liable for duties on goods it declares, and the owner of the goods stays liable too.
- Keep client names, KRA PINs, rates, driver IDs and guest passport details out of public AI tools. Mask them first, or use a tool your company has approved.
- Start with one task, one team and four weeks. Time the task before and after, including the time spent checking.
Who this guide is for
This guide is for people who keep Mombasa's trade and tourism moving from a desk. That means clearing and forwarding staff, transport and haulage offices, warehouse and yard supervisors, accounts teams, safety officers, hotel front offices and reservations staff, and the owners of small firms in any of these.
You do not need a technical background. If your day is full of emails, entries, delivery notes, claims and guest messages, this is for you.
Our page on AI training in Mombasa explains how training is arranged. This guide goes further into the work itself, with prompts you can use today, whether you ever train with us or not.
Why paperwork is the pressure point in Mombasa
Volume is rising faster than most office teams are growing, and every extra tonne of cargo creates messages, documents and questions for someone at a desk.
The Port of Mombasa handled a record 45.45 million tonnes in 2025, up 10.9% on 2024. Containers reached 2.11 million TEUs (the standard 20-foot container unit), up 5.5%.
The more telling number for a Mombasa office is transit cargo. It rose 19.5% to 15.88 million tonnes. Transit to Uganda grew 25.2%, Rwanda 22.8%, DR Congo 16.5% and Tanzania 11%.
Transit work is communication-heavy. Your client may be in Kampala, Kigali or Goma, and the goods cross borders and pass through more hands. Each stage produces a question: where is my cargo, what is holding it, what do you need from me?
Delays add to this load. The Standard reported in February 2026 that the port was facing congestion even as cargo beat forecasts. When boxes wait, phones ring, and the same explanation gets typed again and again.
Mombasa County also carries weight beyond the port. According to reporting on the Kenya National Bureau of Statistics' Gross County Product 2024 report, it ranked fourth among counties in nominal output and contributed 9.9% of Kenya's manufacturing value added.
Hospitality has its own version of the same squeeze. KNBS figures reported by The Star show 2.55 million international arrivals in 2025, up 6.2%, with August the busiest month. Hotel bed-nights rose 12.6% to 11.56 million across the country, and Kenyan residents accounted for 45% of occupancy.
At the coast, The Standard reported in August 2026 that hotels are leaning on domestic visitors, conferences and corporate meetings to stay busy through the year. One Mombasa Island hotel said local tourists make up about 70% of its clients. That changes the writing a hotel needs: fewer brochures for overseas agents, and more quick, tailored replies to a teachers' association, an NGO planning a workshop or a family booking for the school holidays.
Where AI helps and where it does not
The simple test is this: can a person check the output quickly against something real? If yes, AI is a good fit. If the output is a judgement that carries legal or financial weight, AI can only prepare the ground.
| Task | Use AI? | What AI does | What the person still does |
|---|---|---|---|
| Client status updates | Yes | Turns your tracking notes into a clear message | Confirms the facts and the next date |
| Comparing documents for mismatches | Yes, on masked copies | Lists differences between invoice, packing list and bill of lading | Opens the originals and resolves each difference |
| Summarising a long circular or tariff notice | Yes | Gives the main points and who is affected | Reads the original before acting |
| Claims and incident letters | Yes | Drafts from your facts, in order | Checks the facts, the policy wording and the amounts |
| Toolbox talks | Yes, from approved procedures | Turns a procedure into a short talk | Safety officer approves it |
| Excel for accounts and fuel | Yes | Writes formulas and explains them | Tests them on known figures |
| Guest replies and review responses | Yes | Drafts in your house style | Reads every reply before it goes |
| Choosing an HS tariff code | Only as a question list | Suggests what facts you need | Licensed agent decides, using the official tariff |
| Declaring customs value | No | Nothing | Agent and owner |
| Deciding fault in an accident | No | Nothing beyond ordering the timeline | Investigator, insurer, management |
| Deciding who is hired or disciplined | No | Nothing beyond formatting | HR and management, with a right to human review |
Clearing and forwarding: AI drafts, you declare
For clearing and forwarding firms, the safe use of AI stops at the moment a declaration is made, because the law holds people responsible and a chatbot has no licence to lose.
KRA says its Integrated Customs Management System (iCMS) lets it receive declarations before a vessel docks. So the accuracy of what your team prepares early matters a great deal.
