On this page
TL;DR:
- AI is useful for the work around a tender: drafting terms of reference, tightening specifications, building blank scoring sheets, planning supplier checks and summarising signed contracts.
- The middle of a tender is off limits. Bids, quotations, evaluation scores and committee discussions never go into a public AI tool. For public bodies, section 67 of the Public Procurement and Asset Disposal Act, 2015 bars disclosing them.
- People decide. Under section 80, the evaluation committee evaluates bids using the criteria in the tender documents, and every member signs the report. AI does not score, rank or recommend a winner.
- AI is often wrong about Kenyan procurement law and prices. It can invent a section number or a market rate. Check the Act, your manual and real quotations.
- Private companies are not bound by the Act, but the same habits protect them: suppliers trust you with their prices, and the Data Protection Act covers personal data inside bids.
Who this guide is for
This guide is for people who buy things for an organisation in Kenya: procurement and supply chain officers in ministries, counties, state corporations and schools, buyers in private companies, NGO procurement staff, and the user departments who write the first draft of a requirement. It also helps anyone who sits on an evaluation committee.
You need no technical background. If you are looking for training for a whole team, see our page on AI training for procurement teams in Kenya. This post is the hands-on guide: what to do, what to type and where to stop.
The simple rule: AI helps before bids arrive and after the award
Think of a tender as three zones. AI helps in the first and last. The middle belongs to people only.
| Stage | Where AI can help | What stays human only |
|---|---|---|
| Planning and needs | Turning a user department's notes into a clear statement of need | Approving the budget and the method |
| Specifications and terms of reference | Finding vague words, gaps and brand names | Deciding what the organisation truly needs |
| Evaluation criteria | Turning loose phrases into measurable criteria; building a blank scoring sheet | Approving the criteria in the tender document |
| Opening and evaluation | Nothing. Keep bids, prices and scores out of AI tools | Opening, evaluating, comparing and signing |
| Due diligence | Listing what to check; drafting reference-call questions | Running the checks on official registers |
| Contract management | Turning a signed contract into an obligations table; drafting delivery reminders | Signing, varying and terminating contracts |
What the law says, in plain words
Public procurement in Kenya rests on Article 227 of the Constitution, which requires procurement by State organs and public entities to be fair, equitable, transparent, competitive and cost-effective. The Public Procurement and Asset Disposal Act, 2015 (PPADA) puts that into practice, and section 4 applies it to all State organs and public entities.
The section numbers below come from the Kenya Law revised edition of 2016. The Act was amended by the Finance Act, 2017 and the Public Procurement and Asset Disposal (Amendment) Act, 2022. We compared each section below with the Kenya Law text as at 31 December 2022, and the points we rely on did not change. Parliament has looked at further amendment Bills since then, so check the current text on Kenya Law before you rely on any detail.
| Section | What it says | What it means for AI |
|---|---|---|
| 60 | Requirements must be clear, complete and allow fair and open competition. They must not name a trademark, producer or origin unless there is no other precise way, and then must allow equivalents | AI is good at spotting restrictive wording. It also copies manufacturer language, so check it has not added new bias |
| 66 | Bans corrupt, collusive and fraudulent practice and conflicts of interest. A person with a conflict must disclose it and stay out | AI cannot spot your conflicts. That duty stays with you |
| 67 | Bars disclosure of the contents of tenders and of information on their evaluation, comparison or clarification, with listed exceptions. Staff sign a confidentiality declaration | Pasting a bid into a public AI tool shares it with an outside company. The listed exceptions do not cover that |
| 68 | Records for each procurement are kept for at least six years after the contract ends | An AI draft is not the record. File the approved version the normal way |
| 80 | The evaluation committee evaluates responsive tenders using the criteria in the tender documents. Criteria should be objective and quantifiable where possible. Each member signs the report | AI can help write criteria before bids arrive. It never does the evaluating |
The penalties are heavy. Section 176 sets a fine of up to KES 4 million, up to ten years in prison, or both, for an individual who divulges confidential information or inappropriately influences an evaluation. Section 67(5) adds a ten-year bar from working for a government entity or a company where the government holds shares. No faster summary is worth that.
