Does an AI interview predict sales quota attainment?

Updated 1 November 2026

Can an AI interview tell your CRO which sales candidate will hit quota? It is a reasonable question, but the published evidence does not yet establish that link. Some studies show that structured interviews and sales role-plays relate to job performance. Others connect AI-led interviews with hiring or retention outcomes.

The closest evidence is an assessment centre built for sales roles. Its ratings predicted objective field sales performance one year later, and the enhanced version reached a true-score validity of .49 (Kückelhaus, Titze and Blickle, 2022). Structured interviews lead all selection methods at .42 against overall job performance (Sackett et al., 2022).

That distinction matters when a vendor makes a prediction claim. Before you use the claim in a buying decision, ask what score was tested, what outcome it predicted, how long the researchers followed people and whether the result was observed in a real sales team.

Fabric's sales role-play interview on its SDR page is a live role-play assessment. The page describes a sales signal in relation to quota attainment, but the research reviewed here does not establish that a Fabric interview score predicts quota. This article examines what the evidence supports and what would be needed to validate that claim.

What the closest studies measured, and where each stops

Nothing published puts an AI interview score next to a rep's quota. Four studies sit near it, and each stops short. The picture below puts them in order, closest to quota first.

Look at the pattern. The two studies that used sales output did not use AI. The two studies that used AI did not use sales output. Nobody has joined the two.

Every study here reports the same kind of number, so it helps to know what that number can and cannot tell you.

How to read a validity coefficient without overstating it

A validity coefficient describes the relationship between an assessment score and a defined outcome. The outcome matters. A score may relate to a manager's rating of performance but have a much weaker relationship with revenue or quota. A coefficient is not a percentage of candidates who will succeed, and it does not guarantee an individual result.

Two things change what the number means. The first is the outcome it is matched to, which researchers call the criterion. The second is whether the figure is as observed or has been corrected.

A corrected "true-score" figure answers a different question. It asks how well the score would predict a perfectly measured outcome. It does not tell you how well the score predicted the numbers you actually have.

For a buyer, the most useful figure is the observed relationship between the interview score and the real outcome you care about. Corrected or true-score estimates can help researchers account for measurement error, but they should not be presented as if they were the correlation an employer observed in its own sales team.

Try it with your own numbers. The default is an observed correlation of .30 against an outcome measured with a reliability of .60.

Slide the second control left and the corrected figure climbs, even though nothing about the real prediction changed. The same score simply looks stronger, which is why every number below carries its outcome, its study and its year.

The 1991 role-play study: what it reports

Squires, Torkel, Smither and Ingate tested a 30-minute role-play test run entirely by telephone for telemarketing representatives (Journal of Occupational Psychology, 1991). Each person did four role-plays, selling a service contract. Two raters scored them independently.

The combined ratings correlated significantly with both objective and subjective measures of sales performance. The raters agreed well: reliability was .80 in the concurrent study and .76 in operational use. Those two figures describe how consistent the ratings were. They are not validity figures.

Three limits matter.

First, the study was concurrent. It compared role-play scores with the performance of people already in the job. That shows scores track performance. It does not show they forecast it.

Second, the outcome was sales performance measures, not quota attainment.

Third, the public record of the paper gives no validity coefficient. None is quoted here. The full text is closed-access, so a specific correlation you see attributed to it second-hand deserves a request for the page number.

The closest study to a prediction

Kückelhaus, Titze and Blickle studied an assessment centre designed for sales positions (Journal of Personal Selling & Sales Management, 2022). An assessment centre is a set of exercises scored by observers. Most research on them covers managers. This one was built for sales.

The ratings showed good agreement between observers and measured distinct things. Their validity for the overall assessment rating sat in the normal range for managerial assessment centres.

The authors then tried a different approach. Motivation for sales success, combined with social competence, predicted field sales performance one year later. It explained extra differences in objective sales that the exercise scores and the overall rating had missed.

The new approach raised operational validity by 25% over the traditional one. Its true-score validity was .49, against objective field sales measured a year after the assessment.

This is the closest the literature gets to quota, for one reason. It looked forward, and it used real sales output.

Two cautions come with it. The .49 is a true-score figure, so it is corrected for imperfect measurement of the sales outcome. The observed correlation is lower, and it is not stated in the paper's abstract. And this is an assessment centre, not an AI interview.

How good is the best method after the 2022 correction

Sackett, Zhang, Berry and Lievens re-checked decades of selection research (Journal of Applied Psychology, 2022). They found earlier studies had over-corrected for "range restriction". In plain words, you only see outcomes for people who were hired. Researchers stretch the numbers to estimate what the whole applicant pool would show, and the old stretch was too large.

The same methods still rank near the top. Most estimates fell, several by .10 to .20. Structured interviews now come first.

Read the table with its outcome in mind. These are not sales figures. They answer "how well does this method predict how a supervisor rates someone in any job?"

Two things are still worth taking from it. Structured interviews at .42 and work samples at .33 are among the stronger selection methods in the re-analysis. The older figures, .51 for structured interviews and .54 for work samples, are still widely quoted. If a vendor uses those numbers as general benchmarks, ask whether they rely on pre-2022 estimates, what outcome was measured and which correction was applied.

