Appeal from the United States District Court for the Eastern District of Wisconsin. No. 13-cv-00885 — Pamela Pepper , Chief Judge .

ARGUED MAY 19, 2026 — DECIDED JULY 13, 2026

Before BRENNAN, Chief Judge , and ST. EVE and KOLAR, Circuit Judges .

ST. EVE, Circuit Judge . Four former employees of or applicants to work for Infosys Technologies Limited, Inc. and Infosys Public Services, Inc. (together, “Infosys”) sued the company, alleging it discriminated against non-South Asians. Plaintiffs relied on an expert report by David Neumark, a labor economist who analyzed Infosys’s demographic data and Page 2concluded that chance alone could not explain the demographics of Infosys’ employee population, as the foundation of their case. T o Plaintiffs, Neumark's analysis helped prove that Infosys engaged in discrimination. But the district court excluded Neumark's expert opinions under Federal Rule of Evidence 702. The district court then denied Plaintiffs' motion for class certification because it relied heavily on Neumark's analysis. It later denied Plaintiffs' request to supplement their summary judgment briefing with different data. Finally, the district court granted Infosys’s motion for summary judgment. We affirm.

Infosys is an India-based consulting and technology services company, with office s in the United States and around the world. Globally, it employs over 190,000 people, and in the United States it has over 20,000 employees. The four plaintiffs —Gregory Handloser, Brenda Koehler, Kelly Parker, and Layla Bolten—either worked for or applied to work for Infosys, and all four are of non-South Asian background.

Handloser worked as a sales manager at Infosys. The company evaluates the performance of sales managers and other employees every six months. At the start of a review cycle, Infosys informs the sales managers of their target metrics for that cycle. At the end of the cycle, the sales manager provides a numerical evaluation of his performance, as does his supervisor. Infosys uses those entries to calculate the manager’s individual 'Performance Index.' It then weights each manager’s Performance Index against every other sales manager’s score; the output is the “Consolidated Relative Rating” score.

Page 3The best such score is a “one plus” and the poorest is a “four.” During the April-September 2012 review cycle, Handloser gave himself a poor evaluation score, as did his supervisor. As a result, his Consolidated Relative Rating was a “four,” and Infosys terminated Handloser as part of a company-wide reduction in force that targeted employees who had received a “four.”

Koehler applied to work for Infosys as a “Lead VMware/Windows Administrator”—a Milwaukee-based position working on an Infosys project with Harley-Davidson. Koehler had many of the qualifications necessary for the role, but, after interviewing her, Infosys determined that she lacked the necessary “Microsoft Active Directory” skills and decided not to hire her.

Parker worked for Infosys between 2010 and 2012 after the company took over the Harley-Davidson project, where Parker previously worked. Infosys informed Parker her position would conclude by the end of 2012. But Parker’s role continued into 2013, and, in June 2013, she began working for SoftHQ— an IT staffing company that helps Infosys fill its temporary staffing needs. Through this, Parker stayed in her prior role at Infosys, though Infosys informed Parker her new role would end once the staffing need concluded. For her part, Parker claims she believed this was a permanent position. In September 2013, Infosys ended Parker’s position and, thus, terminated Parker.

Bolten began working for Infosys as a “Test Analyst” in the company’s Washington, D.C. office in 2013. In this role, Bolten worked under Geeta Kulkarni. Bolten claims Kulkarni, who is Indian, harassed her on account of her non-Indian status. Specifically, after Bolten took a sick day Kulkarni com- Page 4mented that “stupid Americans” are “lucky” they can take sick days, and at a group meeting, Kulkarni did not let Bolten talk about her background, calling them ' American experiences,' while allowing other test analysts to share stories about their background. Bolten also raises other instances of harassment, for example when, following the Boston Marathon bombing, a coworker noted that the attacke r was Russian and that Bolten was Russian as well, while pointing a finger at Bolten, and that her coworkers, including Kulkarni, often did not speak English at work when around her, including when discussing workrelated matters .

Bolten eventually reported these concerns, and her supervisors claimed they would address them. Bolten also shifted teams at Infosys several times. After one of these reassignments, Bolten moved her belongings to a new desk, but a new team member told Bolten someone else on the team already occupied the desk and directed Bolten to a different desk. Bolten responded, 'you know what? I'm not moving anywhere, I’m not doing anything,” and resigned.

