*915 This case returns to us after a bench trial. Intervenor Food Marketing Institute (FMI) argues the district court 1 erred in finding that Exemption 4 to the Freedom of Information Act (FOIA) is inapplicable to data held by the U.S. Department of Agriculture (USDA).

Most of the relevant facts are set out in our previous opinion. See Argus Leader Media v. U.S. Dep't of Agric. , 740 F.3d 1172 , 1173-75 (8th Cir. 2014). The data in question come from the Supplemental Nutrition Assistance Program (SNAP). The USDA issues SNAP participants a card (like a debit card) to use to buy food from participating retailers. When a participant buys food using their SNAP card, the USDA receives a record of that transaction, which is called a SNAP redemption. Argus Leader Media, a South Dakota newspaper, asked the USDA for annual SNAP redemption totals for stores that participate in the SNAP program (the "contested data"). The USDA refused, citing several FOIA exemptions. In our previous opinion, we held that Exemption 3 did not apply to the contested data, and remanded the case to the district court. Id. at 1176-77 .

On remand the only issue was whether FOIA Exemption 4, which covers "trade secrets and commercial or financial information obtained from a person and privileged or confidential," 5 U.S.C. § 552 (b)(4), applies to the contested data. Argus Leader and the USDA agreed that the contested data were commercial or financial information, and that they were not privileged. 2 To show the contested data were "confidential," the USDA had to prove that releasing the data was likely "(1) to impair the Government's ability to obtain necessary information in the future; or (2) to cause substantial harm to the competitive position of the person from whom the information was obtained." Madel v. U.S. Dep't of Justice , 784 F.3d 448 , 452 (8th Cir. 2015) (quoting Contract Freighters, Inc. v. Sec'y of U.S. Dep't of Transp. , 260 F.3d 858 , 861 (8th Cir. 2001) ). The USDA argued only the competitive position prong, so the question before the district court was whether releasing the contested data was likely to cause substantial harm to the competitive position of SNAP retailers.

The case went to bench trial. Both parties called experts to testify about the risks of disclosing the contested data. In its findings of fact and conclusions of law, the district court adopted a definition of competitive harm from the D.C. Circuit: "[C]ompetitive harm may be established if there is evidence of 'actual competition and the likelihood of substantial competitive injury ....' " Argus Leader Media v. U.S. Dep't of Agric. , 224 F.Supp.3d 827 , 833 (D.S.D. 2016) (alteration in original) (quoting *916 Pub. Citizen Health Research Grp. v. FDA , 704 F.2d 1280 , 1291 (D.C. Cir. 1983) ). Applying that standard, the district court found that the grocery retail industry was highly competitive, but that the USDA had not proved a likelihood of substantial competitive injury. Id. at 833-35. The court found the USDA's claims of competitive injury were "speculative at best" because grocery retailers already had access to large quantities of data about their competitors, and existing models explained the majority of grocery customers' behavior. Id. at 834. The court also found speculative the USDA's assertion that stores with high SNAP redemptions would face stigma. Id. 3

After the district court entered judgment for Argus Leader, the USDA decided not to appeal. FMI, a trade group representing grocery retailers, intervened and filed this appeal. FMI contests the district court's findings of fact and application of the law to those facts. 4 "We accept the district court's factual findings unless they are clearly erroneous, and we review the applicability of the FOIA exemption de novo." Peltier v. F.B.I. , 563 F.3d 754 , 762 (8th Cir. 2009).

As to the facts, we see no clear error. FMI argues that the district court erred in finding that release of the contested data would have little effect on the grocery industry, and failed to give enough weight to its assertions that releasing the data would stigmatize some stores and cause stores to stop accepting SNAP. But record evidence showed that the contested data-which are nothing more than annual aggregations of SNAP redemptions-lacked the specificity needed to gain material insight into an individual store's financial health, profit margins, inventory, marketing strategies, sales trends, or market share. FMI's assumption that stores would be stigmatized was speculative and not supported by any other evidence in the record. There was also no meaningful evidence that retailers would end their SNAP participation if the contested data were released.

Applying the law to the facts, we find no basis for reversal. The trial evidence showed that the grocery industry is highly competitive, but is already rich with publically-available data that market participants (and prospective market entrants) use to model their competitors' sales. The evidence shows that releasing the contested data is likely to make these statistical models marginally more accurate. But the evidence does not support a finding that this marginal improvement in accuracy is likely to cause substantial competitive harm. The USDA's evidence showed only that more accurate information would allow grocery retailers to make better business decisions. If that were enough to invoke Exemption 4, commercial data would *917 be exempt from disclosure any time it might prove useful in a competitive marketplace. A likelihood of commercial usefulness-without more-is not the same as a likelihood of substantial competitive harm. We agree with the district court and conclude that the USDA failed to establish that release of the contested data falls within Exemption 4's ambit.

In an effort to avoid this conclusion, FMI cites to our opinion in Madel . In that case, we affirmed Exemption 4's application to Drug Enforcement Administration (DEA) records pertaining to oxycodone transactions by private companies. See Madel , 784 F.3d at 451-53 (noting that one of the requested documents "contain[ed] information traceable to individual manufacturers and distributors, such as market shares in specific geographic areas, estimates of inventories, and sales" for the entire nation-wide market for oxycodone). We cited DEA declarations that the requested records "could be used to determine the companies' market shares, inventory levels, and sales trends in particular areas" which might allow competitors "to target specific markets, forecast potential business of new locations, or to gain market share in existing locations, thereby gaining competitive advantage." Id. at 453 (cleaned up). While these concerns appear to mirror those raised by the USDA in this case, Madel is distinguishable. In Madel , the data in question were sufficiently specific (records of individual companies' sales of a particular drug) that their release was likely to provide a tangible competitive advantage. The contested data in this case, by contrast, are more general, and add little to the information already available to retailers. Because the Madel data are not analogous to the data in this case, the result is different. 5

For these reasons, we affirm the judgment of the district court.