In 2016, Goldman Sachs changed the way it conducted first-round interviews with university applicants. Instead of relying on interviewers travelling to selected campuses, the bank introduced a system in which candidates answered standardised questions on video. Recruiters and business professionals could watch the recordings and decide whom to invite to the next stage. [1, 2]
The change addressed a problem that had become increasingly difficult for the bank to ignore. Goldman Sachs was competing for graduates with a wider range of employers, including technology companies, while also looking for skills beyond its traditional recruitment base. Campus visits had served the bank for years, but they placed practical limits on how many institutions and applicants its recruiters could reach.
The expansion was substantial. In its 2016 annual report, Goldman Sachs stated that it had interviewed candidates from more than 900 schools for its 2017 internship class, approximately 100 more than the previous year. In a 2019 account of the recruitment reform, Dane Holmes, then the bank’s global head of human capital management, reported that the proportion of campus applicants receiving a first-round interview had risen from fewer than 20 per cent in 2015 to almost 40 per cent in 2018. The number of institutions represented increased from 798 to 1,268. [1, 2]
Students who might previously have been excluded by geography or the bank’s campus schedule now had an opportunity to be considered. Goldman Sachs had widened access without having to send interviewers to every additional university.
Those candidates were also entering a different kind of assessment. They were recording answers for people who would watch them later. There was no interviewer present to ask what they meant, interrupt an explanation or request an example when an answer remained vague. Whatever the candidate managed to put into the recording would become the material available for the first evaluation.
The answer without a second question
An ordinary interview is a conversation, even when the questions are carefully structured. A candidate may begin by describing the background to a project. The interviewer can ask where the actual decision occurred. If the candidate describes a team achievement without explaining their personal contribution, a follow-up can establish who did what. An answer that initially sounds unconvincing may become useful once the interviewer reaches the part of the work the candidate has not yet made clear.
A recorded interview places more of that responsibility on the candidate. The answer must contain enough context, responsibility, reasoning and consequence for a later reviewer to understand it without assistance. The applicant must anticipate what the interviewer would otherwise ask.
Consider a graduate who describes a university project involving a difficult deadline. She explains how the group reorganised its work and submitted the assignment on time. Her account is clear, and the result sounds satisfactory. What remains uncertain is her own role. Did she identify the problem, negotiate the redistribution of work, implement someone else’s plan or simply describe what the group eventually did?
In a live interview, the next question could settle much of this: Which decision was yours?
The candidate might then describe discovering that two team members were working from different versions of the data, explaining the error to the group and proposing a new division of responsibility. That answer would provide evidence about what she noticed and how she acted. Alternatively, she might struggle to identify any decision she personally made.
A reviewer watching the recording receives neither answer. The original explanation may be accurate but incomplete. A note describing limited ownership would therefore need care. The recording has established that the candidate did not make her individual contribution sufficiently clear. It has not established that she made no meaningful contribution.
This limitation is not always decisive. If the purpose of the first stage is to assess concise, independent communication, the ability to construct a complete answer may itself be relevant. If the organisation is trying to assess teamwork, judgement or initiative, an incomplete account may leave the very capability under examination unresolved.
The recording can be watched again, perhaps by several people, and its wording can be compared with the same question answered by other candidates. That consistency is useful. It does not supply the question that was never asked.
Two graduates and an investment decision
A similar problem appears when a selection exercise looks more directly related to the work.
Imagine two applicants for a junior financial analyst position. Both receive a short business case concerning a proposed technology service. The project requires an initial investment of £2 million, and the commercial forecast estimates £3 million in revenue during the first year. Management wants a recommendation on whether to proceed.
The figures initially look attractive. The company would invest £2 million and expects to generate £3 million in revenue. But revenue is the money earned from selling the service, before the costs of operating it have been fully accounted for. The company may still have to pay for customer acquisition, technology support, staffing, maintenance and other ongoing expenses. If those costs are high enough, the project could lose money despite meeting its revenue forecast.
The projected income may also depend on assumptions about how many customers subscribe and how long they remain. A forecast built around strong customer retention becomes less attractive if customers leave earlier than expected. Without examining these assumptions, a recommendation based on the headline figures would be incomplete.
The first applicant recognises the commercial opportunity and gives a prompt answer:
“I would recommend a limited launch rather than committing to the full rollout. We could test demand, monitor costs and expand once the early results support the investment.”
Her recommendation is intelligible. She has identified an approach that limits exposure while leaving room for the project to grow. The response gives the evaluator a decision and a way of explaining it.
The second applicant approaches the information differently:
“Before recommending approval, I would want to understand the operating costs and how the revenue forecast was constructed. The £3 million figure does not tell us whether the service will be profitable. Customer retention could also change the result substantially.”