In 2013, Koehler sued Infosys, alleging the company discriminated against individuals who were not of South Asian background through its hiring and employment practices. After adding Handloser, Parker, and Bolten in an amended complaint, Plaintiffs sought to represent a class of '[a]ll individuals who are not of South Asian race or Indian, Bangladeshi, or Nepalese national origin who applied for a position with, or were employed by Defendants, and were discriminated against on the basis of their national origin or race in the last Page 5four years through the date of final judgment in this action.'¹ The complaint brought disparate treatment and disparate impact claims under Title VII and disparate treatment claims under 42 U.S.C. § 1981.

During discovery, Plaintiffs sought to introduce the expert opinions of Professor David Neumark. Neumark is a labor economist whom Plaintiffs hired to analyze Infosys's employment data. After reviewing Infosys’s demographic data, Neumark concluded, for example, that “89.39% of Infosys’ United States workforce was South Asian,” whereas only 11.45% of the workforce in the relevant domestic industry was South Asian. According to Neumark, chance could not explain this disparity, along with similar disparities he found in other analyses.

But Neumark’s analysis had a challenge. Infosys does not maintain demographic data on its employees and applicants at the level of granularity Neumark needed, so Neumark did not have a complete list of who among Infosys’s employees and applicants was of South Asian background. He did, however, have access to national origin information for Infosys employees. So, Neumark employed his fi ve -step 'namematching” methodology. First, Neumark created a list of the last name of every Infosys employee whose national origin was a South Asian country. Second, Neumark consolidated that list to last names associated with two or more employees. Third, Neumark reviewed the consolidated list and removed Page 6sixty-three last names that, to Neumark, sounded “clearly … to be non-South Asian/non-Indian.' Fourth, with his twiceconsolidated list, Neumark compared the names to the last names of Infosys employees not from a South Asian country and to the last names of Infosys applicants (Infosys did not collect national origin data on applicants). Fifth, if a name matched a last name on his list, Neumark categorized the individual as “South Asian”; if the name did not match, he categorized the individual as “non-South Asian.”

In December 2016, following Plaintiffs' motions for partial summary judgment and class certification, Infosys moved to exclude Neumark's expert opinions. Approximately three and one-half years later, in June 2020, the district court held a hearing on the motion to exclude Neumark’s testimony. Then, in September 2022, nearly six years after Infosys had moved, the district court fi nally granted the motion to exclude. Applying Federal Rule of Evidence 702, the district court found that Neumark lacked the qualifications needed to conduct the name-matching methodology, that his methodology was unreliable, and that, as a result, his opinions were irrelevant. It found him unqualified because he admitted he had no experience or training in identifying whether a name is 'South Asian” or not. And it found his methodology unreliable because Plaintiffs offered no evidence demonstrating the methodology’s reliability, testing, error rate, peer review, or scientific foundation. The court excluded Neumark’s opinions as to both Infosys employees and applicants.

In the same decision, the district court also denied Plaintiffs' motion for class certification. To meet Federal Rule of Civil Procedure 23's requirements, Plaintiffs had repeatedly Page 7relied on Neumark’s analysis in their motion. So, having excluded Neumark's opinions, the district court found that P laintiffs lacked the evidence necessary to satisfy Rule 23 and denied class certification. For the same reasons, the district court also denied Plaintiffs' motion for partial summary judgment.

Stepping back, during discovery Plaintiffs requested “[d]ocuments and ESI relating to the demographics or statistics of Infosys' United States work force, including, but not limited to, EEO reports, studies, or other documents reflecting the statistics of the workforce.” Infosys complied with this request to some extent, but it failed to turn over its “PeopleFluent data.' This is demographic data Infosys contracts with PeopleFluent to collect and analyze as part of Infosys’s obligations as a federal contractor to conduct 'affirmative action' analyses. 41 C.F.R. § 60-2.1 (2020) . Plaintiffs learned in June 2016 that the PeopleFluent data existed and eventually received the data directly from PeopleFluent in December 2016 and January 2017 (after discovery had closed and after Infosys had moved for summary judgment).

Because Infosys had failed to produce this data and Plaintiffs received it late, the parties met before former Magistrate Judge David Jones several times in early 2017. Judge Jones asked Plaintiffs if they want ed to supplement their prior filings with the PeopleFluent data, but Plaintiffs never indicated such an intent.

