Pith. sign in

REVIEW 4 major objections 4 minor 67 references

Blind Targeting: Personalization under Third-Party Privacy Constraints

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that strategic querying—an adaptive Bayesian-optimization method that learns from noisy aggregate queries—can design targeting policies that capture 97-101% of the value of a state-of-the-art method with access to all…

desk verdict Worth reading and worth fixing: the targeting idea is good and the empirical results are striking, but Eq. 6 is the covariance of integrals, not averages, so the core posterior and the headline numbers rest on a math error as written. read the letter →

arxiv 2507.05175 v1 pith:4CAT4FAL submitted 2025-07-07 stat.ME cs.LGecon.EMstat.AP

Anya Shchetkina This is my paper · ORCID
classification stat.MEcs.LGecon.EMstat.AP MSC 62F1562G08
keywords third-partyprivacytargetingBayesianoptimizationGaussianprocessesdifferentialheterogeneoustreatmenteffectsaggregatequeriesupliftmodeling
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper asks whether advertisers can still design effective targeting policies when platforms stop giving access to raw individual-level data and instead offer only a limited number of noisy aggregate queries. It develops a method named strategic querying, which adaptively chooses which regions of the data to ask about based on what is most useful for the targeting decision, and tests it in simulations and on the large-scale Criteo uplift dataset. The central claim is that strategic querying retains 97-101% of the targeting value of Causal Forest, a state-of-the-art method that sees all individual records, across four privacy settings, while using fewer than 50 noisy averages instead of 600,000 raw data entries. A simple uniform querying benchmark drops to 33% of that value in one setting. The paper concludes that privacy protections and data-driven personalization are not inherently incompatible, because targeting mostly requires the sign of treatment effects for fairly broad segments rather than precise individual-level estimates.

What carries the argument

The load-bearing mechanism is the pair of Bayesian-optimization extensions the paper introduces. Integral updating extends Gaussian-process regression to observations that are averages over ranges rather than point values, using the closed-form kernel for integrals of a squared-exponential process, so the posterior predictive mean and variance can be evaluated for any candidate query region. The targeting-aware acquisition function, $\text{TAAF}([\underline{X},\bar{X}]) = \beta V([\underline{X},\bar{X}]) - |m([\underline{X},\bar{X}])|$, operationalizes the observation that an additional query has the largest expected value when the region's estimated treatment effect is near zero and uncertain; these are the regions where the optimal targeting decision is still in doubt. Size constraints on candidate regions are needed in simulations to prevent the algorithm from spending queries on overly broad or overly granular cells. The combination makes the query sequence adaptive and decision-aware, which is what distinguishes strategic querying from a fixed uniform grid.

What would settle it

Re-run the Criteo empirical study without collapsing the twelve covariates to three, letting the query space include all available features, and check whether fewer than 50 noisy aggregate queries still achieve near-parity with a full-data Causal Forest; if the lift ratio drops materially below 97%, the central claim does not generalize beyond the paper's coarse query space.

Watch

Extended reading notes

Core claim

The paper's central discovery is that a targeting policy can be learned from a handful of carefully chosen aggregate queries without meaningfully losing value relative to full-data machine learning. Treating the unobserved individual-level treatment-effect surface as a Gaussian process, the method uses integral updating to form posteriors over averages of arbitrary query regions, and a targeting-aware acquisition function to select the next region: it queries where the posterior mean is close to zero and the posterior variance is high, because those are the regions whose treatment or control assignment could change. In the Criteo application, with twelve covariates collapsed to three and two query budgets (27 and 64) crossed with two differential-privacy noise scales, strategic querying achieves 97-101% of Causal Forest's inverse-propensity-weighted lift and is statistically indistinguishable from it, while uniform querying falls to 33% in the most restrictive setting. The empirical result is presented as evidence that the method works exactly where a non-adaptive benchmark fails: when the granularity of treatment effects is unknown and privacy noise is substantial.

Load-bearing premise

The headline result depends on the treatment-effect surface being smooth enough and low-dimensional enough that a Gaussian-process prior over only three collapsed covariates can capture it from fewer than 50 noisy region averages.