He has identified weaknesses in the financial information. His analysis is relevant, but he has not yet told management what to do. If the question requires a decision, his answer remains unfinished.
Both responses contain potentially useful evidence. The first candidate has proposed a plausible course of action under uncertainty. The second has recognised that the headline figures are insufficient to establish the project’s financial attractiveness. Neither response, as it stands, gives the evaluator a complete basis for comparing their judgement.
The first applicant’s answer would be easy to record as commercially minded, decisive, pragmatic. The second might receive analytical, cautious, slow to commit. Those descriptions would have some basis in what the applicants said, but they would also risk turning a difference in response style into a broader distinction between their abilities.
An interviewer could pursue the first recommendation:
“Which missing cost would most affect your decision to launch, and what would make you stop the pilot?”
The candidate might explain that she would limit the investment until the expected customer contribution covered the operating costs and the company had evidence that retention was sufficient. Her original recommendation would then rest on a more substantial financial argument.
She might instead reveal that she had not considered operating profitability at all. The proposal for a limited launch would still sound commercially sensible, but its apparent financial discipline would be weaker than the first answer suggested.
The second applicant could be asked:
“Management has to decide this week. The missing figures will not be available in time. What would you recommend?”
He might propose a small, reversible commitment, identify the risk that would justify postponement and explain what additional evidence would change his position. His earlier caution would then become part of a workable decision.
Alternatively, he might continue requesting information without explaining how he would act if the uncertainty could not be removed. That would give the evaluator a legitimate concern about his ability to reach a decision under constraint.
These questions would change the comparison. The panel would have evidence of how each applicant responds when the weaknesses of an initial answer are examined. Confidence might survive scrutiny or lose its foundation. Caution might produce sound judgement or remain an obstacle to action.
In a recorded first-round interview, those possibilities remain largely hypothetical until another assessment takes place.
What the recording actually measures
The difference between live and recorded interviewing becomes clearer when we examine the research.
In a 2017 experiment, Markus Langer, Cornelius König and Kevin Krause compared asynchronous digital interviews with live videoconference interviews. Participants were randomly assigned to one of the two conditions, and evaluators rated their interview performance. Those in the recorded interview condition received higher ratings. They also described the experience as less personal and reported greater privacy concerns. [3]
The finding is interesting partly because it challenges an easy assumption that removing direct human contact necessarily produces poorer interview evaluations. In this experiment, the recorded format produced more favourable ratings.
Those ratings concerned performance in the assessment. The study did not establish that the candidates receiving higher scores would subsequently perform better at work. A response may become more favourable to evaluators because the format allows candidates to present themselves more effectively. Whether that improvement reveals greater job-relevant capability requires further evidence.
Preparation time offers one explanation for the difference.
Johannes Basch and colleagues investigated how the time available before a recorded answer affected interview performance. Their 2021 research found that additional preparation time improved ratings, with active preparation helping explain the effect. Participants used the available time to prepare their responses more effectively. The study did not find support for the suggestion that the performance improvement was explained by greater use of impression-management tactics. [4]
For a candidate, an extra minute may be enough to remember a more suitable example, decide which details belong in the answer and bring the important decision forward. That can improve the quality of the information available to the evaluator. A person with genuine experience may simply become better able to explain it.
Preparation may also improve an answer whose reasoning remains relatively weak. The applicant can organise a plausible recommendation without necessarily identifying every assumption on which it depends. The evaluator still has to judge what the organised response demonstrates.
Actual interview systems also differ in how much preparation and revision they allow. A 2022 study by Patrick Dunlop, Djurre Holtrop and Serena Wee examined records from an Australian asynchronous video interview provider. The data covered more than 600,000 candidates and over 12,000 interview templates. Most interviews contained four or five questions. Typical settings gave candidates 30 seconds to prepare and two minutes to record a response. Candidates were rarely permitted to preview questions or record another attempt. The researchers also found that organisations frequently retained the platform’s default settings. [5]
Those figures describe the provider examined in the study. They are not evidence of the exact settings Goldman Sachs used.
They do, however, show how strongly an apparently technical configuration can influence the work expected of a candidate.
A two-minute answer may be sufficient to describe a straightforward decision. It may be less suitable for reconstructing a complicated project involving several departments, uncertain responsibilities and an intervention whose results appeared only months later. The candidate has to choose which parts of the account can survive the time limit.
A well-rehearsed applicant may know that the evaluator expects a concise problem, action and result. Another candidate may spend too much time explaining why the problem was difficult. The first produces a more complete recording. The second may reach the important decision just as the recording ends.
That difference tells the evaluator something about the applicants’ ability to produce a short, self-contained account. Whether it also tells the employer which applicant is better at solving complicated problems depends on the work the role requires.