Returning to 2022, after the district court excluded Neumark's expert opinions, in November, Plaintiffs moved under Federal Rule of Civil Procedure 16(b) to retain a new expert Page 8and conduct new discovery. They also indicated they might seek to supplement their summary judgment briefing with the PeopleFluent data if the district court denied their motion. In December 2022, the district court denied Plaintiffs ’ motion and denied leave to supplement the summary judgment briefing with the PeopleFluent data. The court found that

Judge Jones several times asked the plaintiffs whether or not they wanted to have to go back and to supplement their filings with regard to the summary judgment.… And the plaintiffs rejected all of those [offers] . They went ahead.… [T]he plaintiffs were given the very opportunity that I thought they might ask me for after my ruling. Judge Jones gave them that opportunity and gave them an opportunity more than once and told them to propose a plan, and the plaintiffs didn’t. So I … don’t think it is appropriate at this point for the plaintiffs to go back now and somehow modify their response to the defendants' Motions for Summary Judgment as to the individual plaintiff [s].…

Koehler v. Infosys Techs. Ltd. , No. 13-cv-00885 (E.D. Wisc. Mar. 15, 2023), Dkt. No. 214 at 32:2–33:6.

Finally, in June 2025-nearly nine years after Infosys moved—the district court granted summary judgment to Infosys on all claims.² First, it denied the pattern and practice Page 9claim s Plaintiffs brought under International Brotherhood of Teamsters v. United States , 431 U.S. 324 (1977), because it found a Teamsters claim requires class certification and it had previously denied Plaintiffs' motion for class certification.

It next turned to the four Plaintiffs' individual disparate treatment claims, to which it applied the burden-shifting framework of McDonnell Douglas Corp. v. Green , 411 U.S. 792 (1973). In relevant part, it found Handloser’s, Koehler’s, and Parker’s claims all failed at the pretext step, as none could rebut the nondiscriminatory reasons Infosys offered for the employment actions it took. And it found Bolten's claimbrought as a constructive discharge claim-failed because Bolten could not demonstrate a level of harassment sufficient to meet the constructive discharge’s high bar.

Finally, it denied Koehler's and Parker's individual disparate impact claims, finding that those claims relied on Neumark’s statistical analysis and, absent that analysis, Koehler and Parker did not have evidence sufficient to demonstrate a disparate impact.

* * *

Page 10Plaintiffs appeal the decisions of the district court (1) excluding Neumark's opinions, (2) denying class certification, (3) denying leave to supplement their summary judgment briefing with the PeopleFluent data, and (4) granting summary judgment to Infosys on all claims.

The district court applied the proper legal framework when ruling on Infosys’s motion to exclude Neumark’s testimony, so we review its decision for abuse of discretion. Artis v. Santos , 95 F.4th 518, 525 (7th Cir. 2024).

Federal Rule of Evidence 702 provides:

A witness who is qualified as an expert by knowledge, skill, experience, training, or education may testify in the form of an opinion or otherwise if the proponent demonstrates to the court that it is more likely than not that: (a) the expert's scientific, technical, or other specialized knowledge will help the trier of fact to understand the evidence or to determine a fact in issue; (b) the testimony is based on sufficient facts or data; (c) the testimony is the product of reliable principles and methods; and (d) the expert’ s opinion reflects a reliable application of the principles and methods to the facts of the case.

In Daubert v. Merrell Dow Pharmaceuticals, Inc. the Supreme Court set out a district court’s “gatekeeping” obligations, 509 U.S. 579, 597 (1993), to ensure “the reliability and relevancy of expert testimony,” Kumho Tire Co. v. Carmichael , 526 U.S. 137, 152 (1999). See also Artis , 95 F.4th at 525. To satisfy its obligation, a district court evaluates: “(1) the proffered expert 's qual- Page 11ifications ; (2) the reliability of the expert's methodology; and (3) the relevance of the expert's testimony.' Gopalratnam v. Hewlett -Packard Co. , 877 F.3d 771, 779 (7th Cir. 2017). “Because ‘much depends upon the particular circumstances of the particular case at issue,’ the Rule 702 analysis is casespecific. ” Anderson v. Raymond Corp. , 61 F.4th 505, 508 (7th Cir. 2023) (quoting Kumho Tire Co. , 526 U.S. at 150).

Before engaging with Rule 702 and Daubert , Plaintiffs argue the district court erred by “sua sponte” excluding Neumark’s analysis as to Infosys employees when Infosys only moved to exclude his analysis as to applicants. It is clear, however, that “[a] district court is entitled to rule on expert admissibility sua sponte. As long as the district court has given the parties an opportunity to be heard and applies the correct legal framework under Rule 702, we are satisfied. ” Perez v. K & B Transp., Inc. , 967 F.3d 651, 656 (7th Cir. 2020). We ground this rule in the general, broad discretion a district court enjoys in how it conducts a Daubert analysis: 'W e have not required that the Daubert inquiry take any specific form and have, in fact, upheld a judge’s sua sponte consideration of the admissibility of expert testimony.' Kirstein v. Parks Corp. , 159 F.3d 1065, 1067 (7th Cir. 1998); see also Lewis v. CITGO Petroleum Corp. , 561 F.3d 698, 704 (7th Cir. 2009).