Private companies are not bound by the Act. Even so, your tender letters usually promise confidentiality, your auditors expect a fair and documented process, and bids often contain personal data such as CVs, directors' ID numbers and phone numbers. The Data Protection Act applies to that data everywhere. Our guide on using ChatGPT under Kenya's Data Protection Act explains those duties.
Official AI guidance. As of 11 November 2026, we found no circular from the Public Procurement Regulatory Authority (PPRA) on staff use of AI tools. The Data Protection Commissioner's draft Guidance Note on Artificial Intelligence, dated July 2026, went out for public comment in August 2026. Check whether a final version has been issued. Our guide to AI in Kenya's public sector sets out where the wider AI policy and the AI Bill stand.
What you can paste, and where
Use a traffic-light rule.
| Colour | Examples | Where it can go |
|---|---|---|
| Green | Published tender notices, public laws, your own blank templates, general questions ("what should a cleaning contract cover?") | Any tool your organisation allows |
| Amber | A draft specification or terms of reference before it is advertised, with the budget and supplier names removed | Only a work tool your ICT team has approved, signed in with your work account |
| Red | Bids, quotations, the user department's cost estimate, evaluation scores, committee minutes, due diligence reports, personal data from bids | No AI tool, unless your organisation has approved a specific tool for that purpose in writing |
Why is a draft specification amber? Section 67(1) also covers information whose disclosure would inhibit fair competition. A leaked draft gives one supplier a head start.
What happens to what you type
No single tool is the right one. Use what your organisation approves. As of November 2026:
- Microsoft Copilot Chat (recently renamed from Microsoft 365 Copilot Chat) offers enterprise data protection when you sign in with a work account, shown by a green shield. Microsoft says prompts and responses are not used to train the underlying models, and they are logged for audit.
- Google Gemini. Google says that with Keep Activity on, chats help improve its services and a subset is read by human reviewers. Its advice: "Please don't enter confidential information that you wouldn't want a reviewer to see."
- ChatGPT and Claude have free tiers that draft and summarise well. Consumer accounts on several tools can use your chats to improve their models, depending on your settings. Claude, for example, has a "Model Improvement" setting you can switch off. Check the current settings before you type anything work-related.
If your office cannot send documents to tools hosted abroad, an open-source model on your own computer is another route. See how to run open-source AI models locally.
Six procurement jobs, with prompts you can copy
Each prompt uses the four Cs we teach: Clarity (what you want), Context (the background), Constraints (limits and what to leave out) and Customisation (who will read it). Replace the square brackets with your own details, and remove anything red first.
1. Draft terms of reference for a consultancy
Terms of reference often fail because the deliverables are vague. Section 117 asks that they include expected milestones or performance benchmarks, with timelines.
I am a procurement officer in a Kenyan [public entity / company]. Draft terms of reference for a consultancy to [purpose, for example "review our records management system"]. Context: the user department's notes are below, with names and the budget removed. Constraints: use the headings Background, Objectives, Scope of Work, Deliverables, Milestones and Timelines, Reporting, Required Qualifications, What the Client Will Provide. Make every deliverable specific and checkable. Put milestones in a table with week numbers. No budget figure. Do not cite any law. Customisation: readers are consultants deciding whether to bid, so use plain English. End with five questions the user department must answer before this is advertised. Notes: [paste]
The closing questions are the most useful part. They show what the user department has not yet decided.
2. Stress-test a specification before it is advertised
Section 60(3) asks technical requirements, where appropriate, to consider standards, the life of the item, servicing and maintenance, and disposal. These are the items people forget.
Review this draft specification for [item, for example "a standby generator for a county office"]. It has not been advertised, and supplier names and the budget are removed. Give me a table with three columns: Issue, Why it matters, Suggested rewrite. Look for (1) vague words such as "good quality" or "reputable", (2) any requirement only one maker could meet, (3) missing items: standards, service life, warranty, spare parts, user training, delivery and installation, and disposal. Do not suggest prices or brands. The reader is the head of procurement. Specification: [paste]
Our procurement training page has a separate prompt for removing brand names. This one checks for gaps.
