*979 This appeal arises out of a putative class action filed on behalf of purchasers of Flotek Industries, Inc. common stock. Plaintiffs allege that defendants, Flotek and three of its officers, exaggerated the usefulness of its products. They allege misrepresentations relating to a proprietary software Flotek developed to help market these products. The district court dismissed the complaint, holding that plaintiffs had failed to plead facts giving rise to a strong inference of fraudulent scienter. We AFFIRM.

I

Plaintiffs filed this putative securities class action on behalf of investors who purchased Flotek common stock between October 23, 2014, and November 9, 2015. Defendants are Flotek Industries, Inc. and three of its officers: Chief Executive Officer (CEO) and President John W. Chisholm and Chief Financial Officers H. Richard Walton and Robert M. Schmitz. Flotek sells oilfield products called "Complex nano-Fluid technologies" (CnF), which are supposed to improve the productivity of oil and gas wells. According to the complaint, CnF is Flotek's "hallmark" product, and "[b]y the beginning of the Class Period, the Energy Chemistry Technologies segment [of Flotek] represented over 50% of the Company's revenue due to sales of its CnF products." This lawsuit concerns representations made about a software application Flotek developed to help market CnF to exploration and production companies. The software, called "FracMax," analyzes and presents data concerning hydraulically fractured wells to allow comparison of the productivity of oil and gas wells that use CnF with those that do not.

Flotek introduced FracMax to investors at a June 2014 investor conference. Over the next sixteen months, "[D]efendants attended at least 21 conferences, including Flotek's earning conference calls, analyst-hosted conferences and [c]ompany-hosted investor conferences, where they focused on and praised FracMax and its ability to conclusively validate the efficacy and economic benefits of Flotek's CnF products." Defendants promoted FracMax as an integral component of Flotek's sales strategy and as "key to 'materially broaden[ing] the reach of Flotek's marketing efforts' for its CnF products." "[D]efendants reported that the FracMax database included production data from 80,000 wells across key U.S. basins and that based on this data, Flotek's sales force could demonstrate to potential customers that the use of Flotek CnF chemistries added at least an estimated $ 8 billion in aggregate value for operators when compared to those operators that had not adopted Flotek's CnF products." In a 2015 press release, Chisholm, Flotek's President and CEO, stated, "Our FracMax software technology provides conclusive evidence that our [CnF] suite of completion chemistries provides compelling economic benefits to production companies." Flotek also represented in a quarterly earnings press release that sales of CnFs increased substantially because of FracMax.

Although their complaint largely refers to Defendants generically, Plaintiffs' references to specific representations, to the extent they specify, primarily relate to statements made by Chisholm, who at various times emphasized FracMax's empirical validity. During several conference *980 calls with investors, Chisholm explained that FracMax used publicly available data that companies self-reported to state agencies to compare production from wells that used CnF and wells that did not, suggesting this made FracMax's output more reliable. In September 2015, Chisholm gave a presentation at an investor conference in which he explained that FracMax used data reported to the Texas Railroad Commission, a state agency, and he suggested that the data was "back check[ed] and validate[d]." In his PowerPoint presentation, Chisholm presented an "About FracMax" slide stating that the data was "sourced from operator-provided completion data." This slide further stated that the data was "un-adjusted, providing for comparison and analysis of operators' self-reported data sets." Chisholm also presented images of the FracMax interface in order to compare the productivity of four Texas wells, one that used CnF and three that did not, again emphasizing the difference in production levels.

On November 9, 2015, online financial publication Bronte Capital released a blog post (the Bronte Report) contending that the data in Chisholm's presentation was wrong and suggesting that the data had been intentionally altered to make CnF look significantly more effective. Specifically, the Bronte Report alleged that, as compared with the numbers reported by the Texas Railroad Commission, "Flotek had reduced the production data for the [three] non-CnF wells by 40% to 50% ..., while leaving the production data for the CnF well unchanged." After release of the Bronte Report, Flotek shares declined.