Here, Infosys put Plaintiffs on notice that it was challenging Neumark’s qualifications and his name-matching methodology when it moved to exclude, and Plaintiffs had every opportunity to advance the evidence they now claim—nine years later-they would have advanced before the district court. True, Infosys's motion to exclude focused on Neumark's name-matching methodology as applied to Infosys applicants, but the motion addressed the extensive issues Page 12with the methodology generally and Plaintiffs had an opportunity to fully litigate these issues. And, when the court granted Infosys's motion, it found the flaws on which it based its ruling-fl aws Infosys expressly had raised-applied to both applicants and employees, explaining “[i]t is not clear why Neumark's methodology would be fl awed only with regard to the applicant studies.”

Plaintiffs insisted in their briefs, and particularly at oral argument, that had they known the district court would exclude Neumark’s analysis as to employees, they would have pointed to Neumark's minimal use of the name-matching methodology as a part of his employee analysis. But Plaintiffs could have advanced that argument in opposing Infosys's motion. They knew Infosys challenged Neumark’s use of the methodology, and one argument in response could have been its minimal impact on his overall analysis. Moreover, even if Plaintiffs had made this argument, the district court still would have been well within its discretion to reject it and exclude Neumark’s analysis as to Infosys employees. Implicit in their argument is that even as to Infosys employees the flaws the district court identified with Neumark's analysis had some impact on his findings. Maybe that impact was lessened but we would still defer to the district court’s exercise of its discretion— after receiving the parties' briefing and holding a hearing-in determining that Neumark's analysis as to Infosys employees did not satisfy the requirements of Rule 702.

Turning to Neumark's qualifications , “[w]hether a witness is qualified as an expert can only be determined by comparing the area in which the witness has superior knowledge, skill, experience, or education with the subject matter of the witness’s testimony.” Gayton v. McCoy , 593 F.3d 610, 616 (7th Cir.

Page 132010) (quoting Carroll v. Otis Elevator Co. , 896 F.2d 210, 212 (7th Cir. 1990)). “[W] e ask not whether an expert 'is qualified in general' but whether he is qualified 'to answer a specific question ….'' United States v. Truitt , 938 F.3d 885, 889 (7th Cir. 2019) (quoting Gayton , 593 F.3d at 617); see also Ollison v. Gossett , 136 F.4th 729, 742 n.41 (7th Cir. 2025). There is 'no requirement that an expert be a specialist in a given field.' Hall v. Flannery , 840 F.3d 922, 929 (7th Cir. 2016) (quoting Gayton , 593 F.3d at 618). But an expert must “possess[] … specialized education, knowledge, experience, or skill' in the area for which he is proffered. Id. at 929–30.

Neumark admitted at his deposition that he had no experience in name matching and no expertise in identifying South Asian names. This alone is fatal to Neumark's qualifications . Plaintiffs claim we can overlook this because Neumark only searched for “western” names, in which, they contend, he does have experience identifying (seemingly simply due to his upbringing in a western country).

But, setting aside whether his upbringing alone qualifies him as an expert in identifying 'western' names, Plaintiffs ignore that the two categories are not mutually exclusive. That is, even if Neumark was qualified to identify 'western ' names, he remained unqualified to determine whether a 'western' name was or was not also a 'South Asian' name. 3Page 14 This is ultimately the issue with the analysis Neumark purported to use: He claimed he could identify who among Infosys’s employee and applicant population was South Asian based on last name, yet he admitted he lacked the experience or expertise necessary to identify if a last name was South Asian. Given Neumark’s deposition testimony, there was no error, and certainly no abuse of discretion, in the district court's conclusion that Neumark lacked the qualifications necessary to provide his expert opinions.

Rule 702 also requires a court to evaluate “the reliability of the expert’s methodology.” Gopalratnam , 877 F.3d at 779. We have set out “non-exhaustive” factors a court may consider in reviewing reliability, including '(1) whether the particular Page 15scientific theory can be (and has been) tested; (2) whether the theory has been subjected to peer review and publication; (3) the known or potential rate of error; (4) the existence and maintenance of standards controlling the technique ’s operation; and (5) whether the technique has achieved general acceptance in the relevant scientific or expert community. ” Kirk v. Clark Equip. Co. , 991 F.3d 865, 873 (7th Cir. 2021) (quoting Deputy v. Lehman Bros., Inc. , 345 F.3d 494, 505 (7th Cir. 2003) (citation modified) ); see also Gopalratnam , 877 F.3d at 779 (listing similar factors). No factor is dispositive and Rule 702’s test is 'flexible.' Smith v. Ford Motor Co. , 215 F.3d 713, 719 (7th Cir. 2020)). We care not about the “ultimate correctness” of the expert’s opinions; rather, reliability focuses on “the soundness and care with which the expert arrived at h[is] opinion.' Schultz v. Akzo Nobel Paints, LLC , 721 F.3d 426, 431 (7th Cir. 2013).

