Job Trials Over Interviews
Interviews don't predict performance; make the interview as close to the real job as possible.
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 4
- Confidence
- 87%
A hiring philosophy that treats interviews as poor predictors of on-the-job performance and replaces or supplements them with real work: job trials, an intense 30-60-90 day evaluation, and internships as scaled trials. The race-car logic: to hire the best driver, put them in the car rather than quizzing them.
Origin
Thawar's approach across his career, where he 'almost didn't interview anybody' and just had people come in and work (all pair programming). Backed by the well-known research that interviews poorly predict performance. Scaled at Shopify via a plan to hire 1,000 interns in 2025.
Core principles
- 01Interviews are a known-weak predictor of job performance; real work product is a strong one.
- 02Make the interview process as close to the real job as possible so you assess the actual skills the role needs.
- 03Transparency about fit is good for both sides; a fast 'not a fit' frees the person to be great elsewhere.
- 04A resume tells you what someone did, not why; the 'why' is the interesting signal (curiosity, range).
- 05Internships are the ideal trial: four months of real work product beats any number of interview hours.
How to run it
- 1
Make the interview mirror the real job
Design the evaluation to look like the actual work (Thawar's startup had candidates pair-program, since that's how the team worked). The closer the assessment is to the job, the more accurately it predicts on-the-job skill.
Pro tip Use the race-car analogy: you can't ask a driver many useful questions, you put them in the car. Do the same for the role.
- 2
Add a life-story step to test for curiosity and range
Include an interview step that traces the candidate's past decisions to see whether they're a curious person with range. The signal isn't what roles they held but why they moved between them.
Pro tip A resume tells you what you did but not why; probe the 'why did you switch from X to Y' because that's where curiosity and range show up (cf. David Epstein's 'Range').
- 3
Evaluate hard in the first 30-60-90 days
Treat the first 30-60-90 days as the real trial. Spend that window verifying that what the person brings actually matches what the role needs, with an explicit survey ('how happy are you with the person you hired?') and honest feedback conversations.
Pro tip Be transparent: if it's not a fit, both sides benefit from finding out fast so the person can go be amazing somewhere they fit.
Watch out This deliberately raises early attrition (Thawar saw ~20% attrition before 90 days) in exchange for very low attrition after (under 1%), because people know exactly what they're getting into.
- 4
Scale trials through internships
Where volume makes individual trials impractical, use internships as scaled job trials. A four-month internship gives real work product and lets both sides assess fit far better than eight hours of interviews.
Pro tip Shopify planned 1,000 interns in 2025, i.e. 1,000 job trials, then converts the top performers to full-time; interns in pair programming are often more intense than full-timers.
In the wild
At his startup, Thawar considered two ML candidates: a PhD who taught at a university and was employee-recommended (an apparent no-brainer), and a guy he met at a coffee shop who had never held a software job but was deeply interested in machine learning. He didn't let resumes bias him; both pair-programmed in the real environment.
→ 'Person A' was let go within weeks as a poor culture fit; the coffee-shop hire is still at Shopify today as a phenomenal ML engineer, in what was his first software job.
Thawar noticed a waitress running a busy restaurant with exceptional organization, asked what she did outside of it, and offered her a role at Extreme Labs. She started as receptionist, became his admin, then a recruiter he trained.
→ She rose to run an HR function, parlayed the experience into finishing her university degree, and is now a director of HR at a company; she also recruited other strong people from that restaurant onto good career paths.
Common mistakes
Trusting interviews to predict performance
It's an open secret that interviews poorly predict on-the-job performance, people who interview well underperform and vice versa. Relying on interview signal over real work product leads to bad hires and missed great ones.
Interviewing after an internship
Thawar mocks companies that run a four-month internship and then interview the intern for full-time: eight hours of interviews teach you nothing you didn't already learn from four months of real work. Look at the work product instead.
Letting the resume bias the evaluation
A resume tells you what someone did, not why, and prestige signals (PhD, referrals) can mislead. Judging on the resume rather than real-environment work product cost would have cost Thawar his best ML hire.
Is it for you?
Best for
Hiring managers and founders who can create real work trials (pair programming, internships) and want to hire on demonstrated performance rather than interview polish.
Not ideal for
High-volume hiring where individual trials aren't feasible without an internship pipeline, and roles where a poor early-attrition trial period would be unacceptable to candidates.
From the transcript
“interviews are not a good predictor of performance we know this”
“we want to make the interview process as close to the real job as possible”
“what a great interview process because you now have real work product from somebody for four months”
“it tells you what you did but it doesn't tell you why you did those things”
From the episode
How Shopify builds a high-intensity culture
Farhan Thawar (VP and Head of Eng)