Editorial extensions

If this is right

  • Advertisers can build effective targeting policies in privacy-restricted platforms without access to raw individual-level data.
  • The method's advantage over uniform querying is largest when the query budget is moderate and differential-privacy noise is nontrivial; uniform querying only suffices when the marketer knows the granularity of treatment effects and noise is low.
  • Targeting is an information-light task: recovering the sign of treatment effects for broad segments, rather than precise individual-level estimates, is enough for near-oracle policy value.
  • In simulations, the best variant of strategic querying is not dominated by any other method in any of the 144 settings, suggesting that the targeting-aware acquisition function with region-size penalties and constraints is the reliable configuration.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The 97-101% result is demonstrated after collapsing twelve covariates to three; if the informative heterogeneity lives in the discarded individual variation of the other nine variables, the required query count would likely grow, and testing the method with the full covariate space on Criteo would show how much of the result depends on that collapse.
  • The same recipe—Bayesian optimization over aggregate queries with a decision-aware acquisition function—could be applied to adjacent marketing tasks such as pricing, segmentation, or best-arm selection by replacing the zero-crossing criterion with the relevant policy objective.
  • The paper's comparison implies a broader information-theoretic conjecture: a small number of noisy sign queries can recover a piecewise-constant targeting policy whose value is close to the full-data optimum when segments are broad, which could be tested directly by varying the number of segments in a synthetic policy and measuring the query count needed for parity.
  • Platforms could build the integral-updating Gaussian-process machinery into their query interfaces and let advertisers plug in custom acquisition functions, which would turn privacy-preserving clean rooms from passive reporting tools into active decision-support systems.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper studies personalized targeting when an advertiser can only interact with a platform's data through a limited number of noisy aggregate queries, as in third-party privacy-preserving interfaces. The author proposes a Bayesian optimization procedure, called strategic querying, that (i) uses integral GP updating to learn from queries over ranges rather than points, and (ii) introduces a targeting-aware acquisition function (TAAF) that prioritizes regions with high variance and near-zero posterior mean. The method is compared to a uniform-querying benchmark in simulations and on the Criteo uplift dataset, where strategic querying is reported to achieve 97–101% of the targeting value of a Causal Forest baseline across four privacy settings.

Significance. If the method performs as claimed, this is a practically important result: it would show that effective targeting can be retained under restrictive third-party privacy interfaces, and it would quantify how much of the non-private targeting value can be preserved with fewer than 50 noisy aggregate queries. The paper combines existing ideas from Gaussian process regression for binned data, Bayesian optimization, and policy-aware acquisition in a new application, and it evaluates against an external benchmark (Causal Forest) on a large public dataset. The simulation study systematically varies data-generating parameters and privacy levels, which is a useful contribution. However, the correctness of the GP aggregate-query covariance is central to the method, and the current manuscript contains a concrete specification error in that component; the empirical headline therefore needs to be re-established after the correction.

