How to Build an AI Internal Company Assistant for Your Staff

Every organisation with more than a few dozen staff answers the same questions all day: "how many days of leave do I have?", "what is the procedure for requesting a laptop?", "who approves a purchase above ₦500,000?", "how do I get my pension statement?". HR, finance and IT teams spend hours a week repeating answers that already exist in a policy document nobody can find. An internal assistant turns those documents into a colleague that answers instantly and consistently, at 7am or on a public holiday.

This article covers how to build one for a Nigerian company or organisation: which questions to start with, how to control who sees what, which channel to use, how to handle actions such as leave requests, adoption, a labelled hypothetical example and indicative costs. It is about assistants for employees. The owner-facing assistant that answers questions from sales and finance data is covered in the article on building an AI business assistant, and the knowledge layer both depend on is covered in the AI knowledge base article.

What an internal company assistant does

An AI internal company assistant is a chat-based tool, available only to employees, that answers questions using the organisation's own policies, procedures, handbooks and system guides, cites the source document, and hands over to HR, IT or finance when a question needs a person. Many also perform simple actions such as raising a leave request or logging an IT ticket.

Typical scope in a Nigerian organisation:

• HR: leave entitlements and procedure, payroll dates, pension and health insurance enrolment, disciplinary and grievance procedures, dress code, remote-work rules, onboarding checklists.
• Finance: expense claims, per diem rates, purchase approval thresholds, petty cash rules, supplier onboarding steps, VAT and WHT handling reminders (with a referral to the accountant for specifics).
• IT: password resets (via ticket), software requests, laptop policy, VPN and email setup guides, security rules.
• Operations: SOPs, safety procedures, branch opening and closing routines, escalation contacts.
• Company knowledge: organisation chart, who to contact for what, office locations and hours.

What it must not do: make HR decisions, disclose one employee's personal information to another, or replace the formal channels for grievances and disciplinary matters.

Which questions should it answer first?

Start with the questions that are asked most often, have a single written answer, and carry low risk if the answer is slightly incomplete. Use this framework.

Ask HR, finance and IT to each list their top ten repeated questions. Thirty questions with agreed answers is enough for a first version that staff will find useful.

Architecture: knowledge, access, channel, actions

An internal assistant has four layers: a knowledge base of company documents, an access-control layer that filters by role, a channel where staff talk to it, and an optional actions layer that connects to HR, IT and finance systems.

1. Knowledge base. Policies, SOPs and guides converted to text, chunked by section, tagged with department, audience and effective date, and stored for retrieval. This is the same retrieval-augmented generation approach used in other assistants; the knowledge base article explains it in depth.
2. Access control. Each employee is identified (through Microsoft 365, Google Workspace or your HR system), mapped to a role and department, and only content tagged for that role is retrieved. Management-only documents never reach a junior staff member's query.
3. Channel. Microsoft Teams, Slack or Google Chat for organisations that already use them; a web app or WhatsApp for field staff and branches without laptops.
4. Actions. Tools the model may call: create_leave_request, log_it_ticket, get_my_leave_balance, find_contact. Each connects to the relevant system and returns structured data. Personal data such as leave balance is fetched only for the authenticated user asking.
5. Escalation and logging. Unanswered or sensitive questions are routed to a named person or ticket queue; every interaction is logged for quality review and accountability.

Step-by-step: building the internal assistant

The core steps are: collect questions, agree the source documents, tag content by audience, build the knowledge base, identify users and roles, connect the channel, add escalation, pilot with one department, then roll out. Here is each step.

1. Collect the top questions per department. HR, finance, IT and operations each provide their top ten, with the current answer and the source document.
2. Agree and update the source documents. Where the handbook is outdated or contradicts practice, fix the handbook first. The assistant will expose inconsistencies quickly.
3. Tag content by audience and department. Public-to-all-staff, managers only, HR only, finance only, specific branches. This tagging drives access control.
4. Build the knowledge base. Convert, chunk by section, embed and store with metadata. Test retrieval against the collected questions.
5. Connect identity. Single sign-on through Microsoft 365 or Google Workspace where available; otherwise a staff directory with verified phone numbers for WhatsApp.
6. Write the system prompt. Tone (helpful, brief, professional), the rule to always cite the policy and link to it, what to refuse (other employees' data, legal opinions, disciplinary matters) and the escalation contacts.
7. Add escalation. "I can't answer that; I've sent your question to HR" with a ticket or email, and a way for staff to reach a person directly.
8. Add actions carefully. Start with read-only personal lookups (own leave balance) and ticket creation. Add anything that changes records only after the answers layer is trusted.
9. Pilot with one department for a month. Review every unanswered and corrected question weekly; update documents and tags.
10. Roll out with short training. A five-minute demo, a one-page guide, and visible sponsorship from HR and management.

