On this page
TL;DR:
- This guide is for in-house finance teams in Kenyan companies and NGOs: finance managers, accountants, credit controllers and payables officers.
- The most useful daily jobs are writing Excel formulas, matching M-PESA receipts to invoices, wording variance notes, building cash-flow templates and drafting payment reminders.
- You can get most of this help by sharing column headings and made-up rows. The real figures stay in your own workbook.
- Payroll, bank and M-PESA statements, KRA logins, supplier bank details and unpublished results stay out of public AI tools.
- AI has made payment fraud more convincing. The global overview in the Communications Authority's cyber security report for April to June 2026 describes a rise in deepfake-enabled fraud. Confirm every payment request by a second route.
Who this guide is for
This guide is for the people who close the books inside an organisation. You need basic Excel and no AI background.
Our separate guide on AI for accountants in Kenya covers eTIMS expense matching, ICPAK and client confidentiality in a practice. This one covers the work of a finance office: collections, payables, month-end, forecasts and board papers. It is general guidance, and it does not replace your tax adviser, auditor or lawyer.
What do finance teams actually use AI for?
Finance teams get the most from AI on work they can check against something they already hold. A formula can be tested. A draft reminder can be read. A cash-flow template can be walked through line by line.
| When | Task | How AI helps | What a person still does |
|---|---|---|---|
| Daily | M-PESA and bank receipts | Writes the matching formulas | Clears exceptions and unidentified receipts |
| Daily | Payment reminders | Drafts polite WhatsApp messages and emails | Confirms the balance and any dispute |
| Weekly | Cash-flow forecast | Builds the template and lists what is missing | Supplies figures and judges timing |
| Weekly | Payables | Drafts supplier requests, such as a valid electronic invoice | Approves every payment |
| Month-end | Workbook checks | Explains inherited formulas and lists likely errors | Tests on known answers |
| Month-end | Variance notes | Words the cause you supply | Finds the real cause |
| Quarterly | Board papers | Lists the questions a board member may ask | Answers them and owns the paper |
| Any time | Finance policies | Turns a policy into plain questions and answers | Confirms each answer is correct |
AI should not approve payments, decide a tax treatment, or produce a figure nobody has recalculated. Microsoft itself tells Excel users to avoid Copilot for decisions in sensitive areas, and it names finance as one of them.
What is different about finance work in Kenya?
Four things shape how a Kenyan finance team should use these tools: mobile money, electronic invoicing, fixed monthly tax dates and foreign-hosted software. Generic AI advice usually skips them.
M-PESA sits beside the bank. Customers pay by PayBill or till, often with a reference they typed themselves. One payment may cover several invoices, and one invoice may arrive in several payments. M-PESA charges and transfers to the bank need their own treatment.
Most expenses now need an electronic invoice. KRA's public notice of 8 June 2026 says the allowance for expenses without eTIMS or TIMS invoices covers 2025 returns only. From the 2026 year of income, it says all declared income and expenses must be supported by valid electronic tax invoices. The law does exempt some items. KPMG's 2026 note lists salaries taxed through PAYE, imports and airline passenger tickets, among others, so check any exception with your tax adviser. For a payables team, that makes the supplier's invoice a control point. Our accountants guide has a full reconciliation method.
KRA is asking for more. Tech-ish reported on 4 September 2026 that KRA told businesses using TIMS and eTIMS to keep accurate stock records in those systems. The Star reported in May 2026 that KRA plans to link eTIMS with M-PESA. That link was a plan when reported, so check its status before you budget around it.
Tax dates fall in the middle of the month. KRA's pages say PAYE is due by the 9th and VAT by the 20th of the following month. Those dates belong in any forecast.
Most AI tools run abroad. Section 48 of the Data Protection Act sets conditions for sending personal data out of Kenya. Section 49 adds consent and safeguards for sensitive personal data. Payroll files can hold both kinds.
How do you match M-PESA receipts without pasting the statement?