major comments (4)
  1. [Section 4.2.2, Eq. (6)] The quantity called the 'kernel for averages over a range' is the covariance of the integrals of the GP over the two ranges, not the covariance of the averages. For query outputs defined as averages in Eq. (2), the covariance should include the factor 1/((t-s)(t'-s')). As written, for a single 1-D interval [0,T] with T >> l, Eq. (6) gives a prior variance that grows linearly in T, whereas the variance of a true average must tend to 0 as T grows. This misspecification inflates the posterior variance of large regions. That is exactly the behavior the paper later describes in Section 5.1 as 'regions that are too large naturally have a high variance' and then tries to counteract with size penalties and constraints. More importantly, Section 6.1 states that the empirical application uses TAAF 'with no region size restrictions or penalty,' so the reported 97–101% result relies on either the unnormalized covariance exactly as written or on an implementation that differs from the formula in the text. The authors should correct Eq. (6), re-run the simulation and empirical analyses with the normalized covariance, and clarify which covariance was actually used.
  2. [Section 6.1, query model vs. GP prior] The query output in Eq. (2) is a conditional average over the empirical distribution of X within the queried hyperrectangle, not an unweighted integral of the latent CATE surface. The GP formulation in Section 4.2.2 treats the query as an integral over a range with uniform weighting. This distinction matters in the Criteo application because f0 and f6 are not uniformly distributed and the 'third variable' is a sum of many covariates, so the conditional average is a density-weighted integral. The paper does not describe how the GP accounts for the covariate density inside a queried region. If the implementation ignores this weighting, the posterior predictive mean and variance are computed for a different object than the query actually returns. The authors should state the exact functional relationship between the query statistic and the GP latent function and, if density weighting is ignored, justify that approximation.
  3. [Section 6.2, Table 2] The claim that strategic querying is 'statistically indistinguishable' from Causal Forest is much weaker than the paper's language suggests. The 95% confidence intervals for the ratio to Causal Forest are extremely wide, for example (38%, 151%), (50%, 166%), (58%, 152%), and (49%, 164%) in the four settings. These intervals are consistent with large differences in either direction, so they provide little evidence of equivalence. The phrase '97–101% of Causal Forest performance' should be presented as a point estimate with the associated uncertainty clearly emphasized, and the statistical-indistinguishability claim should be qualified accordingly.
  4. [Section 6.1, variable collapse] The empirical analysis collapses 12 covariates into three: f0, f6, and the sum of all remaining variables. This is justified only by the observation that f0 and f6 have variation across the 25th, 50th, and 75th percentiles in Table 1. The choice is consequential because the reported 97–101% result may depend on the fact that the informative heterogeneity is concentrated in exactly these two variables. The paper does not report any robustness check with respect to the variable collapse or a comparison against using a different subset of covariates. Without such evidence, the generality of the empirical conclusion—that targeting is an information-light task—is not established.
minor comments (4)
  1. [Section 3, Eq. (4)] The differential privacy mechanism is described only through a noise scale that is inversely proportional to the square root of the number of affected users. Please clarify how the constant s relates to a formal DP guarantee (e.g., epsilon and delta) and why this mechanism is appropriate for the range queries used here.
  2. [Section 5.1, Figure 9] The pairwise dominance matrix is difficult to read because the caption does not fully explain the construction of the rows and columns. In particular, the text should define what 'dominated' means in terms of the 95% threshold and should explain why the diagonal entries are zero.
  3. [Section 6.1, hyperparameters] The description of GP hyperparameter estimation says MLE starts at step 10 with fixed values before that, and that the next region is chosen randomly among the top five TAAF values. Please report the sensitivity of the empirical results to these implementation choices, because they can affect the number of effective queries and the exploration-exploitation balance.
  4. [Section 4.2.3, Eq. (9)] The TAAF weighting parameter beta is set to 3 - i/100 in the application, but the scale of beta relative to the posterior variance is not discussed. Since the covariance misspecification in Eq. (6) directly affects the magnitude of V, the interpretation of this schedule is unclear. This should be clarified after the normalization issue is resolved.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the 97-101% result is an external benchmark comparison, not an algebraic consequence of the method's choices.

full rationale

The central derivation chain is not circular. Strategic querying is evaluated against an external benchmark (Causal Forest) on an external dataset (Criteo), under externally imposed query limits and differentially private noise, and the targeting value is computed by IPW on held-out bootstrapped samples. The 97-101% result therefore is not an algebraic consequence of the paper's own GP or acquisition-function assumptions. TAAF is explicitly described as an approximation heuristic (Section 4.2.3), and its motivation via a one-step expected-value calculation is a heuristic analogy rather than a fitted prediction. The simulation uses a Gaussian-process DGP matching the GP prior, which is a model-validation limitation, and the acquisition functions were selected after a disclosed pretest (Appendix), but neither reduces the central Criteo result to the method's inputs by construction. Self-citations to Shchetkina and Berman (2024) are contextual or accompanied by self-contained derivations, so they are not load-bearing. A separate correctness concern exists: Equation 6 as printed omits the inverse-volume normalization of the covariance of range averages, which would miscalibrate the posterior, but this is a formula-error/correctness issue, not a circularity. No circular step is established.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The method is a new algorithm, not a new physical entity, so no invented entities are introduced. The free parameters are mostly standard Bayesian optimization choices and model selection steps. The main additional assumptions are the GP prior, the causal identification assumptions, and the Gaussian noise model. These are standard for the literature but are load-bearing for the method.