KRA's own digest of tax and customs cases is worth reading here. It records courts saying that under section 147 of the East African Community Customs Management Act, 2004, a customs agent is treated as the owner of goods it has declared and is personally liable for the duties. It also records that this does not free the actual owner, and that importers can carry the loss when an agent acts fraudulently.
Two practical points follow.
If you are an agent, an AI error in a code or a value is your error. AI tools can state a wrong tariff code with complete confidence. They may also use an old version of the tariff. Always check against the current official tariff and your own past rulings.
If you are an importer, ask your agent how they use AI and what they check. A careful answer should reassure you. As KRA's digest puts it, a prudent taxpayer monitors its agent.
Sample prompt: the "where is my cargo" update
This is usually the best first task, because it is frequent, low risk and easy to check.
I work at a clearing and forwarding company in Mombasa. Below are my tracking notes for one consignment, with the client's name removed. Write a short update to the client's logistics officer in Kampala. Context: the client cannot see our systems and is under pressure from their own customer. Constraints: under 120 words; say the current stage first, then the reason for any delay, then what happens next and the date of my next update; use only the facts in my notes; do not promise a delivery date; if something is unclear, write [CHECK] instead of guessing. Customisation: plain, polite English suitable for WhatsApp or email. Notes: [paste notes with names, PINs and rates removed]
The line "do not promise a delivery date" matters. AI tends to sound reassuring, and a reassuring promise you cannot keep costs more than a late message.
Sample prompt: the document mismatch check
I will paste text from three shipping documents for one consignment: commercial invoice, packing list and bill of lading. Names, addresses and prices are replaced with labels such as [SHIPPER] and [PRICE A]. Compare them. Give me a table with these columns: field, what each document says, match or mismatch. Check at least: description of goods, number of packages, gross weight, net weight, container numbers, seal numbers, and port of loading and discharge. Do not correct anything. Do not suggest a tariff code. If a field is missing from a document, write "not found".
Scanned documents are a known weak point. AI can misread a poor scan, especially numbers. Treat this prompt as a second pair of eyes after your own check, and never as the only check. A wrong digit in a container number is exactly the kind of thing it can get wrong while looking sure of itself.
Sample prompt: preparing for a tariff question
I need to find the correct tariff classification for a product. Do not give me a code. Instead, list the facts about the product that a customs officer would need to classify it, such as materials, function, how it is presented and whether it is a part or a complete item. Then list the questions I should ask the supplier. Product description: [generic description]
This uses AI for what it does well, which is organising your thinking, and it leaves the decision with the licensed person.
Haulage and transport: letters, safety and shifts
Transport companies spend many hours on writing that follows a pattern: incident reports, claims, driver notices, route updates and monthly summaries. AI handles the pattern, and your staff supply the facts.
In October 2026 we trained two teams at A.O. Bayusuf and Sons, a transport and haulage company in Mombasa, on site, over three days each. Timing is a real planning point for any coastal operation: run teams one after the other so that some staff stay on duty while others train, and agree early starts where shifts need them.
Sample prompt: turning an incident report into a clear timeline
Below are rough notes about a road incident involving one of our trucks. Names and registration numbers are replaced with labels. Turn them into a factual timeline. Constraints: list events in time order; separate what is known from what was reported by someone else; do not state or suggest who was at fault; do not add any detail that is not in the notes; mark gaps with [CHECK]. At the end, list the documents an insurer is likely to ask for so that I can confirm them with our insurer. Notes: [paste]
The instruction not to suggest fault is important. A draft that casually blames the driver can cause real harm if it reaches a client or an insurer.
Sample prompt: a toolbox talk from your approved procedure
Below is our approved procedure for [task, for example coupling a trailer]. Turn it into a five-minute toolbox talk for drivers and turnboys. Constraints: use only the steps in the procedure; do not add safety advice of your own; short sentences that a supervisor can read aloud; give a Kiswahili version that reads naturally; end with three questions the supervisor can ask to check understanding. Procedure: [paste]
"Do not add safety advice of your own" keeps the talk tied to what your safety officer approved. A fluent speaker should also read the Kiswahili before it is used.
Accounts: Excel is often the quickest win
Accounts staff do not need to paste your books into AI to benefit from it. Describe your columns, ask for the formula, and keep the figures in your own file.
I use Microsoft Excel. My sheet has columns for Truck Label, Date, Litres, Fuel Cost in KES and Kilometres Run. Give me formulas to calculate fuel cost per kilometre for each row and the monthly average per truck, and flag any row more than 20% above that truck's average. Explain each formula in one sentence. I am comfortable with simple formulas but not macros.