Send the shape of your data and keep the statement in your own file. The AI tool only needs your column headings and the rule you want.
M-PESA business statements can be downloaded as Excel or CSV files, according to a guide for Kenyan businesses from Tally, which sells accounting software. Column names can differ by account and download type, so use the headings on your own file.
Suppose your workbook has two sheets. All details below are invented.
- MPESA: A Receipt code, B Date, C Amount, D Account reference.
- Invoices: A Invoice number, B Customer, C Amount due.
Prompt 1: headings only
I use Microsoft Excel. Sheet "MPESA" has columns A Receipt code, B Date, C Amount, D Account reference. Sheet "Invoices" has columns A Invoice number, B Customer, C Amount due. Customers type the invoice number as the account reference, and one invoice can be paid in several parts. Give me: (1) a formula for D2 on Invoices that totals the receipts for that invoice; (2) a formula for E2 on Invoices showing "Unpaid", "Part paid", "Paid" (within KES 1) or "Overpaid"; (3) a formula for E2 on MPESA showing "Duplicate code", "No matching invoice" or "Matched". Explain each part in one sentence. List three ways these formulas could give a wrong answer. Do not assume anything I have not said.
You should get something close to this:
Invoices D2: =SUMIFS(MPESA!$C:$C,MPESA!$D:$D,A2)
Invoices E2: =IF(D2=0,"Unpaid",IF(ABS(D2-C2)<=1,"Paid",IF(D2<C2,"Part paid","Overpaid")))
MPESA E2: =IF(COUNTIF($A:$A,A2)>1,"Duplicate code",IF(COUNTIF(Invoices!$A:$A,D2)=0,"No matching invoice","Matched"))
Here is what the result might look like, using made-up figures.
MPESA sheet
| Receipt code | Account reference | Amount (KES) | Result |
|---|---|---|---|
| RCPT01 | INV-201 | 45,000 | Matched |
| RCPT02 | INV-202 | 40,000 | Matched |
| RCPT03 | INV-202 | 30,000 | Matched |
| RCPT04 | INV-204 | 33,000 | Matched |
| RCPT05 | invoice 203 | 18,500 | No matching invoice |
| RCPT06 | John | 7,200 | No matching invoice |
| Total | 173,700 |
Invoices sheet
| Invoice | Amount due (KES) | Received (KES) | Status |
|---|---|---|---|
| INV-201 | 45,000 | 45,000 | Paid |
| INV-202 | 120,000 | 70,000 | Part paid |
| INV-203 | 18,500 | 0 | Unpaid |
| INV-204 | 32,000 | 33,000 | Overpaid |
| Total | 215,500 | 148,000 |
Now do the check that matters. Matched receipts of KES 148,000 plus unmatched receipts of KES 25,700 equal the statement total of KES 173,700. If those do not agree, stop and find out why.
Look at what the formulas cannot do. INV-203 shows as unpaid, yet RCPT05 is probably that customer's payment with a badly typed reference. Someone must look at it, confirm it and fix the reference. RCPT06 goes to a suspense account until somebody traces it. Tally suggests tracing such amounts within about 30 days.
Where this method can fail
- A duplicated receipt code gets counted twice in the invoice total. Clear duplicates first.
- Spaces and different spellings ("INV 201", "invoice 201") will not match. These formulas ignore capital letters, but a missing hyphen or an extra space breaks the match. Ask the AI tool for a helper column that cleans the references.
- One payment covering several invoices shows as "No matching invoice". A person has to split it.
- M-PESA charges belong in their own ledger, and transfers from M-PESA to the bank are transfers, not sales.
- A customer can show you a fake payment message. Rely on your statement.
Ask customers to use the invoice number as the account reference on your PayBill, as Tally also suggests. That should reduce the number of exceptions.
How can AI help with variance notes and board papers?
AI can word a variance note well, but it cannot know why the figure moved. Your team finds the cause. The tool helps you say it clearly and prepare for the questions.
