Policy questions answered instantly,
from your actual documents

The process today

Every company has a handful of policy questions that get asked constantly: how many vacation days carry over, what the expense limit is for a client dinner, whether a particular tool is approved for work data. The answer usually exists somewhere, a handbook, a Notion page, a PDF from two reorganizations ago, but finding it takes longer than asking a colleague, so people ask a colleague instead. The colleague answers from memory, which is fine until the policy changed last month and nobody told them.

This creates a quiet inconsistency problem. Two employees who ask the same question on the same day can get two different answers depending on who they happened to ask, and neither answer may match what is actually written in the current policy document. HR and managers end up fielding the same handful of questions over and over, not because the questions are hard, but because finding the answer in the actual source document is more friction than just asking a person directly.

The cost shows up as interruptions rather than a single large expense: a manager pulled out of focused work to answer something that is written down somewhere, multiplied across a team and across a year. It is rarely dramatic, but it adds up, and it gets worse as the handbook grows and fewer people have read the latest version in full.

What the agent does

The agent reads directly from your handbook, policies and SOPs using retrieval over the actual documents, so an answer is grounded in what is currently written rather than a general impression of company policy. Every answer comes with the source document cited, so an employee, or a manager double-checking, can see exactly where the answer came from and judge whether it is current.

Content lives in Notion, Google Drive, or whatever document folder your team already maintains, and an admin panel lets someone update the underlying content without needing a developer involved, so a policy change made this morning is reflected in the next answer rather than waiting for a rebuild. Version control is built in specifically so an outdated policy document does not keep getting quoted after it has been superseded. The agent runs in Telegram, Slack or a web chat widget, whichever matches where your team already asks questions, and produces a usage report showing the most-asked questions and the gaps in documentation they expose, which is often as useful to HR as the answers themselves.

What stays with humans

Anything the agent cannot answer clearly from the documents gets routed to HR or a manager rather than guessed at, and that routing decision, saying “I don’t know, ask a person” instead of inventing a plausible-sounding policy, is deliberate rather than a fallback of last resort. Deciding what the actual policy should be, resolving an ambiguous or contradictory document, and handling any question that touches a specific employee’s personal situation all stay with a human. The agent answers only from documents it has been given access to; it does not browse the open internet or fill gaps with general knowledge about how other companies typically handle something.

Guards

Every answer is cited to its source document, which makes it possible to check an answer against the original text rather than trusting it blindly, and the usage report surfaces which questions come up most and where documentation is thin or missing. Version control means an edited policy actually replaces the old one in what the agent quotes, instead of both versions floating around. When the documents do not clearly answer a question, the agent escalates rather than answering with confidence it has not earned, which keeps the failure mode visible instead of silent.

Price and timeline

Running cost is usually $15 to $50 a month in model usage depending on question volume, with a budget cap set before launch.

Related

This works well alongside employee onboarding, since new hires generate a disproportionate share of policy questions in their first weeks, with compliance checklists where the underlying documents need to stay audit-ready, and with SOP generation for keeping the source procedures themselves current rather than just answering questions about stale ones. The visa consulting centre case study describes a close relative of this build, a client-facing AI consultant answering from a knowledge base that non-developers can edit live, and SENET shows a memory and retrieval engine built for a sellable AI assistant product. For more on the approach, see the AI agents service page and the automation-everything overview.

Want to see your own handbook answer questions instead of your HR inbox? Get in touch and we will look at what documents you already have.

Tired of doing this by hand? We can take the whole routine off your team, not just this step: Routine takeover, from $400 →