AI for HR: your own people-ops assistant

Fixed prices: €2,500 audit, €20,000 proof of concept, production from €50,000. The audit scopes your handbook and highest-volume questions, the proof of concept runs on your real content, and production delivers the integrated build you own. You know the number before we start.

• Timelines: audit 1–2 weeks, proof of concept 4–6 weeks, production from 8–12 weeks.
• All prices exclude VAT and are published, so no quote round is needed to budget.
• An HR assistant is one of the cheaper first use cases, because the content already exists in your handbook.

Yes, and HR data needs the strictest handling of any use case. The assistant is grounded only in your approved documents, supports EU-hosted infrastructure and role-based access, and never sends employee data to train public models. Privacy is designed in rather than bolted on.

• Role-based access matters most here: a salary policy should not surface to everyone who asks.
• Dutch employers answer to the Autoriteit Persoonsgegevens, Belgian employers to the Gegevensbeschermingsautoriteit — the GDPR obligations themselves are identical.
• Ask any supplier in writing whether prompts and documents are used for model training. They should not be.

Often yes in the Netherlands, depending on what it does. Under the Works Councils Act (WOR, article 27), your ondernemingsraad has a consent right over arrangements for staff monitoring and personnel assessment. An assistant that only answers handbook questions is lower risk than one that logs or evaluates.

• Question logging is the usual trigger — who asked what about sick leave is exactly the data that raises concerns.
• In Belgium, CAO nr. 39 requires informing and consulting the works council about new technology with collective impact.
• Simple mitigation: keep queries anonymous by default, and say so in the rollout communication.

No. It uses retrieval-augmented generation, so every answer is drawn from your own handbook and policies and cites its source. If the handbook does not cover something, it says so and routes to a person instead of guessing. That refusal behaviour is the point.

• Invented HR policy is a real liability — an assistant guessing at leave entitlement creates an expectation you may have to honour.
• Citations let an employee check the source, which also reduces follow-up questions to your HR team.
• Test it before launch with questions your handbook deliberately does not answer.

It answers from your handbook, not from generic training data. Off-the-shelf chatbots know HR in general and your company not at all, and they lock you into their SaaS. We build an assistant grounded in your specific policies, integrated with your systems, that you own outright.

• Generic tools cannot answer a question about your cao, your leave scheme or your own notice periods.
• Ownership covers code, data and model choices, so there is no per-seat licence.
• Off-the-shelf is the better call if your policies are entirely standard — which, for most Dutch and Belgian employers, they are not.

Yes. The production build integrates with your HRIS, document stores and ticketing, so answers stay in sync with live policies and routine requests get handled end to end. Integrations are scoped during the €2,500 audit, which is what keeps the build price fixed.

• Syncing to the live handbook matters: a stale copy is how an assistant starts quoting last year's policy.
• Where a system has no usable API, we fall back to scheduled exports or a direct database read.
• Start read-only. Let the assistant answer before you let it file or approve anything.