Test every formula on a few rows you have worked out by hand. If your company has Microsoft 365 Copilot licences, these tasks can happen inside Excel, but the prompt above works without one.
Coast hospitality: faster replies, still your voice
For hotels, AI is most useful on the messages that come in many times a week. The aim is a quicker reply that still sounds like your property. See our page on AI training for hospitality for how a session is built around this.
Sample prompt: replying to a conference enquiry
I work in sales at a beach hotel on the Kenyan coast. Draft a reply to the enquiry below from an NGO planning a three-day staff workshop in the low season. Context: use only these facts about us: [meeting room sizes, what is included, parking, Wi-Fi]. Constraints: under 150 words; answer each of their questions in order; no prices, because I will add the approved quote; do not promise availability or anything not listed; warm and professional. Customisation: end by asking for the number of delegates and their preferred dates. Enquiry: [paste, with the sender's name removed]
Sample prompt: answering a bad review
Below is a guest review of our hotel. Draft a public reply. Constraints: under 90 words; thank the guest; address the specific complaint; do not admit legal liability; do not mention any other guest, room number or staff member by name; invite them to contact [role and official email]; no copied phrases such as "we value your feedback". Customisation: we are a family-run property and our tone is friendly. Review: [paste]
Two cautions for hotels. AI will happily invent a pool, a shuttle or a late checkout if you let it, so give it your facts and tell it to use only those. And a reply posted without a human reading it can go badly wrong in public, so a person reads every one.
What must stay out of public AI tools
If you would not pin it on the noticeboard at the gate, do not paste it into a public AI tool. In Mombasa offices, that list includes:
- Shipper and consignee names linked to rates, and client price lists
- KRA PINs, import declaration details and bank details
- Driver ID numbers, licences, payroll and medical records
- Injury details from incident reports
- Guest passport scans, card details, health or dietary notes tied to a name
- Anything covered by a client confidentiality clause
This matters for practical reasons as well as legal ones. Google's help page for its consumer Gemini apps says a subset of chats may be seen by human reviewers and tells users not to enter confidential information they would not want a reviewer to see.
The Data Protection Act also applies. A March 2026 briefing on health data by the law firm Clyde & Co sets out core duties that apply more widely: register with the Data Commissioner where required, have a lawful basis, and report breaches to the Commissioner within 72 hours. It also says people should have human review and a way to contest automated decisions with significant effect. Transfers of personal data outside Kenya have conditions too, and the big public AI tools are run by companies based abroad.
This guide is general information, so take professional advice on customs or data questions about your own business.
The everyday fix is simple. Replace real details with labels such as [CLIENT A] and [DRIVER 2] before pasting. Then put the real details back yourself. Our guide to an AI usage policy for Kenyan organisations gives you a short policy to adapt.
Which tools to use, and the free versions
You can learn all of this for free. No single tool is the right one for every company.
- Microsoft Copilot Chat. Microsoft says this comes at no extra cost to business users with a qualifying Microsoft 365 subscription, signed in with a work account. Without a Copilot licence, the only company data it can use is the files you upload. Reaching emails, meetings and company files needs a paid licence. If your office runs on Microsoft 365, check what you already have.
- ChatGPT, Gemini and Claude. Each has a free version, as of November 2026, with usage limits. Each is fine for practising with non-sensitive text.
Ask two questions before staff use any tool at work. Is it approved here? And what happens to what we type?
What goes wrong, and how to avoid it
| What goes wrong | Why it happens in Mombasa offices | How to avoid it |
|---|---|---|
| A wrong tariff code is used | AI sounds sure, and old tariff versions are online | Never accept a code from AI; check the official tariff |
| A delivery date is promised | AI writes reassuringly | Tell it not to promise dates |
| A scan is misread | Poor photos of stamped documents | Check numbers against the original every time |
| A client's rates leak | Staff paste whole email threads | Mask first; paste only the part you need |
| A letter blames someone | Rough notes contain opinions | Tell AI to separate fact from report and not suggest fault |
| A hotel reply invents a feature | No facts were supplied | Give it your fact sheet; say "only these facts" |
| Staff use personal accounts in secret | There is no policy, so nobody asks | Write a one-page policy and name approved tools |
| Time saved is overstated | Checking time is ignored | Time the whole task, including corrections |
Training will not fix everything. AI cannot clear a congested yard, speed up a verification or repair a broken process. If a delay comes from a missing permit or a slow approval chain, fix that first.