Here is an invented management account.
| Line | Budget (KES) | Actual (KES) | Variance (KES) | Variance |
|---|---|---|---|---|
| Staff costs | 4,200,000 | 4,560,000 | 360,000 | 8.6% |
| Marketing | 900,000 | 640,000 | (260,000) | (28.9%) |
| Fuel and transport | 520,000 | 610,000 | 90,000 | 17.3% |
| Bank and M-PESA charges | 180,000 | 214,000 | 34,000 | 18.9% |
| Total | 5,800,000 | 6,024,000 | 224,000 | 3.9% |
If your real figures are confidential, you can still use this approach. Change customer and supplier names to letters. Then divide every figure by one secret number that you do not share. Percentages and patterns stay the same, and the amounts mean nothing to an outsider. Afterwards, go back to your own workbook and restate every figure from it.
Prompt 2: board questions
You are a careful non-executive board member at a Kenyan company with no finance background. Below is a one-page summary of budget against actual. Do not comment on the arithmetic. List the eight questions you would ask, starting with the one that matters most. For each, say which line prompted it. Then list any line where the summary gives no reason for a large movement. Summary: [paste scaled figures with names removed]
Prompt 3: variance note from your own causes
Write the variance commentary for the lines below, for a board paper. Use only the causes in my notes. If a line has no cause in my notes, write [CAUSE NEEDED] and move on. Do not add reasons, forecasts or recommendations. Maximum two sentences per line. Plain English for readers who are not accountants. Notes: [line, amount, cause in your own words]
The [CAUSE NEEDED] marker matters. AI tools like to fill a gap with a tidy reason, and an invented reason in a board paper is a serious error.
How can AI help with a cash-flow forecast?
Use AI to build and test the structure of a forecast, then enter the numbers yourself. The tool is good at remembering lines you might forget. It knows nothing about when your biggest customer really pays.
Prompt 4: a 13-week template
Build the layout for a 13-week cash-flow forecast in Excel for a Kenyan company with [number] staff in [sector]. Context: customers pay by bank transfer and M-PESA; suppliers are paid by bank transfer. Constraints: give column headings for weeks 1 to 13 and a list of rows for opening balance, receipts by type, payments by type and closing balance. Include separate rows for payroll, PAYE due by the 9th and VAT due by the 20th. Give formulas for totals and closing balance. Do not invent any amount. Customisation: add a row showing the lowest closing balance in the 13 weeks, and a list of ten questions I should answer before trusting it.
Then ask a follow-up using invented or scaled numbers: "If receipts in weeks 4 to 7 arrive 30 days late, which rows change, and what formula would show the effect?" You get the method, and you run the scenario in your own file.
A forecast only helps if somebody compares it with what happened. Each week, put actual receipts beside the forecast and note the reason for the largest gap.
Which other finance tasks are worth trying?
Three more jobs can save time and carry little risk when you remove the names.
Prompt 5: asking suppliers for a valid invoice
Write a short, polite WhatsApp message and a short email from a company's payables team to suppliers. Purpose: remind them that we need a valid electronic tax invoice for every purchase, showing our KRA PIN, before we process payment. Explain in one plain sentence that KRA now expects expenses to be supported by these invoices. Do not state any tax rate, penalty or legal section. Keep the WhatsApp version under 80 words. Leave [contact person's title] and [our PIN] for me to fill in.
Have your tax adviser confirm the wording of any new payment condition, and check your contracts before you apply it.
Prompt 6: understanding an inherited workbook
Below are formulas from a spreadsheet someone else built. I am pasting the formulas and headings only. Explain in plain English what each one does. Then list anything that looks risky: hard-coded numbers, ranges that stop too early, references to other files, and errors being hidden with IFERROR. Do not rewrite anything yet. Formulas: [paste formula text]
Prompt 7: a finance policy in plain questions
Below is our staff expense policy. Turn it into a question-and-answer sheet for staff who will read it on their phones. Use only what the policy says. If staff are likely to ask something the policy does not answer, list it under "Gaps for finance to decide". Do not suggest what the answer should be. Policy: [paste policy with names removed]
The gaps list shows finance what the policy still needs to settle.