Access control by role and department

Access control is the feature that makes an internal assistant safe, and it must be enforced in code rather than in the prompt. The retrieval step should filter chunks by the user's role and department before the model ever sees them.

Practical model:

• All staff: general policies, procedures, contacts, IT guides.
• Managers: performance review procedures, approval workflows, budget guidance.
• HR: compensation frameworks, disciplinary case procedures, templates.
• Finance: payment procedures, bank details, audit guidance.
• Branch or unit specific: local SOPs and contacts.

Personal data is a separate category. "What is my leave balance?" is answered by a tool that queries the HR system for the authenticated user only. "What is Adaeze's leave balance?" must be refused unless the asker is Adaeze's manager and the tool enforces that relationship. Log all personal-data lookups.

Choosing the channel: Teams, Slack, Google Chat or WhatsApp

Put the assistant where staff already spend their working day. For office-based Nigerian companies that is often Microsoft Teams or Google Chat; for branch, field and factory staff it is WhatsApp; for tech companies it is Slack.

Many organisations run two channels on one backend: Teams or Google Chat for head office, WhatsApp for branches. The knowledge base and access rules stay the same.

From answers to actions: leave, requests and tickets

Once staff trust the answers, the assistant can take routine actions, which is where the largest time savings for HR and IT come from. Introduce actions in stages.

• Stage 1, read-only personal lookups: own leave balance, own payslip date, own pension provider, own assigned equipment.
• Stage 2, creating requests: leave requests, IT tickets, expense claim drafts, purchase request forms, each routed to the normal approver in the existing system.
• Stage 3, guided workflows: onboarding a new employee (collecting documents, creating accounts, scheduling induction) with the assistant tracking each step and chasing owners.

Every action should confirm before submitting ("Request 3 days of annual leave from 6 to 8 October, to be approved by Mr Bello? Reply YES"), write to the existing HR or ticketing system rather than a separate store, and be logged.

What changes for Nigerian organisations

Building an internal assistant in Nigeria changes in four ways: many staff are not desk-based and reach the assistant by WhatsApp, policies must reflect Nigerian statutory context, employee data is governed by the NDPA 2023, and power and connectivity shape how it is hosted.

WhatsApp for non-desk staff. Bank branch staff, sales reps, drivers, factory and hospitality workers rarely sit at a laptop. A WhatsApp channel with verified numbers gives them the same access to policies and requests. Keep answers short with a link to the full document.

Nigerian statutory context. Pension contributions, health insurance, PAYE, the Labour Act and industrial-relations rules shape HR questions. The assistant should explain your organisation's policy and point staff to HR or a qualified adviser for anything statutory; it must not give legal or tax advice. Verify policy content against current requirements.

Employee data protection. Staff records, salaries, health information and disciplinary notes are personal data under the NDPA 2023. Restrict access by role in code, keep sensitive content out of the general knowledge base, log personal-data lookups, and choose a model provider or private deployment that meets your data-handling requirements. Verify current NDPC guidance; this is not legal advice.

Hosting and cost. Host in the cloud so the assistant works during office power cuts. Model and channel fees are in USD; internal assistants have predictable volumes, so caps are easy to set.

Culture and trust. Staff may worry that questions are monitored. Be transparent: publish what is logged, who can see it and why, and make clear that the assistant is a help tool, not a surveillance tool.

Example (hypothetical): a microfinance bank with 12 branches

Example (hypothetical): A microfinance bank headquartered in Lagos with 12 branches across the South-West and about 300 staff runs HR and IT from head office. Branch staff call or message head office dozens of times a day about leave, allowances, loan-processing procedures and system access, and answers vary by who responds.

The bank builds an internal assistant with a knowledge base of the staff handbook, credit and operations SOPs, IT guides and finance procedures, tagged for all-staff, branch managers, credit officers, HR and finance. Head office uses Microsoft Teams; branches use a WhatsApp channel with verified staff numbers. Staff can ask about procedures, look up their own leave balance and log IT tickets. Questions about individual customers, disciplinary matters or other staff are refused and routed to the right department.

The pilot with two branches exposes that the credit SOP in circulation is two versions old; it is updated before rollout. Weekly review of unanswered questions drives handbook updates for three months, after which HR reports far fewer repeat calls and branch managers get consistent answers. This is an illustrative scenario, not a Linestech client result.

How much does an internal AI assistant cost in Nigeria?

The cost of an internal company assistant in Nigeria is driven by the volume and condition of policy documents, the number of roles and access levels, the channels, identity integration, and whether it performs actions in HR and IT systems. The figures below are indicative 2026 ranges; actual quotes vary with scope, vendor and exchange rate.