3. Build evaluation criteria and a blank scoring sheet
Do this before any bid arrives. Once bids are in, the criteria in the tender document are fixed.
Turn these draft evaluation criteria into criteria that can be scored objectively, for a tender for [service, for example "office cleaning for three buildings"]. Draft criteria: [paste, for example "relevant experience, adequate staff, good methodology"]. (1) Split them into a preliminary stage (pass or fail documents), a technical stage (scored) and a financial stage. (2) For each technical criterion, state the evidence the bidder must provide and exactly how marks are given. (3) Suggest a technical pass mark and explain the trade-off. Do not refer to any bidder. Present it as a table for Word, then give Excel column headings for a blank scoring sheet.
Your head of procurement checks the result against your standard tender document and the Regulations.
4. Comparing bids: what AI can and cannot do
In a public body, AI does not touch bid contents or scores. What it can do is build the empty workbook the committee will fill in by hand, on an office computer, with no AI.
I need an Excel workbook an evaluation committee will fill in by hand, for up to [number] bidders labelled Bidder 1, Bidder 2 and so on. The approved technical criteria and marks are: [paste]. Give me the sheet layout, the formula for each bidder's total, a formula that shows "Qualifies" or "Does not qualify" against the pass mark in cell [cell], and a check that flags any score above the maximum for a criterion. Explain each formula in one sentence. I know basic Excel.
In a private company with an approved enterprise tool, your policy may allow AI to tabulate quotations. Rename suppliers to A, B and C first. Ask only for figures in a table, never a recommendation. Then check every number against the original, because one wrong digit changes the outcome. Our Excel and AI training in Nairobi covers this kind of sheet.
5. Plan supplier due diligence
AI should never be your source of facts about a real company. It can invent directors, awards and court cases. Use it to plan the checks, then run them on official sources.
I am preparing due diligence on a shortlisted supplier for [category, for example "medical consumables"] in Kenya. Do not tell me anything about any specific company. (1) Give me a checklist of what to verify, grouped into registration, tax, past performance, capacity and conflicts of interest. (2) Next to each item, name the document or official source that would prove it. (3) Write eight open questions for reference calls with the supplier's past clients.
Three official checks every Kenyan buyer should know:
- Company registration. An official search (formerly called a CR12) on the Business Registration Service through eCitizen gives you a computer-generated certificate of what the registry holds on a company. Check it against what the bid declares.
- Tax compliance. KRA says Tax Compliance Certificates are valid for twelve months only and can be checked with the Certificate Checker on the iTax portal. A PDF on its own proves little.
- Debarment. PPRA publishes a list of debarred firms on its website, with the date and length of each debarment. Treat it as one check among several, not proof on its own.
6. Turn a signed contract into an obligations table
After signing, most of the risk is in missed dates: delivery, inspection, payment, warranty expiry, renewal notice. AI is good at pulling these out.
Below is a signed supply contract. The supplier's name is replaced with "Supplier", and the value and personal details are removed. Turn it into an obligations register with columns: Clause number, Obligation, Who must act, Deadline or trigger, Consequence if missed. Then list every date and notice period. Quote clause numbers exactly. If a clause is unclear, write [CHECK] and do not guess. No legal advice. The readers are the contract manager and the user department. Contract: [paste]
Use a tool your organisation has approved for contracts. Read every clause the table points to, and take anything marked [CHECK] to your legal officer.
Where AI goes wrong in procurement
| What goes wrong | Example | How to avoid it |
|---|---|---|
| Invented law | AI quotes a section that does not exist, or a rule from the Public Procurement and Disposal Act, 2005, which the 2015 Act repealed | Ask AI not to cite law. Look up every section yourself |
| Invented prices | AI offers a "typical Nairobi price" for laptops or diesel | Prices come from real quotations and past contracts, never from AI |
| Hidden brand bias | A "neutral" specification copied from one maker's datasheet | Ask: could at least three makers meet this? If not, rewrite |
| "Just to see" scoring | Someone pastes two quotations to ask which is better | Treat it as a breach. Red data never goes in |
| No owner | An AI-drafted document goes out unchecked | A named officer approves every document before it leaves the office |
If your organisation has no AI rules yet, start with a short one for the procurement unit: approved tools, the traffic-light list, no AI on bids or scores, every legal reference checked, and a named officer who signs off. Our AI acceptable use policy template for Kenya gives you the full version.