A junior professional preparing a written financial analysis can normally inspect the figures, reconsider assumptions and revise a recommendation. A client-facing analyst presenting a time-sensitive decision to senior management may have much less room. Both are plausible demands of the job, but they call for different kinds of assessment.
The employer needs to know which demand the recording is intended to represent.
Why consistency is not enough
Graduate recruitment involves decisions made before employers have much direct evidence of how applicants will perform in the workplace. Degrees, academic results, internships and project experience provide information, but none offers a complete view of future performance.
Michael Spence’s work on labour-market signalling describes the underlying information problem. Employers must make decisions about people’s future productivity without being able to observe that productivity directly at the point of hiring. They therefore use information available beforehand to form expectations about capability. [6]
A degree may tell an employer something about a candidate’s educational achievements and preparation. A completed project may offer evidence of technical knowledge or persistence. An interview may show how a person organises an explanation or approaches a problem under scrutiny. Each source reveals something under particular conditions.
The employer’s task is to determine how strongly that information is connected to the work the candidate would actually be expected to do.
An interview can be designed to investigate that connection systematically. In a structured interview, candidates receive questions linked to relevant job requirements, and their answers are evaluated against defined criteria. The criteria help interviewers distinguish stronger from weaker demonstrations of the capabilities being assessed.
For the junior analyst position, the criteria might include identifying materially incomplete information, recognising financial risk, comparing alternatives, explaining a recommendation and adjusting it when assumptions change.
A candidate who identifies the missing operating costs could receive credit for that observation. A candidate who proposes a limited launch could receive credit for developing an actionable response. Neither would necessarily deserve a high rating for financial judgement until the assessment had established how the applicant arrived at the conclusion and whether the reasoning could withstand appropriate examination.
The structure helps the evaluator keep those distinctions visible. Asking consistent questions reduces the opportunity for different interviewers to pursue entirely different interests. Defined criteria also give the panel something more reliable than general confidence, likeability or the ease of conversation when comparing applicants.
The research supports taking structured interviews seriously. In a 2022 reassessment of the predictive validity of personnel selection methods, Paul Sackett and colleagues found that structured interviews ranked particularly highly in their ability to predict subsequent work performance. The researchers also concluded that earlier estimates of predictive validity for several commonly used methods had been too optimistic. [7]
The evidence therefore supports using structured interviews while remaining careful about the conclusions they can sustain.
Standardising a question makes candidates easier to compare, but the question must still examine something relevant to the job. If every graduate is asked to deliver an impressive two-minute account of leadership, the organisation may obtain consistent ratings of verbal presentation without establishing how effectively those graduates would perform the analytical or operational tasks ahead of them.
A selection method can be applied with great consistency while producing evidence about a capability the employer has not adequately connected to the role.
The same problem arises when evaluators give a recorded answer more meaning than its content supports.
How the assessment note changes the candidate
Return to the two graduate applicants.
The first recording contains a clear proposal for a limited launch. The second contains a more explicit examination of financial uncertainty but no final recommendation.
An evaluator reviewing several recordings may find the first account easier to use. It offers a conclusion, a course of action and a familiar form of business reasoning. The note can be written quickly. The second response needs a more careful description because the useful observations and the incomplete recommendation have to remain separate.
The comparison could reasonably produce these initial notes:
Candidate A: Recommended a limited launch and proposed monitoring performance. Clear commercial communication. Cost assumptions and approval threshold not established.
Candidate B: Identified missing operating costs and retention assumptions. Demonstrated attention to financial uncertainty. No final recommendation within the recorded response.
Both notes retain the strengths and limits of what was actually observed.
A less careful assessment could produce a very different record:
Candidate A: Strong commercial judgement; confident and decisive.
Candidate B: Hesitant; insufficient business orientation.
The second pair of notes is easier to carry into a hiring discussion. The candidates have become recognisable types. One sounds ready for business, the other more comfortable with analysis than action. Yet the recording has not established that the first would make the better investment decision, or that the second would struggle to make decisions at work.
The wording has moved beyond the observation. In the first candidate’s case, clarity has acquired the status of judgement. In the second, an unfinished recommendation has begun to sound like a more general inability to decide.
A reviewer encountering only those final descriptions may have no reason to return to the original recordings. The distinction between what the candidates demonstrated and what the evaluator inferred is no longer visible.
There is a further complication. Candidates can become skilful at supplying the kind of account that interviewers find convincing.
Research by Nicolas Roulin, Adrian Bangerter and Julia Levashina examined honest and deceptive impression management in employment interviews. Their studies found that interviewers often struggled to identify how candidates were managing the impressions they created, particularly when the tactics were deceptive. [8]
This does not make a polished answer suspicious by default. Candidates have good reasons to prepare and present their experience clearly. Professional communication is valuable, and interviewers should give it appropriate credit.
The difficulty arises when presentation begins carrying a larger claim about competence than the underlying evidence can support.