The next day, Flotek issued a press release conceding that Bronte Capital's analysis was correct, 1 and ascribing the error to data provided by a third party, Drillinginfo, that "caused FracMax to identify the three non-CnF wells as contained in units with multiple wells (as opposed to single well units), which required FracMax to incorrectly apply an allocation algorithm to the production for the non-CnF wells." According to Flotek, because Texas reports oil and gas production by lease, rather than by well, FracMax used an algorithm to apportion production within multiple-well lease units, leading to unwarranted adjustments of the non-CnF wells discussed at the September conference. Flotek also held a conference call, during which "Chisholm admitted that the Company's internal controls 'did not catch' " the errors, and "that the Company had no internal controls in place to check the accuracy of the third-party data from Drillinginfo." He explained that they did not cross-reference the third-party data against the underlying state-agency data and that the company "had, several months ago, evolved to a different allocation program" that would provide greater accuracy.

According to Plaintiffs, a report issued by Iberia Capital Partners that same day "stated that the four wells presented at the September 11, 2015 investor conference were designated by Drillinginfo as being single well leases, not multiple leases, therefore the Texas Railroad Commission data for those wells did not require adjustment," and Flotek shares fell further as a result. Contrary to its earlier statements following the Bronte Report, Flotek then reported that an internal investigation concluded that the error originated in the software, not the data set, as Chisholm had stated earlier, and "most likely resulted from the accidental inclusion of test code by the third-party software developer.... who was hired to write code for FracMax."

*981 Plaintiffs filed this lawsuit, alleging violations of Section 10(b) of the Securities Exchange Act, 15 U.S.C. § 78j(b) ; Securities and Exchange Commission Rule 10b-5, 17 C.F.R. § 240 .10b-5 ; and control person liability for the individual defendants under Section 20(a) of the Securities Exchange Act, 15 U.S.C. § 78t. Defendants filed a motion to dismiss the complaint for failure to state a claim. The district court dismissed the complaint, concluding that Plaintiffs had failed to sufficiently plead scienter. Plaintiffs timely appealed.

II

This court reviews the sufficiency of a complaint de novo. Ind. Elec. Workers' Pension Tr. Fund IBEW v. Shaw Grp., Inc. , 537 F.3d 527 , 533 (5th Cir. 2008). Plaintiffs' "well-pleaded facts are to be accepted as true and viewed in the light most favorable to [them]." Daugherty v. Convergent Outsourcing, Inc. , 836 F.3d 507 , 510 (5th Cir. 2016). "[C]onclusory allegations, unwarranted deductions, or legal conclusions" are not "well-pleaded facts" for purposes of evaluating a complaint. Southland Sec. Corp. v. INSpire Ins. Sols. Inc. , 365 F.3d 353 , 361 (5th Cir. 2004). Where fraud is alleged, Federal "Rule [of Civil Procedure] 9(b) creates a heightened pleading requirement that 'the circumstances constituting fraud or mistake shall be stated with particularity.' " U.S. ex rel. Rafizadeh v. Cont'l Common, Inc. , 553 F.3d 869 , 872 (5th Cir. 2008) (quoting FED. R. CIV. P. 9(b) ). A class-action complaint alleging a violation of Section 10(b) must allege fraud in accordance with the Rule 9(b) heightened-pleading standard. See Southland , 365 F.3d at 362 .

To state a viable securities-fraud claim under Section 10(b) and Rule 10b-5, Plaintiffs must allege that (1) Defendants made a misrepresentation or omission relating to the purchase or sale of a security, (2) such representation or omission related to a material fact, (3) the representation or omission was made with scienter, (4) Plaintiffs acted in reliance on Defendants' representation or omission, and (5) the representation or omission proximately caused Plaintiffs' losses. Neiman v. Bulmahn , 854 F.3d 741 , 746 (5th Cir. 2017) (quoting Nathenson v. Zonagen Inc. , 267 F.3d 400 , 406-07 (5th Cir. 2001) ). At issue in this appeal is whether Plaintiffs have sufficiently pleaded scienter.