On appeal, Plaintiffs offer evidence that, in their view, bolsters name-matching’s reliability. Maybe so. But the problem for Plaintiffs is more fundamental: They presented none of this evidence to the district court. Plaintiffs provided no evidence of the name-matching methodology's testing, error rate, peer review, or general acceptance in the scientific community. And Neumark admitted he had no experience with the technique either. Our review is for abuse of discretion, and the district court did not abuse its discretion in concluding Neumark's methodology was unreliable when Plaintiffs did not offer evidence supporting reliabilit y. Gopalratnam , 877 F.3d at 782 (' A decision is an abuse of discretion only if no reasonable person would agree with the decision made by the trial court.” (quoting Smith v. Hunt , 707 F.3d 803, 807–08 (7th Cir. 2013))). Plaintiffs may regret the choices they made before the district court, but we hold them to those choices.

There is little to say about the district court's denial of Plaintiffs' motion for class certification. Because we affirm the district court's decision to exclude Neumark's expert opinions, we affirm the denial of class certification. At oral argument Plaintiffs conceded that absent Neumark’s analysis they cannot meet Federal Rule of Civil Procedure 23's requirements. We agree.

After the district court excluded Neumark’s expert opinions, Plaintiffs sought leave to supplement their summary judgment briefing with the PeopleFluent data. But the court denied leave, a decision we review for abuse of discretion. See Christensen v. Weiss , 145 F.4th 743, 755 (7th Cir. 2025); Frazier v. Dovenmuehle Mortg., Inc. , 72 F.4th 769, 778 (7th Cir. 2023).

The district court enjoys inherent discretion to manage the case before it and to keep the case moving. Ewing v. 1645 W. Farragut LLC , 90 F.4th 876, 889 (7th Cir. 2024) (“District judges have wide discretion to manage their proceedings ....” (quoting Rainey v. Taylor , 941 F.3d 243, 247 (7th Cir. 2019))). For that reason, a district court may properly deny 'an effort to supplement the record after summary judgment … where the information was available at the time the motion was heard and no reasonable explanation is offered for why it was not submitted.” Perez v. Staples Cont. & Com. LLC , 31 F.4th 560, 569 (7th Cir. 2022) (citing Moro v. Shell Oil Co. , 91 F.3d 872, 876 (7th Cir. 1996)).

The district court here did not abuse its discretion. Plaintiffs knew of the PeopleFluent data as of mid-2016 yet only made the request to supplement their briefing in late 2022 , af- Page 17ter the district court granted Infosys’s motion to exclude Neumark’s expert opinions. Moreover, at the time the parties discussed the PeopleFluent data in early 2017, Judge Jones repeatedly asked Plaintiffs what recourse they might want given Infosys’s failure to timely produce the data and offered Plaintiffs potential avenues to remedy the impact of Infosys's shortcomings (including supplementing their summary judgment briefing) . This occurred fi ve years before Plaintiffs eventually sought to supplement their briefing . Plaintiffs rejected or ignored those opportunities and proceeded anyway. They are, again, not entitled to a do-over.

Plaintiffs do not dispute this but instead contend the district court's error was legal: It decided their request to supplement without applying the correct legal standard—namely, Federal Rule of Civil Procedure 6(b) and the Supreme Court’s 'excusable neglect' test from Pioneer Investment Services v. Brunswick Associates Ltd. , 507 U.S. 380, 395 (1993). Rule 6(b) provides that “[w]hen an act may or must be done within a specified time, the court may, for good cause, extend the time … on [a] motion made after the time has expired if the party failed to act because of excusable neglect.”

But it is unclear why Plaintiffs contend we (and the district court) should look to Rule 6(b) . Setting aside the question of whether a request to supplement summary judgment briefing is a request made under Rule 6(b), Plaintiffs never moved under Rule 6(b) or otherwise put the district court on notice that they believed Rule 6(b) and Pioneer ’s “excusable neglect” test applied. Instead, they moved under Rule 16 to modify the district court’s scheduling order so they could submit a new expert and reopen discovery. In the alternative, the district court asked if Plaintiffs might want to supplement their summary Page 18judgment briefing with the PeopleFluent data, to which Plaintiffs responded they would. Neither Plaintiffs nor the district court ever raised or addressed Rule 6(b). So, Plaintiffs have waived their Rule 6(b) argument.