free parameters (5)
  • β in TAAF = 3 - i/100 (decreasing with query index i)
    The acquisition function in the empirical application uses a hand-chosen β schedule. No principled selection is given; the paper says it is set to encourage exploration.
  • GP hyperparameters (amplitude, lengthscale, noise) = Estimated via MLE starting at step 10; fixed at 1 and s+0.01 before step 10
    The GP hyperparameters are fit to the queried data, not derived from first principles. This is a standard Bayesian approach but introduces fitting that influences the query selection.
  • Region size penalty and constraint choices = Chosen among 4 variants based on simulation performance
    The best variant (TAAF with size penalty and constraint) is selected based on the simulation study, which is a model selection step that could overfit the simulation settings.
  • Cost of treatment c = 0.01
    Assumed, not estimated. The paper acknowledges this creates meaningful heterogeneity. The empirical results depend on this value.
  • Number of candidate regions considered and randomization among top 5 = 5 (top regions)
    The empirical method chooses randomly among the five regions with the highest TAAF values. This is an ad hoc exploration mechanism.
assumptions (4)
  • domain assumption The treatment effect surface can be modeled as a Gaussian process with a squared exponential kernel.
    The entire Bayesian updating and acquisition function machinery rests on this assumption, which is stated in Section 4.2.2 and used in Section 5.1.
  • domain assumption Ignorability and overlap hold, so the query output is an unbiased estimate of the CATE in the queried region.
    Section 3 assumes random assignment, ignorability, and overlap to equate the expected query output to the CATE.
  • domain assumption The differential privacy noise is zero-mean Gaussian with variance as in Eq. 4.
    The posterior updating uses the noise scale σ_q as the observation noise, which is Gaussian by the paper's assumption.
  • standard math Smith et al. (2018)'s closed-form kernel for averages over ranges is correct and applicable.
    The paper builds directly on this result in Eq. 6 without re-deriving it.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Blind Targeting: Personalization under Third-Party Privacy Constraints." pith.science (2026). https://pith.science/paper/4CAT4FAL

@misc{pith2026250705175,
  author       = {Pith},
  title        = {Pith review of: Blind Targeting: Personalization under Third-Party Privacy Constraints},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4CAT4FAL}},
  note         = {Machine review of arXiv:2507.05175}
}
read the original abstract

Major advertising platforms recently increased privacy protections by limiting advertisers' access to individual-level data. Instead of providing access to granular raw data, the platforms only allow a limited number of aggregate queries to a dataset, which is further protected by adding differentially private noise. This paper studies whether and how advertisers can design effective targeting policies within these restrictive privacy preserving data environments. To achieve this, I develop a probabilistic machine learning method based on Bayesian optimization, which facilitates dynamic data exploration. Since Bayesian optimization was designed to sample points from a function to find its maximum, it is not applicable to aggregate queries and to targeting. Therefore, I introduce two innovations: (i) integral updating of posteriors which allows to select the best regions of the data to query rather than individual points and (ii) a targeting-aware acquisition function that dynamically selects the most informative regions for the targeting task. I identify the conditions of the dataset and privacy environment that necessitate the use of such a "smart" querying strategy. I apply the strategic querying method to the Criteo AI Labs dataset for uplift modeling (Diemert et al., 2018) that contains visit and conversion data from 14M users. I show that an intuitive benchmark strategy only achieves 33% of the non-privacy-preserving targeting potential in some cases, while my strategic querying method achieves 97-101% of that potential, and is statistically indistinguishable from Causal Forest (Athey et al., 2019): a state-of-the-art non-privacy-preserving machine learning targeting method.

Figures

Figures reproduced from arXiv: 2507.05175 by the authors.