III

A

The Private Securities Litigation Reform Act (PSLRA) specifically requires that a complaint in a securities case support allegations of scienter with "facts giving rise to a strong inference that the defendant acted with the required state of mind." 15 U.S.C. § 78u-4(b)(2)(A). " 'Scienter' is 'a mental state embracing intent to deceive, manipulate, or defraud.' " Goldstein v. MCI WorldCom , 340 F.3d 238 , 245 (5th Cir. 2003) (quoting Ernst & Ernst v. Hochfelder , 425 U.S. 185 , 193 n.12, 96 S.Ct. 1375 , 47 L.Ed.2d 668 (1976) ). For purposes of 10(b) liability, a defendant must have acted with, at minimum, severe recklessness. Warren v. Reserve Fund, Inc. , 728 F.2d 741 , 745 (5th Cir. 1984). "Severe recklessness," is "limited to those highly unreasonable omissions or misrepresentations that involve not merely simple or even inexcusable negligence, but an extreme departure from the standards of ordinary care." Rosenzweig v. Azurix Corp. , 332 F.3d 854 , 866 (5th Cir. 2003) (quoting Nathenson , 267 F.3d at 408 ). The omissions or misrepresentations at issue must also "present a danger of misleading buyers or sellers which is either known to the defendant or is so obvious that the defendant must have been aware of it." Id.

*982 (quoting Nathenson , 267 F.3d at 408 ). Though not in themselves sufficient to support a strong inference of scienter, "[a]ppropriate motive and opportunity allegations may ... 'meaningfully enhance the strength of the inference of scienter.' " Shaw , 537 F.3d at 533 (quoting Southland , 365 F.3d at 368 ).

To evaluate scienter in a securities-fraud case, a court must (1) take the well-pleaded allegations as true; (2) evaluate the facts collectively, including facts contained in "documents incorporated in the complaint by reference and matters subject to judicial notice," "to determine whether a strong inference of scienter has been pled"; and (3) "take into account plausible inferences opposing as well as supporting a strong inference of scienter." Id. To withstand a motion to dismiss, "an inference of scienter must be more than merely plausible or reasonable-it must be cogent and at least as compelling as any opposing inference of nonfraudulent intent." Tellabs, Inc. v. Makor Issues & Rights, Ltd. , 551 U.S. 308 , 314, 127 S.Ct. 2499 , 168 L.Ed.2d 179 (2007).

Scienter must be alleged with respect to "the individual corporate official or officials who make or issue the statement (or order or approve it or its making or issuance, or who furnish information or language for inclusion therein, or the like) rather than generally to the collective knowledge of all the corporation's officers and employees acquired in the course of their employment." Southland, 365 F.3d at 366 ; see also Shaw , 537 F.3d at 533 ("[T]his court has rejected the group pleading approach to scienter and instead looks to the state of mind of the individual corporate official or officials."). Furthermore, such allegations ordinarily must be based on more than an individual's position within the company, absent special circumstances. Neiman , 854 F.3d at 749 .

B

Plaintiffs identify several alleged misrepresentations they argue give rise to a sufficient inference of scienter, both because Defendants were severely reckless in making statements counter to information that should have been obvious to them, and because, after learning of that contrary information, they continued to make those statements. The alleged misrepresentations are: (1) Defendants' repeated use of the term "conclusive" in describing FracMax's effect, which Plaintiffs characterize as tantamount to assuring that FracMax's data was irrefutable; (2) Defendants' reference to FracMax data as "un-adjusted" when, in fact, FracMax used an algorithm to adjust certain data; (3) Defendant Chisholm's presentation at the September 11, 2015, conference of direct comparisons between several wells that used CnF versus several that did not, which Defendants concede relied on incorrect data for the non-CnF wells; and (4) Chisholm's representation at this same conference that this data was "back check[ed] and validate[d]," when Defendants later admitted that "Flotek had no internal controls in place to ensure the integrity of the FracMax database." The district court determined that these representations failed to generate a strong inference of scienter. We consider each representation in turn, and ultimately agree with the district court's conclusion that each fails to raise a strong inference of scienter. 2

