The settlement reached the data supplier
On September 14, 2021, the SEC announced a securities-fraud settlement with App Annie Inc. and its co-founder and former chief executive Bertrand Schmitt. App Annie was ordered to pay $10 million and Schmitt $300,000. Both accepted a cease-and-desist order without admitting or denying the findings, apart from jurisdiction. Schmitt also received a three-year prohibition on serving as an officer or director of a covered public issuer. [1][2]
The SEC described the action as its first charging an alternative-data provider with securities fraud. The relevant product estimated mobile-app downloads, usage and revenue. Trading firms used that information to inform securities transactions. The action concerned representations and conduct in generating and supplying those estimates; it was not a finding that purchasing any alternative dataset is illegal. [1]
Two customer relationships, two sets of promises
The SEC found that App Annie obtained confidential app-performance data from companies on assurances about aggregation and anonymization. It also represented to trading-firm subscribers that its estimates were generated consistently with those permissions and with effective controls against misuse of confidential information. According to the agency, from late 2014 to mid-2018 the company instead used non-aggregated, non-anonymized information to alter estimates. [1]
Those two relationships create the case’s economic structure. Data contributors receive an analytics service or another benefit in exchange for sharing information. Subscribers pay for a useful measurement of businesses they do not control. The supplier can monetize the gap between what contributors know about themselves and what outside investors can observe.
That arrangement is capable of legitimate value creation. A model can combine permitted observations into an estimate that is more timely than a quarterly filing. But the permission to collect data for one purpose is not automatically permission to supply it for another, and the label “estimate” does not settle how much confidential information survives the transformation.
Why accuracy can conceal a provenance problem
A statistical model typically produces an estimate from an observed sample and a set of assumptions. An adjustment made after seeing a company’s confidential actual result can reduce measured error without improving the model’s genuine ability to infer performance from permitted inputs. The output can become more accurate and less faithful to the process subscribers were told they were buying.
Consider a hypothetical model that estimates revenue at $80 when a confidential source shows $100. Replacing the estimate with $90 halves its absolute error from $20 to $10. That improvement says nothing about whether the original method learned a better relationship. If future users cannot lawfully access the same confidential source, the apparent gain may not be reproducible. These figures illustrate the issue and are not App Annie’s reported financial results.
The SEC order described post-model alteration practices, including an error-halving process, and found that customers did not know about the relevant use of confidential information. It also documented the distinction between a stated internal exclusion policy, the narrower policy first written down in 2017 and its implementation. In June 2018 the company stopped the described alterations and began excluding public-company data consistently with its representations. [2]
For an investment user, provenance is part of the product’s economics. A signal that is accurate but unavailable under the promised permissions carries a different risk from a lawful but noisy estimate. Pricing, contractual warranties and expectations of future availability can all depend on that distinction. Model error and legal exposure are not alternative ways of measuring the same uncertainty.
Aggregation and anonymity do different work
Aggregation combines observations. Anonymization seeks to prevent an observation from being attributed to an identifiable contributor. Neither word, standing alone, explains whether a small group can be reverse-engineered or whether a company’s actual result materially shapes its own published estimate. The unit being protected can be a person, a publisher, an app or a corporate issuer.
For example, a dataset containing a thousand observations may still reveal one company’s performance if that company supplies nearly all the economic activity. A model with many inputs can also rely disproportionately on a single direct measurement. Those are general statistical possibilities, not additional findings about App Annie. They show why a count of records or a claim of sophisticated modeling is an incomplete description of information exposure.
Timing changes value as well. A reasonably accurate estimate arriving before earnings can be more useful to a trader than the same estimate arriving afterward. That commercial demand creates a premium for freshness, but freshness can also narrow the distance between an estimate and undisclosed actual performance. The legal issue depends on the actual information and conduct, not on whether the vendor calls the product artificial intelligence, analytics or research.
The enforcement record and a later staff warning
The order found violations of Exchange Act Section 10(b) and Rule 10b-5 through consent, not a contested trial. [1]
The public record does not establish that every subscriber knowingly possessed improperly sourced information, that every resulting trade was profitable or that all trading customers committed insider trading. Those would be separate claims requiring evidence about each party and transaction.
In April 2022, the SEC’s Division of Examinations issued a broader risk alert about investment advisers’ handling of material nonpublic information. Staff identified deficiencies involving alternative-data diligence and policies. The document expressly states that it is staff guidance with no new legal force or obligations. It supplies later regulatory context, rather than a further judgment against App Annie. [3]
The connection is informational. A supplier’s explanation of its process can be important to the subscriber’s own assessment, but a polished explanation is not the same as evidence about actual data flows. A change in a dataset’s contributors or transformations can alter both statistical properties and legal assumptions while leaving the product name unchanged.
What changed after the historical conduct
The three-year officer-and-director bar ran from entry of the September 14, 2021 order. Its stated term therefore reached September 2024; describing that particular bar as permanent would be wrong. The October 4, 2026 check located no later official extension or vacatur. The cease-and-desist remedy is a different provision and should not be treated as automatically ending with the bar. [2]
Sensor Tower announced on March 18, 2024 that it had acquired data.ai, the business formerly known as App Annie, for an undisclosed amount. That is a company-reported corporate development, not an SEC determination about the purchaser’s current practices. The acquisition does not erase the historical order or prove that the old conduct persists. [4]
The case’s lasting importance is the separation of three questions: how useful a number is, how it was produced and what users were told about its production. A financial signal can fail the latter two tests while appearing to excel at the first. That makes a data supplier’s credibility part of the market infrastructure surrounding investment decisions, even when the supplier never executes a securities trade itself.
Sources
- SEC — App Annie settlement announcement, September 14, 2021Filing / reportBack to text: ↑1↑2↑3↑4
- SEC — App Annie and Schmitt order, Release 92975, September 14, 2021Filing / report · PDFBack to text: ↑1↑2↑3
- SEC Division of Examinations — Investment Adviser MNPI Compliance Issues, April 26, 2022Filing / report · PDFBack to text: ↑
- Sensor Tower — data.ai acquisition announcement, March 18, 2024SourceBack to text: ↑