Figure 5
Figure 5. Extensions of Bayesian optimization framework [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 9
Figure 9. Pairwise method comparison TAAF, No Penalty, Constraint TAAF, No Penalty, No Constraint TAAF, Penalty, Constraint TAAF, Penalty, No Constraint Var, No Penalty, Constraint Var, No Penalty, No Constraint Var, Penalty, Constraint Var, Penalty, No Constraint Regret, No Penalty, Constraint Regret, No Penalty, No Constraint Regret, Penalty, Constraint Regret, Penalty, No Constraint Uniform Competitor Method TAAF, No Penal… view at source ↗
Figure 10
Figure 10. Strategic vs uniform querying: performance by parameter levels [PITH_FULL_IMAGE:figures/full_fig_p023_10.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

67 extracted references · 61 canonical work pages

  1. [1]

    B., Mironov, I., Talwar, K., and Zhang, L

    Abadi, M., Chu, A., Goodfellow, I., McMahan, H. B., Mironov, I., Talwar, K., and Zhang, L. (2016). Deep learning with differential privacy. In Proceedings of the 2016 ACM SIGSAC conference on computer and communications security , pages 308--318

    2016

  2. [2]

    AWS C lean R ooms differential privacy

    Amazon Web Services (2024). AWS C lean R ooms differential privacy. Accessed via AWS public documentation

    work page 2024

  3. [3]

    Andrews, I., Kitagawa, T., and McCloskey, A. (2024). Inference on winners. The Quarterly Journal of Economics , 139(1):305--358

    2024

  4. [4]

    Athey, S., Catalini, C., and Tucker, C. (2017). The digital privacy paradox: Small money, small costs, small talk. Technical report, National Bureau of Economic Research

    work page 2017

  5. [5]

    Athey, S., Tibshirani, J., and Wager, S. (2019). Generalized random forests. The Annals of Statistics , 47(2):1148--1178

    2019

  6. [6]

    Bleier, A., Goldfarb, A., and Tucker, C. (2020). Consumer privacy and the future of data-based innovation and marketing. International Journal of Research in Marketing , 37(3):466--480

    work page 2020

  7. [7]

    Bondi, T., Rafieian, O., and Yao, Y. J. (2023). Privacy and polarization: An inference-based framework. Available at SSRN 4641822

    work page 2023

  8. [8]

    Bureau, U. C. (2023). Why the C ensus B ureau chose differential privacy. 2020 Census Brief C2020BR-03, U.S. Census Bureau. Accessed: 2025-06-23

    work page 2023

Show all 67 references
  1. [9]

    Campbell, J., Goldfarb, A., and Tucker, C. (2015). Privacy regulation and market structure. Journal of Economics & Management Strategy , 24(1):47--73

    2015

  2. [10]

    Chen, Y.-W., Ascarza, E., and Netzer, O. (2024a). Policy-aware experimentation: Strategic sampling for optimized targeting policies. Columbia Business School Research Paper , (5044549)

    2024

  3. [11]

    M., and Tan, V

    Chen, Z., Karthik, P., Chee, Y. M., and Tan, V. Y. (2024b). Fixed-budget differentially private best arm identification. arXiv preprint arXiv:2401.09073

    2024 arXiv

  4. [12]

    Cheng, S. (2024). Unlock the value of your sensitive data with differential privacy, now generally available. Snowflake blog; Product and Technology category

    2024

  5. [13]

    and Nissim, K

    Cohen, A. and Nissim, K. (2018). Linear program reconstruction in practice. arXiv preprint arXiv:1810.05692

    2018 arXiv

  6. [14]

    Dew, R. (2024). Adaptive preference measurement with unstructured data. Management Science

    2024

  7. [15]

    Dew, R., Padilla, N., and Shchetkina, A. (2024). Your MMM is broken: Identification of nonlinear and time-varying effects in marketing mix models. arXiv preprint arXiv:2408.07678

    2024 arXiv

  8. [16]

    Diemert, E., Betlei, A., Renaudin, C., and Amini, M.-R. (2018). A large scale benchmark for uplift modeling. In KDD

    2018

  9. [17]

    and Nissim, K

    Dinur, I. and Nissim, K. (2003). Revealing information while preserving privacy. In Proceedings of the twenty-second ACM SIGMOD-SIGACT-SIGART symposium on Principles of database systems , pages 202--210

    2003

  10. [18]

    E., and Lynch, J

    Dub \'e , J.-P., Bergemann, D., Demirer, M., Goldfarb, A., Johnson, G., Lambrecht, A., Lin, T., Tuchman, A., Tucker, C. E., and Lynch, J. G. (2024). The intended and unintended consequences of privacy regulation for consumer marketing: A M arketing S cience I nstitute report. ...