One candidate may be fluent because she understands the work deeply and can explain it well. Another may know the expected language of an effective answer while having only limited experience behind it. A third may have substantial relevant ability but struggle to make it visible within a short recording.
A carefully designed assessment has to distinguish those possibilities as far as its available evidence allows. Sometimes the recording supplies enough information. Sometimes it establishes a concern worth investigating further. Occasionally it shows that an essential requirement has not been demonstrated under conditions sufficiently similar to the job.
What the process cannot fairly do is treat its own difficulty understanding an answer as proof that the candidate lacked the relevant ability. The difficulty may arise from the candidate’s explanation, the assessment format, the evaluator’s expectations or some combination of the three.
After the first-round recording
Goldman Sachs’s recruitment reform was successful in one documented and consequential respect. The organisation made first-round interviews available to a larger proportion of applicants and reached graduates from substantially more institutions. This widened the pool from which its recruiters could identify people for further assessment. [1, 2]
The reported figures do not establish whether video-based first-round assessment predicted later job performance more accurately than the earlier approach. The bank’s own account also describes a process in which recorded responses were followed by in-person interviews for selected candidates. The recording was one part of a broader selection procedure.
That distinction is useful for understanding what an early assessment needs to accomplish. It may be designed to identify sufficiently promising candidates for closer examination, rather than establish everything required for a final hiring decision. Its information can therefore be limited without being useless.
There is still a consequential choice at the point where someone decides which recordings justify a second stage.
A reviewer may see a candidate who produces a concise, persuasive recommendation while leaving its financial assumptions unexplored. Another may identify a significant risk but fail to give the complete answer the format appears to demand. Advancing either candidate can be defensible. Rejecting either may also be defensible if the missing capability is essential to the role and the assessment has tested it appropriately.
The quality of the decision depends partly on whether the reviewer can say what was established and what remains unknown.
A first-stage note could preserve a missing recommendation as a limitation in the recorded exercise while recognising the financial reasoning that was demonstrated. Another could credit a clear commercial proposal while identifying the assumptions that still require examination. Those distinctions give the next interviewer a useful starting point. They also prevent a strong first impression from quietly exempting one candidate from the scrutiny another must undergo.
The recorded answer remains exactly as the candidate gave it. An evaluator can watch it again, compare it with other responses and return to the evidence on which an initial judgement rested. What the recording cannot show is how the candidate would have responded to the question that was never asked.
Where that missing answer could change the hiring decision, the uncertainty needs to remain visible. Otherwise, a candidate may progress because the recording was easy to assess, or disappear because the relevant ability never became sufficiently clear within the format the employer chose.
Sources
Goldman Sachs (2017). 2016 Annual Report.
https://www.goldmansachs.com/investor-relations/financials/current/annual-reports/2016-annual-reportHolmes, D. E. (2019). Expanding the Pool: How Goldman Sachs Changed the Way It Recruits. Harvard Business Review, May–June.
https://hbr.org/2019/05/expanding-the-poolLanger, M., König, C. J., & Krause, K. (2017). Examining digital interviews for personnel selection: Applicant reactions and interviewer ratings. International Journal of Selection and Assessment, 25(4), 371–382.
https://doi.org/10.1111/ijsa.12191Basch, J. M., Brenner, F., Melchers, K. G., Krumm, S., Dräger, L., Herzer, H., & Schuwerk, E. (2021). A good thing takes time: The role of preparation time in asynchronous video interviews. International Journal of Selection and Assessment, 29(3–4), 378–392.
https://doi.org/10.1111/ijsa.12341Dunlop, P. D., Holtrop, D., & Wee, S. (2022). How asynchronous video interviews are used in practice: A study of an Australian-based AVI vendor. International Journal of Selection and Assessment, 30(3), 448–455.
https://doi.org/10.1111/ijsa.12372Spence, M. (1973). Job Market Signaling. The Quarterly Journal of Economics, 87(3), 355–374.
https://doi.org/10.2307/1882010Sackett, P. R., Zhang, C., Berry, C. M., & Lievens, F. (2022). Revisiting meta-analytic estimates of validity in personnel selection: Addressing systematic overcorrection for restriction of range. Journal of Applied Psychology, 107(11), 2040–2068.
https://doi.org/10.1037/apl0000994Roulin, N., Bangerter, A., & Levashina, J. (2015). Honest and deceptive impression management in the employment interview: Can it be detected and how does it impact evaluations? Personnel Psychology, 68(2), 395–444.
https://doi.org/10.1111/peps.12079
Source note: The account of the Goldman Sachs recruitment reform draws on the bank’s annual report and Dane Holmes’s published description of the changes. The two financial analyst candidates and their assessment notes are hypothetical examples. The interpretation of how recorded answers become evaluation evidence is the author’s analysis.