Seeing no error and certainly no abuse of discretion in the district court's exercise of its inherent discretion, we affirm its decision to deny Plaintiffs leave to supplement their summary judgment briefing.

Finally, we turn to the district court's decision granting summary judgment to Infosys on three sets of claims brought by Plaintiffs: (1) the Teamsters pattern and practice claim s; (2) the individual disparate treatment claims; and (3) the disparate impact claims brought by Koehler and Parker. We review de novo the district court’s decision granting Infosys’s motion for summary judgment. Vassileva v. City of Chicago , 118 F.4th 869, 873 (7th Cir. 2024). We 'constru[e] all facts and draw[] all reasonable inferences in the light most favorable to [Plaintiffs] ,' as the non-moving party. Johnson v. Dominguez , 5 F.4th 818, 824 (7th Cir. 2021). “Summary judgment is appropriate if ‘the movant shows that there is no genuine dispute as to any material fact and the movant is entitled to judgment as a matter of law. ’” Id. (quoting Fed. R. Civ. P. 56(a)).

The primary theory of discrimination Plaintiffs pursue is that Infosys “engaged in a pattern and practice of discriminating against” non-South Asian employees and applicants. That is, such discrimination is Infosys’s “standard operating procedure.” Matthews v. Waukesha County , 759 F.3d 821, 829-30 (7th Cir. 2014). Plaintiffs invoke the Supreme Court's decision Page 19in International Brotherhood of Teamsters , which explained that the burden of proof for a “pattern -or-practice” claim brought as a class action requires showing “that unlawful discrimination has been a regular procedure or policy followed by an employer or group of employers.” 431 U.S. at 360.

The district court correctly recognized that once it had denied Plaintiffs' motion for class certification, Plaintiffs could no longer maintain their Teamsters claims. Teamsters makes clear that a pattern and practice claim is conventionally brought as a class action. See id. at 359-60; see also Franks v. Bowman Transp. Co. , 424 U.S. 747, 751 (1976). Since Teamsters , both the Supreme Court and our court have confirmed this view. See Cooper v. Fed. Rsrv. Bank of Richmond , 467 U.S. 867, 876 (1984); Gilty v. Vill. of Oak Park , 919 F.2d 1247, 1252 (7th Cir. 1990) (noting plaintiff 'filed an individual claim [,] … not a class action. Consequently, … evidence of a pattern and practice ‘ can only be collateral to evidence of specific discrimination against the actual plaintiff ’” (quoting Williams v. Boorstin , 663 F.2d 109, 115 n.38 (D.C. Cir. 1980)); see also id. (citing approvingly position that “[p] attern -and-practice suits, by their very nature, involve claims of classwide discrimination” (quoting B. Schlei & P. Grossman, Employment Discrimination Law 1322 & n. 95 (2d ed. 1983))); Matthews 759 F.3d at 829.

Because we affirm the district court's exclusion of Neumark's report and denial of class certification, Plaintiffs cannot maintain their pattern and practice claims brought under Teamsters .

In addition to the Teamsters claims , each plaintiff also brought individual disparate treatment claims under Title VIIPage 20 or 42 U.S.C § 1981. Because “[t]he legal analysis for discrimination claims under Title VII and § 1981 is largely identical,” Lewis v. Ind. Wesleyan Univ. , 36 F.4th 755, 759 (7th Cir. 2022), we analyze Plaintiffs ' claims together. Plaintiffs must show (1) membership in a protected class, (2) an adverse employment action, and (3) causation. Id. As is often the case, only causation is in dispute, which ' require [s] that [race or national origin] be a ‘motivating factor in the defendant’s challenged employment decision.’” Id. ( quoting Comcast Corp. v. Nat'l Ass'n of Af. Am.-Owned Media , 589 U.S. 327, 332 (2020)). The fundamental question “is simply whether the evidence would permit a reasonable factfinder to conclude that the plaintiff 's race, ethnicity, sex, religion, or other proscribed factor caused the discharge or other adverse employment action. Evidence must be considered as a whole, rather than asking whether any particular piece of evidence proves the case by itself ….” Ortiz v. Werner Enters., Inc. , 834 F.3d 760, 765 (7th Cir. 2016).