Recurring costs, mostly in USD: model usage (predictable for internal volumes), channel fees (WhatsApp conversation charges; Slack or Teams subscriptions you may already pay), hosting at roughly ₦150,000–₦800,000 per year, and maintenance or a retainer of ₦20,000–₦150,000 per month for content updates and tuning. Policy clean-up is often quoted separately. Compare two or three written quotations on the same scope: documents, roles, channels, identity, actions, logging and code ownership.

Adoption: getting staff to actually use it

An internal assistant only saves HR and IT time if staff use it instead of calling. Adoption is a change-management task, not a technical one.

• Launch with a real need: open enrolment for health insurance, a new leave year, a policy change.
• Make it the first stop: HR and IT reply to routine questions with "ask the assistant" for the first month, with the answer attached so nobody is stranded.
• Show sources every time. Staff trust an answer that links to the policy.
• Publicise wins: "the assistant answered 400 questions last month; the three it could not answer led to these policy clarifications".
• Keep a feedback button and act on it visibly.
• Train managers first; they set the example in branches and departments.

Mistakes to avoid

• Launching on outdated handbooks. The assistant will surface every contradiction to every employee. Fix documents first.
• Access control in the prompt only. "Do not show HR documents to non-HR staff" in a prompt is not security. Filter at retrieval in code.
• Answering personal-data questions about other employees. Enforce identity and manager relationships in the tools.
• Letting the assistant opine on legal or disciplinary matters. Explain policy, refer to HR or a qualified adviser.
• Ignoring non-desk staff. If the branches cannot reach it, most of the repeat questions remain.
• No escalation path. A dead end sends staff back to phoning HR and kills trust.
• Hidden logging. Undisclosed monitoring undermines adoption and may raise data-protection issues.

Conclusion

An internal company assistant succeeds on three things: current, agreed source documents; access control enforced by role in code; and a channel that reaches every employee, including the ones in branches and on the road. Start with thirty repeated questions, pilot in one department, add personal lookups and request creation only once answers are trusted, and be transparent about logging. Indicatively, a first version costs from around ₦1,000,000 in Nigeria plus USD-denominated running costs, and the policy clean-up it forces benefits the organisation regardless.

If HR, IT and finance in your organisation spend their days repeating the same answers, Linestech can help you assess your policy documents, design role-based access and channels, and scope an internal assistant your staff will actually use.

Frequently asked questions

What is the difference between an internal assistant and a general chatbot like ChatGPT?

A general chatbot knows nothing about your policies and cannot be trusted to state your leave rules. An internal assistant answers only from your organisation's documents, cites them, enforces who may see what, and can create requests in your HR and IT systems. Staff pasting policies into a public chatbot also creates data-protection risk that a controlled internal assistant avoids.

Can it answer personal questions such as my own leave balance?

Yes, if it is connected to your HR system through a tool that looks up data only for the authenticated employee asking. It should refuse requests about other employees unless the asker has a defined right to see them, such as a line manager, and that relationship is enforced in code, not in the prompt.

Which is better for staff in branches: WhatsApp or a web app?

WhatsApp is usually adopted faster because staff already use it, and it works well on basic phones and poor connections. It requires the WhatsApp Business Platform, verified staff numbers and per-conversation fees. A web app gives more control and richer formatting but needs its own login and more data. Many organisations run both on one backend.

How does it handle questions with legal or statutory implications?

It explains your organisation's written policy, cites the document, and refers the employee to HR or a qualified professional for anything requiring interpretation of law, tax or pension rules. It should never present itself as giving legal advice. Make sure your policies are reviewed against current Nigerian requirements before they go into the knowledge base.

What data protection issues should we think about?

Employee records are personal data under the NDPA 2023. Restrict content by role, keep sensitive HR files out of the general knowledge base, log and limit personal-data lookups, tell staff what is logged, and check the data-handling terms of your model provider or consider a private deployment. Verify current NDPC guidance; this is not legal advice.

How long does it take to build?

A starter assistant answering from an existing, current handbook on one channel can be piloted in three to five weeks. Adding role-based access, single sign-on and a WhatsApp channel typically takes six to ten weeks. Actions in HR and IT systems add time depending on those systems' APIs. Policy clean-up is often the longest part.

Should we build it or use a SaaS HR assistant?

If your HR platform already offers an assistant that reads your policies and you are happy with USD per-user pricing and its data-handling terms, using it can be quick. Build when you need multiple channels (especially WhatsApp for branches), Nigerian-specific workflows, integration with local systems, or stricter control of where employee data is processed.

Sources and further reading

Figures, platform rules and regulations change. These are the primary references behind this article and the places to check before you act on it.

• Nigeria Data Protection Commission (NDPC)
• WhatsApp Business Platform — Meta for Developers
• MDN Web Docs — web platform reference
• OpenAI platform documentation
• Anthropic developer documentation