A real interview, repeatable for every candidate.

Criterio Talent is a conversational interview platform for companies. A virtual interviewer runs the conversation over voice, with audiovisual presence, following the script your team defined: it asks, listens, follows up where it needs to, and closes.

Every interview produces three things: a turn-by-turn transcript, an assessment per competency against the client’s rubric, and the verbatim fragment behind each conclusion. With that, comparing candidates stops being a comparison of impressions.

What changes between a high-volume frontline role and a management position is not the product: it is the depth it is configured with.

You paste the job description and an assistant builds the interview definition: competencies, items and levels. It is not a generic template with the title swapped — it comes out of the role you just described.

From there you adjust it by talking: add a competency, rewrite an item, raise the bar. Changes that remove something ask for confirmation before they apply, because deleting a competency mid-conversation is the kind of thing nobody wants to find out about later.

The script stays versioned and gets reused across candidates. That is the half that makes a process comparable: everyone answers the same definition, and when it changes, the change is visible.

The draft is saved as you work, so an interrupted session does not cost you a fresh start.

The difference from async video shows up on the second turn. When someone answers “we had a problem and we sorted it out together”, there is nothing there to assess: a situation with no action of their own and no outcome. Criterio asks for what is missing.

And it knows when to stop. Each item has its own time budget, follow-ups included, and an explicit rule: if a second follow-up does not surface what was being looked for, it closes and the interview moves on. Pushing a third time does not produce better evidence — it produces an uncomfortable person.

This is the capability that separates it fastest from a recorded questionnaire: the conversation adapts to what it just heard.

Duration is a decision about the process, not a fixed feature of the product. It is declared in the script, from five minutes up, and each question gets the room it needs for the answer to go somewhere instead of stopping at the headline.

That is what lets the same system cover a high-volume role and a position that asks for a good deal more detail. You configure the depth; you do not change tools.

The maximum for each implementation is agreed before you start, and that is the one that governs day to day. Raising it later is configuration, not a project.

Each competency in the script gets a level, the reason it was assigned, and the fragment of the conversation that supports it. Whoever reads the report goes from the verdict to the sentence, and from the sentence to the recording.

And there is a decision here worth knowing about, because almost no system makes it: what is published in quotes is not what the model wrote. It is the verbatim slice of the transcript. The model’s text is only used to locate the segment; what reaches the screen is cut from the real turn.

The consequence is the part that matters: an invented quote cannot be published, however much the model insists, because there is nothing to cut it from.

A role can have ordered rounds — a first round, a technical one, one with the director — each with its own name, its own place in the process and its own interviewer profile. Your ATS declares them when it syncs the role, or they are created through the API.

The same person walks through all of them without turning into three different candidates: their candidacy is one, and it accumulates everything they answered along the way. The round name is frozen into each interview once it runs, so renaming “First round” to “Technical” does not rewrite the record of someone who answered it in March.

Who moves to the next round is decided by a person from the platform, or by the client system through the API. At that moment the individual link for that round is issued, and your team or your ATS gets it to the candidate through the channel where you already talk to them.

A round orders the process, the invitation and the file; the questionnaire comes from the role, and today all of its rounds run the script published for it. Each round also keeps a stable key the integration refers to, so renaming it on screen does not break the ATS connection.

Within a role, candidacies sit side by side per requirement: the same competencies, the same instrument, and answers you can go and read. The position board aggregates the report for the whole process and exports it under the client’s brand.

Advancing or rejecting is done by a person, and it is recorded with their name, the time and their reason. That is where the team’s judgement pays off most: not in repeating the same first conversation thirty times, but in deciding once everything is on the table.