Can we offer this customer payment terms? Where is the current contract template? How do we process a return after the warehouse has accepted the goods?
Every question gets the same answer: “Ask Marina.”
Marina has worked here for years. She remembers exceptions, knows the agreements, and understands which of the three instructions still applies. She genuinely helps the company.
But right now, Marina is in a meeting. Two sales representatives are waiting for her answer, a new employee has put a task aside, and a customer cannot understand why a simple question is taking so long.
If this sounds like your company, start with a small review. You do not need to document every process or upload the entire company archive into an AI system. Choose one team, collect its recurring questions, and identify which answers could be available without constant help from an experienced colleague.
This article explains how to do that, where an AI knowledge base can help, and how to test its value before a larger rollout.
Why a strong employee becomes essential to everyone else's tasks
Nobody usually appoints Marina as the keeper of every answer.
First, she helps a new employee. Then she explains a difficult case to another department. Eventually, colleagues turn to her even for routine questions because it is faster than finding a document.
An informal system gradually develops around one person:
- instructions are stored in folders, but Marina knows which ones;
- rules change, but Marina remembers the latest decision;
- exceptions are discussed in messages, and only Marina can find the relevant exchange;
- documents contradict each other, and Marina explains what to do in practice.
This is convenient when there are few questions. As the team grows, one employee's availability starts affecting everyone else's speed.
A useful signal for a manager: a task is delayed because an employee cannot get a verified answer, rather than because the task itself is difficult.
What the business loses while the team waits for guidance
The first visible cost is the experienced specialist's time. They repeat explanations and switch between their own work and other people's questions.
But the consequences extend further.
| Situation | What happens to the work | Possible business consequence |
|---|---|---|
| A representative waits for discount approval rules | The customer proposal is not ready | Response time increases |
| A new employee cannot find the procedure | The task stops or proceeds through guesswork | More mentoring and corrections are needed |
| Two employees receive different explanations | The same question is handled differently | Customers hear conflicting terms |
| The experienced specialist goes on vacation | The team postpones unusual cases | A backlog develops |
| An old instruction is used | Work follows outdated rules | Rework is required |
Waiting alone does not prove a financial loss. An employee may switch to another task, and the customer may wait for the answer.
Identify what actually stopped and what happened as a result. This is more useful than treating every hour of waiting as a direct financial loss.
Why a shared folder of instructions may not be enough
Suppose the company has documented its processes and made the documents available to the team. Yet the questions to Marina continue.
The employee may still face a lengthy process:
- Work out the name of the relevant document.
- Find it among similar files.
- Confirm that the version is current.
- Read the relevant section.
- Determine whether the rule applies to their situation.
If messaging a colleague is faster, the employee will probably do that.
Not every company needs AI. Sometimes it is enough to remove outdated files, assign document owners, and create an accessible page with answers to common questions.
An AI knowledge base is worth testing when there is a lot of information, questions are phrased in different ways, and searching regularly delays work.
What an answer that helps an employee continue could look like
Consider a hypothetical question from a sales representative:
The customer wants 20-day payment terms. Can I include them in the proposal?
Instead of giving a generic explanation that payment terms depend on company policy, the assistant should find the applicable internal rule.
For example:
Under the current policy, payment terms longer than 14 days require approval from the head of finance. Do not confirm them to the customer before approval. Include the order value, requested payment period, and company details in the approval request.
Source: “Payment Terms Approval Procedure,” section 3.2.
This is a fictional example, not a universal rule for offering payment terms.
The point is that the employee receives the condition, the next step, and a source to verify. They do not need to find Marina just to learn the standard procedure.
A useful answer lets someone complete a task or understand who should receive it and what information to include.
Start with questions the team keeps asking
Choose one department or process for the initial review. Examples include preparing sales proposals, processing returns, or onboarding new sales representatives.
For one working week, ask employees to record questions they take to colleagues. Existing work discussions can also help, provided their access rules are respected.
Create a small table:
| Field | What to record |
|---|---|
| Employee's question | The actual wording, without rewriting it to match an instruction |
| Task | What the person was trying to do |
| Usual responder | Employee or role |
| Where the answer lives | Document, system, conversation, or only a colleague's experience |
| Frequency | Number of occurrences during the selected period |
| What is delayed | Customer response, estimate, approval, or other work |
| Who confirms the rule | Process owner |
| What to do without an answer | Who should receive the question and what information to include |
Then divide the questions into three groups.
The answer exists and is current. This material can be prepared for search.