One method of satisfying this standard is the familiar burden-shifting framework of McDonnell Douglas Corp. , 411 U.S. 792. “To establish a prima facie case of employment discrimination, the employee must show that ‘(1) she is a member of a protected class, (2) she was meeting the defendant’s legitimate expectations, (3) she suffered an adverse employment action, and (4) similarly situated employees who were not members of her protected class were treated more favorably.'' Lewis , 36 F.4th at 759 (quoting Tyburski v. City of Chicago , 964 F.3d 590, 598 (7th Cir. 2020)). If a plaintiff does so, the burden shifts to the defendant to “to articulate a legitimate, nondiscriminatory reason for the adverse employment action.' Id. at 760. Finally, the burden again shifts to the plaintiff to “show why the employer’s explanation is pretextual.” Id. We Page 21have often explained that “[b]ecause the prima facie and pretext inquiry … overlap, if a defendant offers a nondiscriminatory reason for its actions, we can proceed directly to the pretext inquiry. ” Barnes v. Bd. of Trs. of Univ. of Ill. , 946 F.3d 384, 389 (7th Cir. 2020). We take the Barnes approach here.

“[P]retext … is not ‘just faulty reasoning or mistaken judgment on the part of the employer; it is [a] lie, specifically a phony reason for some action.’” Id. at 38990 (quoting Argyropoulos v. City of Alton , 539 F.3d 724, 736 (7th Cir. 2008) (citation modified) ). Plaintiffs ultimately must show, by a preponderance of the evidence, that Infosys’s stated reasons for the employment actions it took 'were false , not that they were unfair or baseless.” Anderson v. Mott St. , 104 F.4th 646, 653 (7th Cir. 2024). Our focus is on whether Infosys “honestly believed the stated reason” for the employment action. Id. at 654.

With these principles in mind, Handloser, Parker, and Koehler all fail to raise an issue of fact as to whether Infosys’s reasons for the employment actions it took against each of them were pretext for discriminatory animus.

Infosys fired Handloser because he received the poorest possible score during the September 2012 review cycle and the company-wide reduction in force targeted all employees who had received that score. He points to his prior, strong performance—but Infosys does not dispute that he was previously performing well. Rather, it explains that the reduction in force focused only on the one review cycle at issue and that Handloser performed poorly during that cycle.

Handloser also points to an email from a supervisor, in which the supervisor, according to Handloser, admitted Handloser would have a strong case against Infosys should it Page 22terminate him. But his reading of the email is selective. Indeed, it confirms that Handloser evaluated himself as underperforming. Though the email speculates that Handloser might sue Infosys should it terminate him, it does not suggest his lawsuit would have any merit or otherwise suggest Infosys’s reason for terminating him was pretext for racial discrimination. In sum, Handloser has not demonstrated a genuine dispute of fact as to whether Infosys’s stated reason for his termination was pretext for discrimination against him on account of his non-South Asian status.

Parker contends Infosys discriminated against her twice: (1) in failing to hire her when she applied for a full-time position and (2) in discharging her when Infosys ended her position through SoftHQ. But, starting with the first, Infosys presented evidence that when Parker applied for the full-time role, it interviewed her and determined she lacked the qualifications necessary for the role. Parker does not advance evidence which demonstrates this was pretext. She contends she was already performing the same role before Infosys interviewed her, so she could not have lacked the necessary qualifications. But Infosys points to interview logs where Parker’s interviewers concluded she did not possess the skills necessary for the role. All Infosys must show is an honestly held belief, Cunningham v. Austin , 125 F.4th 783, 789-90 (7th Cir. 2025), and Parker’s evidence does not undermine that Infosys honestly believed she was unqualified.

Next, Infosys presented evidence showing it believed Parker's position through SoftHQ was temporary, it informed Parker of this during her time in the role, and it discharged her when it decided not to renew her role at the end of the contract period. According to Parker, Infosys never told her Page 23that her position was temporary and Infosys signed the “task order” noting Parker’s position was temporary after it had already informed Parker her position was ending. Parker’s evidence, at most, shows a disjointed and ill-advised process. It does not raise an issue of fact as to pretext for racial animus. Rather, Parker engages in speculation to connect the disjointed discharge process to racial animus, which we do not permit. See Robertson v. Dep’t of Health Servs. , 949 F.3d 371, 378 (7th Cir. 2020) (“In determining whether the employer’s reason can be characterized as pretextual, we do not evaluate whether the employer’ s proffered justification was accurate or even whether it was unfair. Our sole focus is on whether the employer’s stated reason can be characterized as a falsehood rather than an honestly held belief.”).

