SimplBrain
How an Employee Assistant Should Distinguish Policy From Helpful Background
SimplSolutions Editorial Team · 8 min read
Published

Employees rarely ask an assistant for a document. They ask what they are allowed to do.
That distinction creates a governance problem. A company may have a current policy, an older FAQ, a process guide, a manager’s note, and informal discussion about the same topic. All may contain useful words. They do not all have the same authority.
An employee assistant should therefore decide what kind of knowledge it is reading before deciding what to say. Without that classification, helpful background can quietly become apparent policy.
The core problem: relevance is not authority
A source can be relevant without being binding.
An FAQ may explain a policy in plain language but lag behind a revision. A manager note may describe a common practice but apply only to one team. A procedure may explain the steps for requesting an exception without granting the exception itself.
If an assistant blends these sources into one fluent answer, the employee may not be able to tell which sentence is a requirement, which is an interpretation, and which is merely advice.
That is a control failure even when every sentence sounds reasonable.
The NIST AI Risk Management Framework is intended to help organizations manage AI-related risks and incorporate trustworthiness considerations into the design, development, use, and evaluation of AI products, services, and systems. For an employee assistant, that principle has a practical implication: answer quality includes knowing when a statement is authoritative, not just finding text that appears relevant.
A four-part authority model
A useful operating model separates business knowledge into four classes.
1. Binding policy
Binding policy states what the organization requires, permits, prohibits, or formally grants. It should have a defined owner, an effective date or version, and a scope that the assistant can identify.
When a binding policy applies, it should control the assistant’s answer. The assistant should present it as policy only when the source has been approved for that purpose and the employee’s situation falls within its scope.
The assistant should not silently expand a policy because another document sounds more accommodating.
2. Approved procedure
An approved procedure explains how to carry out a policy. It may identify an approval route, required form, timing, or responsible team.
A procedure can answer questions such as “How do I submit this request?” It should not be treated as permission to do something the governing policy does not allow.
The distinction matters because procedural convenience is not a change in entitlement.
3. Contextual guidance
Contextual guidance helps an employee understand intent, examples, common scenarios, or practical considerations. It can make a policy easier to apply, but it does not automatically create a new requirement or exception.
The assistant can use this material to explain a policy when the relationship is clear. It should label the difference: “The policy requires X. The accompanying guidance suggests Y as a typical way to comply.”
4. Unsupported commentary
Unsupported commentary includes informal notes, unapproved interpretations, stale summaries, speculation, and statements with no identifiable owner or governing status.
Such material may signal that a policy is confusing or that a team has developed a local practice. It should not be presented as company policy.
If it conflicts with an authoritative source, the assistant should not average the two positions. It should preserve the conflict and route the question for review.
Worked example: travel reimbursement
Consider a fictional travel question. An employee asks:
“Can I book a premium-economy seat for a nine-hour international trip, and do I need pre-approval?”
The assistant finds three sources:
- Current travel policy: The approved policy says standard economy is required for international flights under a stated duration threshold. It also says exceptions require written approval from a designated finance owner before booking.
- Older FAQ: A previous FAQ says premium economy is acceptable for international flights longer than eight hours and suggests that manager approval is sufficient.
- Informal manager note: A team note says, “We usually let people book premium economy on long trips—just keep the receipt.”
A weak assistant might combine the eight-hour threshold from the FAQ with the informal reimbursement practice and answer that the employee can proceed after asking a manager.
A governed assistant should reason differently:
- The current approved travel policy is the governing source for what the company permits.
- The older FAQ is useful background but cannot override the current policy.
- The manager note describes a local practice, not a company-wide authorization.
- The answer should follow the current policy and identify the required approval path.
A responsible response would be:
“The current travel policy governs this request. Based on that policy, premium economy requires the specified written approval before booking, and the applicable flight-duration rule should be checked against the policy’s current wording. The older FAQ and a manager note describe different practices, but they do not override the current policy. If the policy’s duration threshold or approver is unclear for your itinerary, contact the designated finance or travel-policy owner before booking.”
This answer is intentionally narrower than the most convenient answer. It does not declare the trip eligible based on the older FAQ. It does not convert a manager’s informal note into an exception. It gives the employee a useful next action without inventing certainty.
What the assistant should say when sources disagree
Source conflict should change answer behavior, not merely trigger a footnote.
The assistant should:
- State which source governs and why, if that authority is recorded.
- Separate policy requirements from procedural instructions and explanatory context.
- Avoid using an older summary to fill a gap in a current policy unless the relationship is explicitly approved.
- Identify missing facts that could change the answer, such as employee group, location, date, transaction type, or approval status.
- Escalate when the governing document is ambiguous, expired, outside scope, or contradicted by another source with comparable authority.
Escalation is not a failure to answer. It is the correct outcome when the assistant cannot establish that a conclusion is authorized.
A useful answer can still tell the employee what is known, what is uncertain, and who owns the decision.
Classification should happen before generation
Operators should not rely on wording alone to make this distinction. A source should carry governance information that an assistant can use, such as:
- Source type: policy, procedure, guidance, or commentary.
- Owner: the function accountable for the content.
- Approval status: whether it has been formally approved for employee use.
- Effective date and review date.
- Scope: the populations, locations, processes, and situations covered.
- Supersession status: whether it replaces or has been replaced by another source.
- Escalation contact: the person or team responsible for unresolved questions.
These attributes are more useful than a generic label such as “official.” A document can be official in the sense that it came from a company mailbox and still be unsuitable as a policy authority.
The assistant’s answer should inherit these distinctions. A policy claim can be stated directly when applicable. A procedure can be presented as a way to comply. Context can be framed as explanation. Commentary should be withheld, qualified, or routed for review.
What operators should measure
An assistant that cites the right document can still give the wrong kind of answer. Review should therefore examine behavior, not only retrieval.
Useful review questions include:
- Did the assistant identify the governing source before answering?
- Did it mistake a procedure or FAQ for permission?
- Did it distinguish current material from superseded material?
- Did it expose uncertainty when the employee’s facts were incomplete?
- Did it provide an appropriate escalation path?
- Could an employee tell which sentence was policy and which was context?
These checks support the broader goal of trustworthy AI risk management: making system use and evaluation part of governance, rather than treating deployment as the end of the process. NIST describes the AI RMF as intended for voluntary use. As an editorial matter, organizations should define their own authority model and accountability structure for employee-facing answers.
A practical standard for a governed Business Brain
A governed Business Brain should preserve the difference between “the company requires,” “the approved process is,” “the guidance suggests,” and “someone said.” That distinction is essential for employee trust and operational accountability.
The goal is not to make an assistant evasive. The goal is to make it precise about the strength of its answer.
For operators, the next step is to review a small set of high-impact employee questions—travel, expenses, leave, procurement, security, or access—and classify the sources an assistant would encounter. Record which source governs, which materials explain it, and what must trigger escalation.
Then test whether the assistant preserves those distinctions in plain language.
The takeaway
An employee assistant should never infer policy from relevance, recency alone, or confident wording. It should classify knowledge by authority first.
Binding policy controls the answer. Approved procedure explains execution. Contextual guidance supports interpretation. Unsupported commentary is not authorization.
When the hierarchy is unclear, the assistant should say what is known, identify the uncertainty, and escalate to the accountable owner. That is not less helpful. It is what makes the answer safe to rely on.