*983 1. Description of FracMax as "Conclusive"

Plaintiffs suggest that Defendants' use of the term "conclusive" to describe FracMax was obviously misleading because Flotek employed "no internal controls" over the information used for FracMax provided by third-party Drillinginfo, yet the term "conclusive" suggests the data is infallible. However, in context, Chisholm's use of the term "conclusive" may have had innocent intentions and may not have been inconsistent with a lack of internal controls. See Shaw , 537 F.3d at 538 (statements subject to "many interpretations, including innocent ones," do not "contribute to a strong inference of scienter" (cleaned up)). For example, Chisholm stated in a press release that the "FracMax software technology provides conclusive evidence that our [CnF] suite of completion chemistries provides compelling economic benefits to production companies." Plaintiffs do not allege that CnF products provide no economic benefit whatsoever, but instead allege the benefit was overstated. Therefore, Chisholm's generalized endorsement of FracMax as evidencing the "compelling economic benefits" of CnF products is not unreasonable, given that CnF products undisputedly provided some economic benefit. At the very least, such statements are not "highly unreasonable omissions or misrepresentations ... involv[ing] ... an extreme departure from the standards of ordinary care." Rosenzweig , 332 F.3d at 866 (quoting Nathenson , 267 F.3d at 408 ). Plaintiffs argue that Flotek's lack of internal controls should have made it obvious to Defendants that using the term "conclusive" was misleading. That argument also fails. Plaintiffs do not allege that Defendants should have known the Drillinginfo data was unreliable. While it was perhaps unwise to rely completely on third-party data while referring to the product using it as "conclusive," it was not reckless to do so. See MCI WorldCom , 340 F.3d at 254 (holding complaint presenting "what could best be described as allegations of mismanagement," or even "gross mismanagement," failed to allege severe recklessness of any individual). At bottom, Plaintiffs' allegations concerning Defendants' characterization of FracMax data as conclusive fail to generate an inference of scienter that is "at least as compelling as ... opposing inference[s] of nonfraudulent intent." Tellabs , 551 U.S. at 314 , 127 S.Ct. 2499 .

2. Characterization of FracMax-Provided Information as "Un-Adjusted"

Plaintiffs also contend that a slide show Chisholm presented at least once, which stated that FracMax "data is un-adjusted," was misleading and Chisholm should have known it was incorrect, because FracMax used an allocation algorithm that necessarily made adjustments to Texas production data. In response to the fallout from the Bronte Report, Flotek disclosed that it used an "allocation algorithm" for Texas data, because data provided by the Texas Railroad Commission was not broken down by well, but instead by lease, and a lease might contain multiple wells. Therefore, according to Plaintiffs, any statement positing that the data was not adjusted was patently false, supporting an inference of scienter. However, there is no specific allegation that Chisholm knew at the time he made the statement at issue that FracMax utilized an algorithm. Moreover, as the district court correctly pointed out, the use of an algorithm does not make the claim that the data was "un-adjusted" misleading. See Rosenzweig , 332 F.3d at 866 (danger of misleading buyers or sellers must be "either known to the defendant or is so obvious that the defendant must have been aware of it" (cleaned up)).