Which tools should a finance team use?
Start with whatever your organisation has approved. If you already pay for Microsoft 365, Copilot Chat is the sensible first place to look, since finance work already lives in Excel. For practice on invented data, every tool below has a free route or comes with a plan you may already hold.
| Tool, as of November 2026 | Cost and access | What the provider says |
|---|---|---|
| Microsoft Copilot Chat, work account | No additional cost with an eligible Microsoft 365 subscription | Enterprise data protection applies, shown by a green shield. Prompts are not used to train foundation models. Prompts are logged for audit |
| Copilot in Excel | Needs an eligible plan. Microsoft's FAQ lists a commercial Copilot subscription and Copilot Chat-eligible business plans, among others. Its Copilot Chat page says availability in Microsoft 365 apps varies by market and licensing | Its editing features work only when Calculation Options are set to Automatic. Review everything it does |
| Microsoft Copilot add-on | Listed at US$30 per user per month billed yearly. The Business add-on, for up to 300 users, is listed at US$21, or US$18 under a promotion from 1 July to 31 December 2026 for existing Microsoft 365 Business customers. Both need a qualifying Microsoft 365 plan | Adds work-data-grounded chat and in-app features |
| Free Microsoft Copilot | Free at copilot.cloud.microsoft | A chat tool. Microsoft's comparison does not list Word, Excel or Outlook for it |
| ChatGPT Free | Free, with limits on uploads | Engadget reported in August 2026 that free accounts are enrolled in model training by default. You can turn this off under Settings, Data Controls |
| Gemini, personal account | KES 0 on the free plan | A subset of chats are read by human reviewers, and reviewed chats can be kept for up to three years. Google asks you not to enter confidential information |
| Gemini, Google Workspace account | Depends on your Workspace plan | Google says content is not human reviewed or used for training outside your domain without permission |
| Claude Free | Free, with usage limits | Anthropic says chats are used to improve models if you allow it, so check Model Improvement in privacy settings. Conversations flagged for safety review can also be used. Incognito chats are excluded from training |
Work accounts are a stronger place to start, but they still need a decision from your organisation. Microsoft says web queries from Copilot Chat go to Bing under separate terms, and that AI processing can reach other regions when demand is high. Ask your IT team what is switched on, and read our guide to Microsoft Copilot training for teams in Kenya for the wider picture.
What must never go into a public AI tool?
Keep anything that identifies a person, reveals your bank position or gives access to a system out of public AI tools. Finance data can be commercially secret even when it names nobody.
Never paste:
- Payroll, salary schedules, P9 forms and staff bank details
- Bank statements and M-PESA statements, which can show payers' names
- iTax, eCitizen, bank and M-PESA portal logins, tokens and one-time codes
- KRA PINs and ID numbers of staff, customers or suppliers
- Customer and supplier ledgers with names attached
- Unpublished results, forecasts and board papers
- Tender prices, bids and contract terms before they are public
- Details of a fraud under investigation or a staff disciplinary case
| Data | Where it can go |
|---|---|
| Public KRA notices and published accounts | Any tool |
| Column headings and invented rows | Any tool |
| Formula text with no data in it | Any tool, after checking for names in sheet titles |
| Scaled figures with names removed | A tool your organisation has approved |
| Real ledgers and management accounts | Only a work account your organisation has approved and assessed |
| Passwords, tokens, one-time codes | Never |
The law applies to the personal data in these files. Section 63 of the Data Protection Act lets the Data Commissioner impose a penalty of up to KES 5 million, or 1% of an undertaking's turnover in the previous year, whichever is lower. Section 65 lets a person claim compensation for damage. Our guide to the Data Protection Act and AI explains the duties in more detail.
