Most interviews measure how well someone interviews. A structured interview measures whether they can do the job. The difference is not the questions. It is deciding what good looks like before you meet anyone, then scoring every candidate against that instead of against each other.
The market gives you less room than it used to. The Ministry of Statistics and Programme Implementation says "Unemployment Rate (UR) in usual status (ps+ss) was 3.2% for both male and female of age 15 years and above during July 2023 to June 2024." Good applicants are employed and in other processes.
What goes on the scorecard?
Four to six attributes the role needs. For each one, write what a weak, acceptable and strong answer sounds like. Do this before screening starts.
If you cannot describe a strong answer in advance, you are not ready to interview for that attribute. You will rate confidence instead of competence, which is how teams hire people who present well and deliver late.
| Attribute | Weak | Strong |
|---|---|---|
| Statutory depth | Names PF and ESI, cannot explain a threshold or a deadline | Explains PF, ESI, professional tax and TDS, and what changes across states |
| Cycle ownership | Did data entry inside someone else's cycle | Closed a full monthly cycle end to end, including corrections |
| Error handling | Says mistakes do not happen | Describes a specific error, how it was found, what changed after |
| Scale | Single location, under 100 staff | Multi-state, multi-location, headcount close to yours |
Write the right-hand column before you interview. It takes 20 minutes and it stops the panel converging on whoever was most articulate. The statutory items are checkable against the Employees Provident Fund Organisation and Employees State Insurance Corporation rules.
Why ask every candidate the same questions?
Because unstructured interviews drift. The first candidate gets asked about a project. The third gets asked about their weekend. The comparison between them is worthless.
Fix the question set per attribute and keep the order. Follow up on substance, not rapport. When an answer is general, push once for specifics: what exactly did you do, what did you decide, what happened. One good follow-up tells you more than 3 new questions.
Which questions must stay out of the room?
Some questions are common in Indian interviews and should not be. Age. Marital status. Pregnancy or family plans. Caste. Religion. Mother tongue. None bear on capability, and asking exposes you.
This is not abstract. Female labour force take-up for ages 15 and above moved from 23.3% in 2017-18 to 41.7% in 2023-24, and female worker population ratio from 22.0% to 40.3% over the same period. You are interviewing a far more mixed pipeline than the questions in your old template assumed. If a rule is genuine, test it directly. Instead of asking a woman whether she plans to have children, state the travel or shift expectation and ask whether she can meet it. Same information, obtained legitimately, asked of everyone. The Ministry of Labour and Employment sets the equal-opportunity framework these sit inside.
How many questions fit the time?
| Length | Questions | What it can assess |
|---|---|---|
| 30 minutes | 4 to 6 | Two attributes, properly |
| 60 minutes | 8 to 12 | Three to four attributes |
| Over 90 minutes | Split it | Fatigue beats signal after this |
Cramming 12 questions into half an hour produces 12 shallow answers. Fewer questions, each pushed to specifics, is the better trade.
When should you score?
During, then at once after. Memory across 6 applicants in a day is not reliable, and the strongest memory is usually the most recent interview rather than the best one.
Take notes against the scorecard as you go. Commit a score within minutes of the candidate leaving. Where more than one person interviews, score independently before discussing. A panel that talks first produces one opinion held by 3 people.
What do you do when panel scores disagree?
Treat it as information, not a problem to average away. A split usually means one panellist tested the attribute and the other tested rapport.
Go back to evidence. Ask each scorer what the candidate actually said that produced their score. If one side has a quote and the other has an impression, you have your answer. If both have quotes and still disagree, the attribute deserves a short follow-up rather than a coin toss.
How do you know the interview predicts anything?
Three numbers. Interviews per offer. Offers per acceptance. And how 6-month performance ratings compare against interview scores. The third is the only real test.
If strong interview scores do not track with strong performance at 6 months, the scorecard measures the wrong attributes. Change the scorecard rather than interviewing more people against it. For wider context the same PLFS release notes that "The overall WPR among persons of the same age group in usual status (ps+ss) has increased from 56.0% during July 2022 to June 2023 to 58.2% during July 2023 to June 2024", per the Ministry of Statistics and Programme Implementation.
Keep the whole trail in one place so the decision is reconstructable: who interviewed, what was scored, on what evidence. 247HRM applicant tracking holds scorecards, notes and stage history on the candidate record, with panellist review tracking so each panellist files their own review before anyone discusses it.