*984 3. Presentation of Incorrect Well Data

The starkest example of Flotek's provision of false information is that Chisholm presented demonstrably false information at the September 2015 conference in the form of incorrect well data. However, even these statements fail to support a strong inference of scienter, because Plaintiffs fail to plead that any Defendant knew of these errors at the time, and the misstatements were not sufficiently obvious that Defendants were severely reckless in presenting the information. As the district court pointed out, although Plaintiffs pleaded that the three non-CnF wells used in the September 2015 presentation were downwardly adjusted, they "failed to plead facts establishing this discrepancy is true throughout FracMax's 80,000 well database," and "a trend as to six wells out of 80,000 is not sufficient on its own to establish a strong inference of scienter." Plaintiffs contend that the fact that the discrepancies inured to Flotek's benefit weighs in favor of a finding of scienter, as it suggests that Chisholm made intentional misrepresentations rather than mistakes. But the mere fact that an error favors the defendant is insufficient to support a strong inference of scienter. Cf. Shaw , 537 F.3d at 543 (characterizing allegations that defendants would benefit from representations as supporting an inference of motive rather than going directly to severe recklessness, and noting that motive alone is insufficient for scienter).

It does appear that it would have been very easy to check if this data was correct-Flotek verified the alleged errors and responded to the Bronte Report within a day. However, while this suggests negligence, there is no indication that Defendants had reason to know of any deficiencies in quality control problems before the data was made public. Cf. Abrams v. Baker Hughes Inc. , 292 F.3d 424 , 432 (5th Cir. 2002) ("Plaintiffs point to no allegations *985 that the defendants knew about the internal control problems, only that they should have known or that their lack of knowledge based on their corporate positions demonstrates recklessness."). Plaintiffs rely on the fact that Flotek revised its allocation algorithm, which they claim implies that there was an issue with the previous algorithm. However, we have previously declined to draw a similar inference, concluding in Abrams that "[t]he fact that Baker Hughes was overhauling its accounting system ... does not command an inference that company officials should have anticipated finding a problem or assumed that financial data reported under [the] old system was inaccurate." Abrams , 292 F.3d at 433 . Thus, Chisholm's use of incorrect data at the September 2015 conference also fails to give rise to a strong inference of scienter.

4. Statement that Data Was Back-Checked and Validated

Finally, Plaintiffs claim that Chisholm represented that the data presented at the September 2015 conference was "back check[ed] and validate[d]." Plaintiffs allege that this was false, as "Flotek had no internal controls in place to ensure the integrity of the FracMax database." Indeed, as Chisholm admitted, there was a failure in quality control, because Flotek did not cross-reference "[the Drillinginfo data] with the [Texas] Railroad Commission data." Despite Chisholm's subsequent recognition that Flotek "should have had the quality control in place that could have validated [the Drillinginfo data] and it wasn't," Plaintiffs nonetheless fail to allege that Chisholm knew of this lack of quality control at the time he made the statement, or that it would have been so obvious that he should have known. Cf. Abrams , 292 F.3d at 432 ("Plaintiffs point to no allegations that the defendants knew about the internal control problems, only that they should have known or that their lack of knowledge based on their corporate positions demonstrates recklessness."). Further, the statement is ambiguous because Chisholm does not say whether Flotek itself back checks and validates the data, or instead relies on a third party to do so, which Chisholm may well have believed was a part of the process. See Shaw , 537 F.3d at 538 (analyzing "more likely, nonculpable inference, absent any other details" about the representation). Accordingly, this allegation also fails to give rise to a strong inference of scienter.

C

In the absence of specific allegations that any Defendant knew of the falsity of any of the statements discussed above at the time they were made or were severely reckless with respect to the statements' truth, Plaintiffs invoke Chisholm's position as an inventor of the FracMax technology as sufficient to "permit a plaintiff to plead scienter." 3 Neiman , 854 F.3d at 749 . Our case law provides that an inference of scienter may be drawn solely from "a defendant's position in the company" only if certain "special circumstances" are alleged. Id. We have stated these "special circumstances" as follows: (1) "the smaller the company the more likely it is that corporate executives would be familiar with the intricacies of day to day operations"; (2) "the transaction at issue may have been critical to the company's continued vitality"; (3) "the misrepresented or omitted information at issue would have been readily apparent to the speaker"; and (4) "the defendant's statements were internally inconsistent with one another." Id. at 749-50 (quoting Local 731 I.B. of T. Excavators & Pavers Pension Tr. Fund v. Diodes, Inc. , 810 F.3d 951 , 959 (5th Cir. 2016) ).