A note on e-GP
Presenting the 2026/27 Budget Statement on 11 June 2026, the Treasury Cabinet Secretary said all public entities must procure only through the Government Electronic Procurement (e-GP) system from 1 July 2026. Follow the latest PPRA circular. Bids on e-GP are still confidential, and downloading them into an AI tool is still disclosure.
Questions people ask
Can I use ChatGPT to evaluate tenders in Kenya?
No. In a public body, section 80 of the PPADA gives evaluation to the evaluation committee, and section 67 bars disclosure of tender contents and evaluation information. AI can help you prepare criteria and a blank scoring sheet before bids arrive.
Can AI write a tender document for me?
It can draft parts, such as the terms of reference or the scope of work. Public bodies should start from the standard tender documents and their manual, then use AI to improve the wording. A procurement officer approves the final version.
Is it safe to paste a draft specification into AI?
Only into a work tool your organisation has approved, after removing the budget and supplier names. A draft that leaks before advertising can give one supplier an unfair head start.
Does AI know the Public Procurement and Asset Disposal Act?
Only partly. It may quote the repealed 2005 Act, miss the 2017 and 2022 amendments, or invent section numbers with full confidence. Treat every legal reference from AI as a lead to check.
What is the fastest safe win for a procurement officer?
Contract obligations tables and terms of reference. Both are repetitive, easy to check against the source, and save real time when done with masked documents in an approved tool.
Sources
All sources last checked 11 November 2026.
- Kenya Law, Public Procurement and Asset Disposal Act, Cap. 412C (consolidated text as at 31 December 2022)
- Kenya Law PDF of Cap. 412C as at 31 December 2022 (FAOLEX copy), used to check sections 4, 60, 66, 67, 68, 80, 117 and 176 against the 2017 and 2022 amendments
- Public Procurement and Asset Disposal Act, 2015, Kenya Law revised edition 2016 (County Assemblies Forum repository copy), used for sections 4, 60, 66, 67, 68, 80, 117, 176 and 182
- Mzalendo, Public Procurement and Asset Disposal (Amendment) Bill, 2024 (National Assembly Bill No. 48 of 2024)
- Constitute Project, Constitution of Kenya 2010, Article 227
- Kenya Law, Public Procurement and Asset Disposal Regulations, 2020
- Microsoft Learn, "Microsoft Copilot Chat Privacy and Protections"
- Google, Gemini Apps Privacy Hub
- Anthropic, "Is my data used for model training?"
- Digital Policy Alert, "Office of the Data Protection Commissioner opened consultation on draft Guidance Note on Artificial Intelligence"
- Kenya Revenue Authority, "Learn About Compliance"
- Business Registration Service, Frequently Asked Questions
- Business Registration Service, "How to apply for an official search (CR12)"
- Public Procurement Regulatory Authority, "Debarred Firms"
- The Star, "All government procurement to go through e-GP from July 1" (11 June 2026)
- Eastleigh Voice, "Treasury CS Mbadi announces end of manual procurement with full shift to eGP system" (11 June 2026)
Where FuKazee fits in
FuKazee trains teams in practical AI for the work in front of them, and we have trained more than 3,000 people since 2022. We assume nobody in the room knows AI yet, we avoid jargon, and we show the free version of every tool beside Microsoft 365 and Copilot. We teach the four Cs first, then how to check every output. Nobody becomes an expert in a few days, so we focus on safe habits staff keep using.
For procurement and supply chain teams, we build sessions around your manual, your templates and masked examples from your own spending categories, and we are clear about where AI must stop. See how it works on our page for AI training for procurement teams in Kenya.
To talk about training for your procurement unit, 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.
