The verdict in three sentences
A retrieval-augmented (RAG) HR chatbot pays off once an HR team handles more than 300 repetitive questions a month. For a 3,500-employee group in Singapore, the 2026 market range is roughly SGD 45,000 to 110,000 to build and SGD 600 to 2,500 a month in model and hosting costs. The hard part is not the AI but access control: a regional director and a warehouse operator should not read the same internal memos.
Why HR is buried under the same 600 questions
In a multi-entity Singapore group, questions arrive by email, Slack, Teams and phone: annual leave balance, medical and dental claim limits, flexible work arrangements under the Tripartite Guidelines, public holiday rules, CPF contribution questions, overseas posting allowances. In a typical sample of 600 monthly requests, more than 70% are answered by a document that already exists. The problem is that the 400 documents (employee handbook, policy circulars, insurance guides, travel and expense rules) live across SharePoint sites and inboxes.
A RAG assistant indexes those documents, retrieves the relevant passages and drafts a sourced answer with a link to the original policy. It does not guess: when nothing relevant is found, it routes the employee to the right HR contact.
| Question type | Share of volume | Can the assistant resolve it? |
|---|---|---|
| Leave and absence | 28% | Yes, with live balance from the HRIS |
| Medical and insurance benefits | 18% | Yes, from the benefit guides |
| Payroll, CPF and bonuses | 17% | Partly, personal data excluded |
| Policies and circulars | 14% | Yes |
| Learning and internal mobility | 10% | Yes, catalogue indexed |
| Sensitive individual cases | 13% | No, handed to an HR partner |
What an HR chatbot costs in Singapore in 2026
Budget is driven by document volume, directory integration and how finely access rights are split. These are 2026 market ranges for this kind of project in Singapore.
| Item | 2026 range | Comment |
|---|---|---|
| Scoping and document audit | SGD 6,000 to 12,000 | Clean-up of 400 documents, obsolete versions removed |
| Ingestion, chunking and indexing | SGD 10,000 to 25,000 | PDF, Word, scanned forms with OCR |
| Directory sign-in (Entra ID or Okta) | SGD 7,000 to 18,000 | Single sign-on with the employee account |
| Rights by population | SGD 9,000 to 20,000 | HQ, regional offices, managers, non-managers |
| Web interface and Teams or intranet integration | SGD 8,000 to 20,000 | Mobile view for frontline staff |
| Testing, HR acceptance and go-live | SGD 5,000 to 15,000 | Reference set of 200 questions |
| Monthly model and hosting cost | SGD 600 to 2,500 per month | Depends on volume and model choice |
A realistic timeline is 8 to 12 weeks: 2 weeks of scoping, 4 to 6 weeks of build, 2 to 4 weeks of acceptance testing with an employee panel.
Security and compliance: the non-negotiables
The PDPA applies, and the PDPC expects a clear purpose, retention limits and safeguards when employee data is processed by AI systems. Four rules prevent bad surprises. First, no individual payroll data in the document index: leave balances are read live from the HRIS after authentication. Second, every document carries a population tag (all staff, managers, a given entity, leadership) and the assistant filters before it even searches. Third, conversation logs are kept 90 days maximum and anonymised for analytics. Fourth, the model provider must contractually guarantee that data is not used for training, with data residency settled in the contract.
At Kolonell, this kind of assistant falls under custom applications of medium complexity: 4,000,000 to 7,000,000 FCFA (6,100 to 10,670 EUR) depending on scope. Details are set in the quote after the document audit.
Mini case study
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
Rachel, Head of HR for a 3,500-employee logistics group in Singapore, logs 600 questions a month. Each one takes an HR executive 12 minutes on average, or 120 hours a month. With 60% automatic resolution, the assistant frees 72 hours a month. At a loaded cost of SGD 55 an hour, that is about SGD 3,960 a month. With SGD 1,500 in monthly model costs, net savings are around SGD 2,460 a month, or SGD 29,520 a year. A SGD 75,000 project pays back in about 30 months on HR time alone, and much faster once you count employees getting an answer in 20 seconds instead of two days.
FAQ
Can the chatbot invent an HR rule that does not exist?
RAG sharply reduces that risk because every answer cites its source document. A threshold is also set: below 70% relevance, the assistant hands over to an HR executive instead of answering.
Do we need to replace our HRIS?
No. The assistant connects to the existing HRIS through its API to read leave balances, and that integration usually accounts for 15 to 25% of the total budget.
Which languages does it answer in?
It can answer in English and, if required, in Mandarin, Malay or Tamil. Languages covered and expected quality are set in the quote after testing on a sample of 50 questions.
How fast is a new policy picked up?
A new circular dropped into the source folder is indexed in under 15 minutes. The previous version is archived automatically to avoid contradictory answers.
What happens with sensitive questions?
Harassment, health or grievance topics are detected and routed straight to a human, with no content stored in analytics. These cases make up about 13% of volume.
Let's scope your project. Send us your monthly HR question volume and the list of documents to cover, and we will scope the build and an 8 to 12 week plan. Detailed quote within 48 h. WhatsApp +221 77 596 93 33.
Mohamed Bah
Fondateur, Kolonell
Passionate about digital and entrepreneurship in Africa, Mohamed has been helping Sénégalese businesses with their digital transformation since 2020. Founder of Kolonell, he believes every SME deserves a professional and accessible online présence.
