A GenAI assistant grounded in company policy that answers employees' benefits, travel, and allowance questions instantly — in the tools they already use.
Across a multinational workforce, employees struggled to find quick, accurate answers to everyday questions about benefits, travel, and allowances — the information existed, but it was buried across long policy documents.
The company wanted one trusted place to get accurate, on-policy answers, available where employees already work, without overloading HR with repetitive queries — and without the assistant ever inventing policy.
The assistant is grounded in the company's own internal policy documents using retrieval, so every answer reflects actual policy rather than a generic guess.
Rather than a new portal to learn, we surfaced the assistant through everyday channels employees already use, such as Teams.
We built in guardrails to keep answers accurate, on-policy, and within scope — essential when the topic is people's benefits and pay.
Employee questions about benefits, travel, and allowances have specific, often company- and even region-specific correct answers determined by policy documents — a general-purpose assistant's 'typical' answer about, say, travel reimbursement policy could easily be wrong for this specific employer. The assistant is built on a RAG pipeline over the company's actual policy documents, with the same retrieval-quality considerations that matter for any RAG system: documents are chunked along policy-section boundaries (so a policy and its exceptions stay together), and every answer is grounded in retrieved policy text rather than the model's general training knowledge.
A multinational consumer-goods enterprise has policy variation across regions and employee categories — a travel allowance policy that differs by country, benefits that differ by employment level. The retrieval layer needs to surface the policy text relevant to the specific employee asking (which may require knowing their region/category as context for the query) rather than retrieving a policy that's correct somewhere but not for this employee. This context-aware retrieval was one of the more involved parts of the build — getting it wrong means confidently giving an employee the wrong country's policy.
An assistant employees have to remember to go to a separate portal to use gets used rarely. The assistant is integrated into the workplace tools employees already use day to day (chat platforms like Slack or Teams), so asking a benefits question is as easy as asking a colleague — which was a deliberate adoption-driven design decision as much as a technical one.
Employees got fast, accurate answers in the tools they already use, and HR was freed from a stream of repetitive questions — a single, trusted source for everyday policy questions.
Tell us what you're building. We'll tell you the fastest honest path to shipping it.
Start a conversation →