If you are building something bigger, note the regulator's direction. PPC Land reports that the ODPC's July 2026 draft Guidance Note on AI lists credit scoring for loan decisions and fraud detection in payment systems among the uses needing a Data Protection Impact Assessment. It was a draft when reported. Digital Policy Alert shows the consultation closed on 17 August 2026 and lists no final version. We could not open the ODPC's own copy.
How has AI changed payment fraud?
Criminals now use AI to produce convincing emails, voices and video, so a request that looks and sounds right proves very little. Every finance team needs a checking routine that does not depend on how real the request seems.
The cyber security report for April to June 2026, from the National KE-CIRT/CC at the Communications Authority, describes a rise in deepfake-enabled fraud and impersonation aimed at government, financial organisations and businesses. This is in the report's global threat overview, and that section gives no count of Kenyan cases. It lists AI-written emails, fake login pages, deepfake videos and voice impersonation used to get people to authorise fraudulent transactions. Kenya's national cyber response team advised organisations to verify sensitive requests independently before approving transactions.
The best-known case happened in Hong Kong. In January 2024 a finance employee at the engineering group Arup was deceived by a fake version of the company's chief financial officer and other staff. The employee made 15 transfers totalling HK$200 million, about US$25.6 million. Arup confirmed that fake voices and images were used.
A short routine covers most of this risk:
- Confirm any new or changed supplier bank or M-PESA details by calling a number you already hold. Never use the number in the message.
- Require two people to approve payments, with one of them outside the chain of the person asking.
- Treat urgency and secrecy as warning signs, especially from senior people.
- Agree that no payment is approved on the strength of a WhatsApp message, an email or a video call alone.
- Check receipts on your own M-PESA or bank statement. Never rely on a customer's screenshot or forwarded message.
- Keep a record of each check and who did it.
Run a surprise test once a quarter and see who follows the routine.
What goes wrong when finance teams use AI?
The usual failures are small and easy to miss, because the output looks tidy.
| What goes wrong | What it looks like | How to avoid it |
|---|---|---|
| A total taken from chat | The sum looks right and is wrong | Recalculate in Excel, always |
| An invented reason | A variance note gives a cause nobody confirmed | Use only causes you supplied |
| Old or foreign rules | A tax rule from another country or an earlier year | Check KRA's current notice |
| Silent workbook changes | Rows sorted or a formula overwritten by Copilot | Work on a copy and compare with the original |
| Formula that works on the sample only | It breaks on a blank, a space or a text number | Test on five rows you worked out by hand |
| Copied data in chat history | Old chats hold customer balances | Practise on invented data and clear chats under your policy |
| Nobody reviews | "The AI did it" | Same review as work from a junior |
How should a finance team start?
Run one pilot through one month-end, on one task. Choose something repetitive that you can check, such as M-PESA matching or variance notes.
Before month-end. Confirm which tools and accounts are approved. Write a one-page rule covering the never-paste list and who signs off. Our guide to an AI usage policy for Kenyan organisations shows what to include. Time the task as it stands today.
During month-end. Run the old method and the new one side by side. Compare results. Log every error and how it was caught.
After month-end. Time the new method again and include checking. Decide whether to keep it. Add the prompts that worked to a shared library.
Where training will not help. It will not fix a chart of accounts that makes no sense or collect missing supplier invoices. You still need someone who understands the business behind the figures.
Questions people ask
Can a finance team in Kenya use ChatGPT at work?
We found no Kenyan rule that bans it. The Data Protection Act applies to any personal data you enter, and your employer's confidentiality rules apply too. Use a free tool for formulas, public notices and invented data, and keep real records in an approved work account.
Can AI reconcile our M-PESA statement?
It can write and explain the matching logic very well. Do the matching itself in Excel, your accounting system or an approved work tool. A person must clear the exceptions, because a badly typed reference needs judgement.
Does Copilot work in Excel for free?
Free Microsoft Copilot is a separate chat tool, and Microsoft's comparison does not list Excel for it. As of November 2026, Copilot in Excel needs an eligible plan, and availability depends on your licence and your administrator. Copilot Chat with a work account still lets you ask for formula help in chat.
