Solutions
Judge every graduate by the same criteria, and improve those criteria every year
Applications spike, but you cannot add interviewers overnight. The AI Interviewer takes the first round, 24 hours a day, 365 days a year, in parallel. You define the criteria yourself, so you can set them against retention after joining and improve them for next year.
The problem
Assessing people with no work history, at volume, in a short window
Applications arrive in a narrow window, and you cannot add interviewers overnight. Graduates have no employment record, so even when you write the criteria down and hand them out, interviewers still look at different things.
- Applications concentrate right after information sessions, and the lead time to a first interview stretches
- Several interviewers are involved, so the same criteria still produce different scores
- Application essays are written with generative AI, so the writing no longer shows how the person thinks
- Whether that year’s criteria were right is hard to review with any evidence
Building the criteria
Put it in a form you can score from what is said
What do you look at instead of a work history? You can take the Japanese Ministry of Economy, Trade and Industry’s Fundamental Competencies for Working Persons as they are. The framework has three competencies and twelve elements, and was defined in 2006 to capture the basic capabilities a workplace needs, at the graduate stage.
Write those elements as five levels in the form of BARS (behaviourally anchored rating scales) and you can score what is said in the interview directly. One element from each of the three competencies looks like this.
Taking initiative
Initiative
Found a problem nobody had raised, brought others along, and got it started
Volunteered for work outside their remit and carried it to the end
Went beyond what was asked within their own remit
Did what was assigned, as it was assigned
Can name an activity, but not what they themselves did
Thinking it through
Problem identification
Questioned the premise of the question and redefined the problem worth solving
Separated several causes and judged which one mattered
Restated the problem concretely and named candidate causes
Can restate the problem, but the causes stay undifferentiated
Describes what happened without framing it as a problem
Working in a team
Listening
Checked what the other person had left unsaid and resolved the mismatch before moving on
Restated the other person’s intent in their own words and confirmed it before answering
Heard the other person out and answered on that basis
Listens, but the answer drifts back to what they had prepared
States what they wanted to say, with little connection to what was said
Sources
Element names — Ministry of Economy, Trade and Industry, “Fundamental Competencies for Working Persons” (proposed 2006; three competencies, twelve elements)
Scale format — Behaviorally Anchored Rating Scales (BARS). Smith, P. C., & Kendall, L. M. (1963) “Retranslation of expectations: An approach to the construction of unambiguous anchors for rating scales,” Journal of Applied Psychology. A rating scale whose levels are fixed by descriptions of behaviour.
The L1–L5 wording — written by us, following that format. It is not officially defined.
Whether you use a framework or not, what separates AI interview services is whether these five levels can be rewritten in your own words. You can keep the public framework as it is, or replace it with your own competency definitions and behavioural models. Where criteria are free text, the definitions you already use in performance reviews go straight in.
What to settle
Three things to settle before you start a graduate intake
01
Who decides the criteria
Borrow the vendor’s standard frame, or write definitions like the ones above yourself. This is what changes which type of service fits.
02
The same yardstick as your interviewers
If you can write L1–L5 yourself, the AI and your people judge on the same basis, and the second round inherits the scores with the quotes behind them.
03
Whether it reaches past the offer
For graduates, retention is the point. Set the score on each criterion against who is still there a year later and you learn whether that year’s criteria were right.
Running it
How the year turns
A graduate intake is not a single-year job. One turn runs from the end of selection, through the review of that year, to the changes you make for the next.
01
Information session and application
Put a QR code in the session materials and the session itself becomes the way in. You can also have the AI explain the business and training before the interview. No diary to negotiate, so they reach the interview while their interest is high.
02
First round
The same criteria reach everyone. Each score keeps the remark it rests on, and the recording, so you can trace why it was given.
03
Handover to the second round
Points worth probing are passed on. The interviewer changes; the yardstick does not, and the conversation picks up where it left off.
04
Placement
Put the same criteria to the people already inside and you see, from real data, which criteria your strong performers score on. Set that against an offer holder’s scores and you have material for placement.
05
Next year’s criteria
Set the scores from selection against who stayed, judge which criteria did the work, and revise them. The change is yours to make; there is no vendor model update to wait for.
Into next year’s selection
Shown as an example: one graduate-hiring year in five stages. Where the lines fall depends on your own selection process.
Stages 4 and 5 only turn when the criteria are yours. A borrowed standard frame still runs the first round, but that frame was built separately from your performance-review system, so it is not the same yardstick as the one used after joining. What you can improve for next year is only what you defined yourself.
Related pages
More detail, by topic
The threads of a graduate intake run across several pages. Start with whichever is closest to your question.
Run first interviews in parallel when applications concentrate, without adding interviewers.
When scores differ by interviewer. One set of criteria for everyone, with the reasons on record.
How far you can narrow the field before the interview, when the paperwork is not enough.
How to choose an AI interview tool →
Five layers and six questions. Who holds the criteria changes which service fits.
AI interview services: nine compared →
Nine services in four types, read on questions, criteria and verification, from public information.
Rewrite your criteria, then test them on past interviews →
Re-score past interviews against rewritten criteria. What the full transcript can test, and what it cannot.
Frequently asked questions
Can an AI assess a student with no work history?
How well it works depends on whether you have decided what to look at. Answers are scored against each criterion and the remark behind each score is kept, so a person can check why the score was given. The criteria themselves are yours to define, and what replaces a degree or a CV can be settled by you, or worked out together with us.
Will it show when the application essay was written with generative AI?
Not from the paperwork alone. The AI interview probes as the conversation goes, so the material is not the written text but whether the person can answer on the spot. Gaps between a prepared answer and a follow-up show up in the probing.
Do we have to use the Fundamental Competencies framework?
No. What this page shows is an example built on a public framework. If you have your own competency definitions or behavioural models, you can set those as the criteria instead. Written in the same words as your performance reviews, selection and life after joining share one yardstick.
What size of intake does this suit?
Flex is bought per interview, with a 100-interview minimum, and is the easiest place to start. Around 100 is also where the effect becomes visible. For steady use there is Starter (240 interviews a year), and for higher volumes Growth (500 a year). See pricing
Tell us about next year’s graduate intake
Based on your expected application volume and the criteria you use today, we will show where to start. Free, about 30 minutes.