Like Parker’s, Koehler’s failure to hire claim fails because Koehler does not rebut Infosys's nondiscriminatory reason for not hiring her: It interviewed her and decided she lacked the necessary “Microsoft Active Directory” skills needed for the role. While Koehler may have subjectively believed she was qualified , “subjective belief[,] … does not, without more, demonstrate pretext. Johnson v. Nordstrom, Inc. , 260 F.3d 727, 733 (7th Cir. 2001). Nor has she shown pretext in her contention that the job posting did not list ' Active Directory' as a required skill. The job listing noted general knowledge of “Window’s architecture” as a requirement, of which 'Active Directory' undisputedly is a component. Finally, Koehler claims that last minute scheduling changes and a comment in an Infosys interview log that she was “not reachable” minutes before her interview demonstrate pretext as they show that Infosys had decided not to hire her before it even interviewed her. But this evidence is the very “conjecture and speculation” we will not engage in, particularly absent other evidence of Page 24discrimination. Ennin v. CNH Indus. Am., LLC , 878 F.3d 590, 597 (7th Cir. 2017). So long as Infosys “‘honestly believed” it made the correct decision—“even if its decision was inaccurate, unfair, foolish, trivial, or baseless'-Koehler's claims cannot succeed. Cunningham , 125 F.4th at 790 (citation modified) ; see also Brooks v. Avancez , 39 F.4th 424, 435 (7th Cir. 2022) (“Our sole focus is on whether the employer’s stated reason can be characterized as a falsehood rather than an honestly held belief.” ( quoting Robertson , 949 F.3d at 378)).

Bolten's claim is slightly different. Bolten resigned, so hers is one of constructive discharge, which requires proof that 'the employer' s discriminatory conduct forced the plaintiff 'to resign because her working conditions, from the standpoint of a reasonable employee, had become unbearable.'' Patterson v. Ind. Newspapers, Inc. , 589 F.3d 357, 366 (7th Cir. 2009) (quoting Fischer v. Avanade, Inc., 519 F.3d 393, 409 (7th Cir. 2008) ). A sufficiently 'unbearable' work environment is one that has become “intolerable.” Chapin v. Fort-Rohr Motors, Inc. , 621 F.3d 673, 679 (7th Cir. 2010). “Conditions amounting to constructive discharge must be pervasive and extreme.' Beverly v. Abbo tt Lab’ys , 107 F.4th 737, 745 (7th Cir. 2024). Isolated incidents, even if severe, do not amount to constructive discharge. See id. (collecting cases); Wince v. CBRE, Inc. , 66 F.4th 1033, 1043 (7th Cir. 2023).

Bolten's constructive discharge claim fails because she cannot demonstrate a sufficiently intolerable work environment. Bolten’s evidence included (1) the comments about her being “American,” (2) the comment about her being Russian and connecting that to the Boston Marathon bomber, (3) that her colleagues spoke Hindi at work when discussing work projects, and (4) that when she moved desks, she was told to Page 25sit in a different place. Indeed, some of this is crude and mean -spirited. But none involves physical threats or humiliation or otherwise amounts to the types of evidence we require for constructive discharge. See, e.g. , Fields v. Bd. of Educ. of City of Chicago , 928 F.3d 622, 625–26 (7th Cir. 2019) (no constructive discharge where “no evidence [plaintiff] was subjected to a threat of violence or other conditions that are more severe than those required to establish a hostile work environment '); Wince , 66 F.4th at 1043 (no constructive discharge where record showed “severe” but isolated “racist episode”); Beverly , 107 F.4th at 745–46 (collecting cases).

Last, we turn to Koehler’s and Parker’s individual disparate impact claims. “Under a disparate impact theory, an employer is held liable when a facially neutral employment practice disproportionately impacts members of a legally protected group.” Downing v. Abbott Lab’ys , 48 F.4th 794, 815 (7th Cir. 2022) ( citation modified ).

The disparate impact claims fail without Neumark’s analysis or the PeopleFluent data . Typically, we require some form of statistical evidence to make out a prima facie case of disparate impact. See Puffer v. Allstate Ins. Co. , 675 F.3d 709, 717 (7th Cir. 2012) (“ The plaintiff must also establish causation by offering statistical evidence … sufficient to show that the practice in question has caused the exclusion of applicants for jobs or promotions because of their membership in a protected group.' ( citation modified )); Gilty , 919 F.2d at 1254 (“Disparate impact cases, because of their emphasis on consequences, necessarily require a focus on statistical evidence. ”). And, we have rejected disparate impact claims where the sample size was too small, or the quality of the statistical evi- Page 26dence is insufficient . See, e.g. , Downing , 48 F.4th at 815–16 (rejecting sample of 10 as too small and noting no disparate impact claim “w here the affected group is too small for any valid statistical comparisons, which immunizes most single decisions from disparate impact challenges” ( citation modified )). Without statistical evidence, Koehler and Parker are left with their experiences. But that alone cannot sustain their claims.

* * *

The judgment of the district court is AFFIRMED.