How an AI Interviewer Talks With Everyone in Your Company, and How Their Words Are Cited

Before automating a process or replacing a system, a manager needs to know how the work actually gets done today. Written procedures tell part of the story. The rest is known by the people who do that work every day.

A cost analyst knows that one customer’s orders are settled against a different price list, and that the adjustment is made by hand before month-end close. A warehouse clerk knows what to do when goods arrive without a purchase order. None of that is written in a procedure, and it is exactly what a new system has to keep.

Gathering that knowledge in meetings takes months and depends on who took the notes. BrainNova gathers it with an AI interviewer that talks with each person separately and keeps a citation for everything each of them said.

One link and one script per role

Each person receives a link by email or WhatsApp. They open it on a phone or a computer, with nothing to install, and answer at their own pace, typing or speaking. If they have to leave, the interview is paused and they can pick it up later from the same link.

The interviewer follows a script written for each person’s role. A warehouse clerk’s script asks about receiving, stock counts and returns. A finance director’s script asks about closing decisions and the criteria behind approving a payment.

Three pieces behind every question

Every turn the interviewer takes is built from three pieces, from the most general to the most specific.

• The interviewing technique, the same for every study. One question per turn, no flattery, no putting words in the person’s mouth and no offering two answers to pick from.
• The category, which sets the kind of conversation. With the people who do the work, the process is mapped step by step, with how long it takes, how often it happens and what holds it up. With the people who lead, the conversation turns to their decisions and the criteria behind them.
• The role script, with the person’s context and the topics the study needs to cover, in order of priority.

Before the person reads a turn, the system checks it against the rules of the technique, and if any rule fails, the turn is written again.

How it asks and how it closes

The interviewer asks, follows up and asks for examples. When someone answers “it depends,” it asks for the last time it did, what happened and who decided. When someone describes a normal day, it then asks about the other days of the week, because exceptions show up on the days that were not normal.

Could you tell me about the last time an order went out at a price different from the list?

As the conversation goes on, it keeps track of what it has understood and what is still missing. If several turns in a row circle the same detail, it steps back to what explains that episode or moves to another topic. If a priority topic in the script has not come up yet, it steers the conversation there before closing.

The close happens in steps. First it asks whether anything important was left unasked. If the person says no, it says goodbye. If they bring something up, it explores it, asks whether there is anything else, and only then says goodbye. A phrase like “that’s it,” said about a single task, leads to asking what comes next and does not end the interview.

How the person interviewed is treated

From the first turn, the person knows they are talking with an AI, on whose behalf and for which study. That first message is written by the system and is the same for everyone. It also says how long the conversation takes.

• They can skip a question or leave the interview at any time.
• If at any point they ask who they are talking to, the interviewer tells them it is an AI.
• What they say is not used to train models.

Someone walking into an interview about their job often wonders whether their position is at stake. That is why the interviewer never talks about replacing anyone, or about automating, optimizing or cutting, and never asks how much free time the person has left or which part of their work is repetitive. Every turn is checked for those words before the person reads it.

The rule protects the person and it protects the data. Someone who feels their job is under threat recites the official procedure and holds back what they actually do, which is exactly what the study needs to learn.

Every statement with who said it

When the study closes, every conversation is kept in full, written or transcribed, with name, role and time. The statements in each conversation are separated, and each one stays tied to who said it and in what role. That unit is called a mention.

“I adjust that customer’s prices by hand before close, because the system doesn’t bring them in.” (cost analyst)

Mentions show where people disagree. The purchasing coordinator says an order is approved once the department head signs it. The treasury analyst says it does not get paid until it arrives with the director’s signature. Both mentions are kept, each with its source, and no one picks one in advance. The same step having two descriptions inside the same company is already a finding.

Mentions feed the data map, which shows what data each process uses, who produces it, who consumes it and where the sources contradict each other. With that map, a manager decides what to automate first, what cannot be lost when changing systems and which differences to settle with each area, and every decision traces back to what each person said.