Is it safe to upload a management account to Copilot Chat?
Only if your organisation has approved it. Microsoft says Copilot Chat with a work account has enterprise data protection and is not used to train foundation models. Your organisation should still decide what belongs there, and you should look for the green shield before you upload anything.
Can AI catch fraud in our payments?
In our view a chat tool is a poor fraud detector for live payments. It can suggest tests to run on your data, such as duplicate amounts or new suppliers paid just below an approval limit. Your first protection remains two-person approval and calling back on a number you already hold.
Will AI replace finance teams?
Parts of the work will shrink, especially first drafts and routine matching. Someone still has to answer to the board, the auditors and KRA, and decide what to do about exceptions. People who can check AI's work will stay useful.
How long does it take a finance team to learn this?
A focused session can establish safe habits and a few repeatable tasks. Practice over the next two or three month-ends is what makes them stick. Nobody becomes an expert in a few days.
Sources
- Kenya Revenue Authority, public notice: Filing of 2025 Income Tax Returns (8 June 2026) (last checked 10 November 2026)
- Kenya Revenue Authority, PAYE (last checked 10 November 2026)
- Kenya Revenue Authority, Value Added Tax (last checked 10 November 2026)
- Tech-ish, KRA orders businesses to keep stock records in TIMS and eTIMS (4 September 2026) (last checked 10 November 2026)
- The Star, KRA to link real-time tax systems to M-Pesa (May 2026) (last checked 10 November 2026)
- KPMG Kenya, Tax Alert: eTIMS and the Shift to Data-Driven Income and Expense Validation (January 2026) (last checked 10 November 2026)
- Tally Solutions, How to reconcile M-Pesa and bank payments with your business accounts (last checked 10 November 2026)
- Data Protection Act, No. 24 of 2019 (ODPC copy) (last checked 10 November 2026)
- PPC Land, report on the ODPC draft AI guidance and its DPIA annex (3 August 2026) (last checked 10 November 2026)
- Digital Policy Alert, ODPC consultation on the draft Guidance Note on Artificial Intelligence (last checked 10 November 2026)
- Communications Authority of Kenya, National KE-CIRT/CC Cybersecurity Report, April to June 2026 (last checked 10 November 2026)
- CFO.com, company defrauded by deepfake CFO identified as Arup (17 May 2024) (last checked 10 November 2026)
- Microsoft Support, Frequently asked questions about Copilot in Excel (last checked 10 November 2026)
- Microsoft Support, difference between Microsoft Copilot (free) and Copilot in Microsoft 365 (last checked 10 November 2026)
- Microsoft, Copilot Chat plans and availability (last checked 10 November 2026)
- Microsoft Learn, Microsoft Copilot Chat Privacy and Protections (last checked 10 November 2026)
- Engadget, ChatGPT: what's free in 2026 and what isn't? (August 2026) (last checked 10 November 2026)
- ChatGPT plans and pricing (last checked 10 November 2026)
- Google, Gemini Apps Privacy Hub (last checked 10 November 2026)
- Google, Gemini subscriptions (last checked 10 November 2026)
- Google Workspace, Generative AI Privacy Hub (last checked 10 November 2026)
- Anthropic, Is my data used for model training? (last checked 10 November 2026)
- Claude plans and pricing (last checked 10 November 2026)
Where FuKazee fits in
FuKazee has trained more than 3,000 people, and we build AI training for teams in Kenya around the work in front of them. For finance teams, that means your own month-end tasks, practised on invented or scaled data so nothing sensitive enters the room.
We assume nobody knows AI yet and we avoid jargon. We show the free version of each feature, we do not push one tool, and we spend real time on Microsoft 365, Copilot and Excel. We teach the four Cs (Clarity, Context, Constraints, Customisation) first, then how to check every output. See what a programme covers on our page for AI training for finance teams in Kenya, or look at Excel and AI training in Nairobi if formulas are your main need.
To talk through what your finance team needs, 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.