A four-week start for one branch or team
Week 1: set the rules. Write a one-page policy: approved tools, what must never be pasted, and who signs off. Choose one task, such as client status updates. Time how long it takes now over ten examples.
Week 2: learn on real work. Train the team using masked examples from their own desks. Teach the four Cs we use: Clarity, Context, Constraints and Customisation. Save the prompts that work in a shared file.
Week 3: use it daily. Everyone uses the shared prompts. A supervisor checks a sample each day. Keep a log of every mistake AI made and how it was caught. That log becomes your best training material.
Week 4: measure. Time ten more examples, including checking. Ask clients or supervisors if quality held. Then decide whether to add a second task or stop.
Checklist before you roll out a tool
- [ ] We know where the data goes and whether it is used for model training
- [ ] We have turned off training on our data where the option exists
- [ ] Staff sign in with work accounts, not personal ones
- [ ] We have a masking method everyone understands
- [ ] Each output type has a named person who approves it
- [ ] We have a log for AI mistakes
- [ ] We have checked our client contracts for confidentiality terms
- [ ] We have asked our insurer or legal adviser about anything unusual
Questions people ask
Can AI do customs clearance in Mombasa?
No tool should make the declaration for you. AI can help prepare and compare documents, but a licensed agent must check every field. KRA's digest of court decisions shows that agents and owners can both be held liable for what is declared.
Is it safe to paste a bill of lading into ChatGPT?
Do not paste it unmasked into a public tool. It usually holds client names, addresses and cargo details. Replace those with labels first, or use a business tool your company has approved and checked.
Which job in a logistics company benefits first?
Customer service and operations coordinators who write the same updates all day usually see results first. Accounts staff who use Excel also gain quickly. Pick the task with the most repetition and the easiest checking.
Can small clearing firms afford this?
Yes, at the start. Free versions are enough to learn drafting, summarising and formula help, as long as sensitive data stays out. Pay for a tool only once you know which task it will improve. Our post on corporate AI training costs in Kenya covers how to budget for training.
Can AI write Kiswahili well enough for drivers and guests?
It often produces a usable draft, and quality varies by tool and topic. A fluent speaker should read anything that matters, especially safety instructions and public messages.
Should hotels let AI reply to guests automatically?
We advise against it for reviews and complaints. A person should read every public reply. Simple booking confirmations from your own system are a different matter, because they use fixed wording and your real data.
How long before staff get good at this?
Basic drafting and summarising can be learned in days. Good judgement about what to trust takes longer and comes from practice on real work. Nobody becomes an expert in a few days.
Sources
All sources last checked 9 November 2026.
- Xinhua via Hellenic Shipping News, "Kenya's main Mombasa port records 10.9-pct rise in cargo volumes in 2025" (2 February 2026)
- African Sustainability Matters, "Kenya's Port Transit Cargo grows 19.5% in 2025" (9 February 2026)
- The Standard, "Mombasa port cargo outperforms forecasts amid lingering congestion" (5 February 2026)
- The Eastleigh Voice, report on KNBS Gross County Product 2024 (15 January 2025)
- The Star, "August tops international arrivals as holiday travel drives tourism growth" (4 May 2026)
- The Standard, "Coast hotels depend on domestic tourists to remain afloat" (24 August 2026)
- Kenya Revenue Authority, E-Case Digest (customs agent and owner liability under EACCMA sections 147 and 148)
- Kenya Revenue Authority, "KRA reforms aimed at enhanced trade facilitation"
- Clyde & Co, "Health Data Protection in Kenya" (17 March 2026), for general Data Protection Act duties
- Google, Gemini Apps Privacy Hub
- Microsoft Support, "How Copilot Chat works with and without a Microsoft 365 Copilot license"
- Microsoft, Copilot Chat
- ChatGPT plans and pricing
- Claude plans and pricing
- Google Gemini subscriptions and free plan
Where FuKazee fits in
FuKazee is a Nairobi-based AI training company. We have no office in Mombasa, and we come on site by arrangement or teach live online. In October 2026 we trained two teams at A.O. Bayusuf and Sons in Mombasa, on site. Since 2022 we have trained more than 3,000 people.
We assume nobody in the room knows AI yet, and we avoid jargon. We show the free version of every tool, and we include Microsoft 365 and Copilot because so many offices run on them. Exercises come from the work your team does, with real personal and sensitive data masked first.
See how it works for your sector on AI training in Mombasa and AI training for logistics teams.
To talk about training for your team, 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.