Our "special circumstances" doctrine does not apply here, and therefore our analysis regarding the alleged misrepresentations, supra Part III.B, holds. Plaintiffs concede that they "do[ ] not contend Flotek is a small company," and instead ask us to find special circumstances based on Defendants' knowledge of the FracMax product. However, we have never found special circumstances permitting an inference of scienter based solely on a defendant's position when the company was large. Compare Dorsey v. Portfolio Equities, Inc., 540 F.3d 333 , 342-43 (5th Cir. 2008) (finding special circumstances where defendant company had no employees itself and was managed by company with only eight employees) and Nathenson , 267 F.3d at 425 (inference warranted where company had roughly thirty-five employees), with Neiman , 854 F.3d at 750 ("[W]ith over 60 employees, [defendant] was approximately twice as large as the companies in the cases where this court has found a 'special circumstance.' ").

The other special circumstances factors also cut against applying that doctrine *986 here. Although FracMax was important to Flotek's sales of CnF, it cannot be said to be critical to its "continued vitality," as required. 4 See Diodes , 810 F.3d at 959 (finding omissions regarding a labor shortage not sufficiently important where the plaintiff did not allege that the labor shortage "jeopardized the company's existence"). Nor would it have been "readily apparent to the speaker"-primarily, Chisholm-that the statements were incorrect, as discussed above in analyzing each statement individually. See supra Part III.B. Finally, despite Plaintiffs' contrary contention, no internal inconsistency exists between the statements that well data was un-altered and that FracMax data was conclusive, on the one hand, and the later revelations about the accuracy of that data and the lack of internal controls, on the other. The earlier statements do not directly contradict that later information. As discussed, the context of the use of these terms reveals that the term "conclusive" was used to tout the economic benefits of CnF products, and the term "un-adjusted" was used in conjunction with references to FracMax as an "analytical model." As such, these statements do not conflict with later revelations that flaws existed in the data and that Flotek relied on a third party to provide accurate information. See supra Part III.B.1-2.

D

Finally, we review the complaint holistically, as the district court did, to determine "whether all of the facts alleged, taken collectively, give rise to a strong inference of scienter," Tellabs , 551 U.S. at 323 , 127 S.Ct. 2499 . The district court correctly pointed out that "Plaintiffs ask this Court to assume scienter based solely on the importance of FracMax to Flotek's business, Defendants' positions within the company, and the fact that the alleged 'mistake' happened in a way that made Flotek's core product, CnF, look more profitable." We agree that such allegations are insufficient to raise a strong inference of scienter and, at most, indicate "simple or even inexcusable negligence," a lesser showing than is required here. See Rosenzweig , 332 F.3d at 866 . This conclusion is further buttressed by Plaintiffs' pervasive use of group pleading-referring generally to "Defendants" rather than specific individuals-a practice this court has expressly rejected. See Shaw , 537 F.3d at 533 ("[T]his court has rejected the group pleading approach to scienter and instead looks to the state of mind of the individual corporate official or officials."). For these reasons, the district court's holistic analysis was also correct.

IV

Plaintiffs contend that the district court erred by dismissing their Section 20(a) claims against the individual defendants for control-person liability. "Control person liability is secondary only and cannot exist in the absence of a primary violation." Southland , 365 F.3d at 383 . Because Plaintiffs have not established a primary violation, their Section 20(a) claims fail.

*987 * * *

For these reasons, the judgment of the district court is AFFIRMED.