A manager's rating is not revenue

For sales, the outcome you choose changes the answer. Vinchur, Schippmann, Switzer and Roth pulled together the research on what predicts how salespeople perform (Journal of Applied Psychology, 1998, a meta-analysis). They reported two outcomes separately: supervisor ratings and objective sales.

General cognitive ability correlated .40 with supervisor ratings of sales performance and only .04 with objective sales. Achievement ran the other way: .25 against ratings and .41 against objective sales.

Potency, the forceful and sociable end of extraversion, scored about the same on both (.28 and .26). Age predicted ratings (.26) but not objective sales (-.06). Two measures did well against ratings, and the abstract reports them as corrected figures: biodata at .52 and sales ability inventories at .45. Neither has an objective-sales figure.

The lesson is practical. Anything sold to you as "predicts sales performance" without naming the outcome is answering a question you did not ask: cognitive ability looks strong against one outcome and absent against the other. These are 1998 figures, so read the pattern between the two columns, not the exact size of any one number.

What AI interview research has measured

Two studies connect AI-run interviews to real outcomes. Both sit one step from quota.

Jabarian and Henkel ran a pre-registered field experiment, meaning the study plan was published before the results existed (working paper, September 2026 version, not peer-reviewed). It covered 70,884 applications across 48 job postings and 41 client accounts in the Philippines, for entry-level customer service roles at a recruitment outsourcing firm.

Of the applicants assigned to the AI voice agent, 5.89% (2,363 of 40,103) were still working one month after starting. That figure counts every applicant who was interviewed, including the many who were never hired. The outcome is 30-day retention in customer service, not a sales number.

Chakraborty, Chiong, Dover and Sudhir, writing in Marketing Science, built an AI model to score 195 fifteen-minute sales role-play videos, as reported by Yale SOM in January 2025. Selecting on the AI's assessment gave a workforce of about 40% better quality than choosing at random.

The strongest signal was information density, meaning how much the seller actually said. The participants were university students selling a software product, and the outcome was rated sales ability, not sales output.

AI interview research has connected interview processes with hiring outcomes and has examined AI scoring of sales role-plays. It has not established that a specific AI interview score predicts quota attainment. That is the evidence gap a buyer should ask a vendor to close.

Five questions to ask any vendor who says their tool predicts quota

Use these on every vendor, Fabric included. Tick each one the vendor answers with specifics.

A vendor who answers all five is rare. The answers tell you more than the headline number does.

What Fabric's sales simulation can assess, and what remains unproven

Fabric's sales simulations work for any role that cold calls or emails. The AI plays the prospect and pushes back as the candidate talks, adapting the way a prospect would. The cold-call exercise produces a numeric sales score, and it runs in languages other than English. Recruiters can review every result.

In the terms used on this page, a standardised, scored, two-way sales exercise is a structured interview with a work-sample component. The Sackett table puts structured interviews first and work samples fourth, at .42 and .33 against overall job performance. Work-sample exercises are also what assessment centres are built from, and the sales assessment centre in the Kückelhaus study, extended with measures of motivation and social competence, predicted field sales a year out.

That is a claim about the assessment method, not proof of a quota prediction. This article does not establish a quota-validity coefficient for Fabric or any other individual tool. Ask for validation using the score, the role and the sales outcomes your team actually tracks.

FAQ

What actually predicts whether a sales rep will hit quota?

Among the predictors reported for objective sales in the 1998 meta-analysis, achievement, a part of conscientiousness, is strongest at .41 (Vinchur, Schippmann, Switzer and Roth, 1998, meta-analysis of salespeople). General cognitive ability reaches .40 against supervisor ratings but only .04 against objective sales, so the two outcomes behave very differently.

How valid are structured interviews for hiring?

Structured interviews are the top-ranked selection method at .42 against overall job performance (Sackett, Zhang, Berry and Lievens, Journal of Applied Psychology, 2022). That is a revision downward, because earlier estimates overapplied corrections for range restriction.

Do sales role-plays predict sales performance?

In a 1991 concurrent study of telemarketing representatives, ratings from a 30-minute telephone role-play correlated significantly with objective and subjective measures of sales performance, with rater reliability of .80 (Squires, Torkel, Smither and Ingate, Journal of Occupational Psychology, 1991). The public record gives no validity coefficient, and the design cannot show forecasting.

Is a validity of .42 any good?

Yes, .42 is the highest estimate in the 2022 re-analysis of selection methods against overall job performance. It accounts for about 18% of the variance in that outcome, so most performance differences remain unexplained. If a vendor quotes .55 or higher as a general structured-interview benchmark, ask which study, criterion and correction produced the figure; do not assume every higher coefficient is automatically outdated.

Validate the score against performance in your own sales team

Run the exercise on one open SDR role and keep every score alongside the criteria and evidence used to produce it. Confirm the current score-editing and audit controls with Fabric before setting up the pilot. Once the new reps have ramped, compare the interview results with the attainment measures your team tracks. That gives you local evidence about whether the exercise is useful; it does not establish predictive validity on its own.

Book a demo and run the cold-call simulation against one open SDR role, or see how the sales role-play interview is set up first.

This article is for informational purposes only. Fabric's Interview Engine screens, scores, and records Round 1 interviews; it does not make the final hiring decision. The recruiter or hiring panel using Fabric remains responsible for all hiring decisions.