Best practices on preparing and doing interviews

Here’s a sneak peek of the best practices we’ve compiled drawing on the work of Prof Iris Bohnet and others.

Before: it's all about a day in the life of...
1. Decide how you want to run the interview(s)  – panel or one-on-one.	Research suggests that successive interviews with different members of the team are better than panel interviews (largely because they mitigate the risk of groupthink on the part of the reviewers). More contact is also better for getting a representative sample of the candidate. If you do multiple interviews, you should shuffle the order of who sees whom and when, allowing you to mitigate ordering effects that can influence scores (like being more generous to the first few, or kinder after a terrible interview).
2.  Decide who you want your interviewers to be – and be clear if you’re not going to weigh views equally.	Experiments indicate 3 is an optimal crowd. Other research also points to the value of having diverse panels – that is, a balance across inherent characteristics (gender, race, age) as well as job-related (i.e. seniority, background).
3. Come up with the what: 5-7 things that person needs to be able to do on the job (be specific) – convert these into work tests or scenario-based questions and score them separately on a scale.	Work-related tests have amongst the highest predictive validity and since they’re forward-looking, you’re able to see how a candidate would operate in the job, rather than what they may or may not have done in the past. Scoring each part separately is critical to improving accuracy and objectivity – it wards against a whole host of behavioural biases including the halo effect (i.e. letting one small positive experience spill over into our overall evaluation) and the peak-end effect (i.e. our predisposition to remember the peak (or trough) and end of an experience – brushing over the rest).  Additionally, you may care most about the candidate's ability to do a particular task, so be clear at the outset to help focus interviewers' efforts. Equally, you may want to rate responses to rapport-building questions, but assign them limited or no weight in the final score – scoring them explicitly will help to ensure the information is captured, but doesn’t disproportionately influence other scores.
During: execute your plan
4. Tell candidates what’s about to happen.	Help ease interviewees in by telling them that you’ll be taking them through a structured interview which could feel a little wooden, but that it’s designed to help them shine.
5. Ask questions/post tasks in the predetermined order.	Some tasks may be more challenging than others, so keeping the same order ensures candidates aren’t unfairly discouraged. It also helps with comparison.
6. Score each question throughout the interview.	Our memory can play tricks on us, so scoring immediately and individually reduces judgemental errors. Try do score it as you go, or at least as soon as you can when the candidate leaves. You can help yourself by always leaving time between candidates to score.
7. Make sure to score individually (for panel interviews, that means scoring before conferring).	We’re social beings which means we’re influenced by others’ views. This can stifle the value you get out of having multiple people assess (bringing their unique perspectives). Independent assessment is critical to avoiding groupthink and benefitting from the wisdom of the crowds.
After: let the numbers fall where they will
8. Submit the scores for analysis and let the numbers guide you – make sure to keep to your original weights.	If you’ve kept to your plan, the candidates’ total scores will reflect a balanced account of their performance. Sometimes this can lead to surprising results, but try to avoid the temptation to change.
9. If there’s a high degree of disagreement between reviewers on a candidate or if two candidates are very closely graded (i.e. if one reviewer was decisive), consider an additional interview.	Our experiments showed that three is an optimal crowd for most assessments, but you might want to call in a fourth in these instances.