    2024

  11. [19]

    and Misra, S

    Dub \'e , J.-P. and Misra, S. (2023). Personalized pricing and consumer welfare. Journal of Political Economy , 131(1):131--189

    2023

  12. [20]

    Dwork, C. (2006). Differential privacy. In International colloquium on automata, languages, and programming , pages 1--12. Springer

    2006

  13. [21]

    Dwork, C., Kohli, N., and Mulligan, D. (2019). Differential privacy in practice: Expose your epsilons! Journal of Privacy and Confidentiality , 9(2)

    2019

  14. [22]

    and Schuster, A

    Friedman, A. and Schuster, A. (2010). Data mining with differential privacy. In Proceedings of the 16th ACM SIGKDD international conference on Knowledge discovery and data mining , pages 493--502

    2010

  15. [23]

    M., and Martindale, C

    Garfinkel, S., Abowd, J. M., and Martindale, C. (2019). Understanding database reconstruction attacks on public data. Communications of the ACM , 62(3):46--53

    2019

  16. [24]

    Garnett, R. (2023). Bayesian optimization . Cambridge University Press

    2023

  17. [25]

    and Tucker, C

    Goldfarb, A. and Tucker, C. (2012). Shifts in privacy concerns. American Economic Review , 102(3):349--353

    2012

  18. [26]

    and Tucker, C

    Goldfarb, A. and Tucker, C. E. (2011). Privacy regulation and online advertising. Management science , 57(1):57--71

    2011

  19. [27]

    Ads D ata H ub for marketers

    Google (2025). Ads D ata H ub for marketers

    2025

  20. [28]

    Greenhill, S., Rana, S., Gupta, S., Vellanki, P., and Venkatesh, S. (2020). Bayesian optimization for adaptive experimental design: A review. IEEE access , 8:13937--13948

    2020

  21. [29]

    Gu, Z., Johnson, G., and Kobayashi, S. (2025). Can privacy technologies replace cookies? A d revenue in a field experiment

    2025

  22. [30]

    C., and Roth, A

    Hsu, J., Gaboardi, M., Haeberlen, A., Khanna, S., Narayan, A., Pierce, B. C., and Roth, A. (2014). Differential privacy: An economic method for choosing epsilon. In 2014 IEEE 27th Computer Security Foundations Symposium , pages 398--410. IEEE

    2014

  23. [31]

    and Ascarza, E

    Huang, T.-W. and Ascarza, E. (2023). Debiasing treatment effect estimation for privacy-protected data: A model audition and calibration approach. Available at SSRN 4575240

    2023

  24. [32]

    and Ascarza, E

    Huang, T.-W. and Ascarza, E. (2024). Doing more with less: Overcoming ineffective long-term targeting using short-term signals. Marketing Science , 43(4):863--884

    2024

  25. [33]

    Huang, T.-W., Ascarza, E., and Israeli, A. (2024). Incrementality representation learning: Synergizing past experiments for intervention personalization . Harvard Business School

    2024

  26. [34]

    Ibragimov, M., Simester, D., and Timoshenko, A. (2020). Improving targeting policies using transfer learning. Available at SSRN 5146292

    2020

  27. [35]

    Inc., A. (2017). Differential privacy overview. Technical report, Apple Inc. Accessed: 2025-06-23

    2017

  28. [36]

    Johnson, G. (2024). Unearthing privacy-enhancing ad technologies ( PEAT ): The adoption of G oogle's P rivacy S andbox. Available at SSRN

    2024

  29. [37]

    A., Shriver, S

    Johnson, G. A., Shriver, S. K., and Du, S. (2020). Consumer privacy choice in online advertising: Who opts out and at what cost to industry? Marketing Science , 39(1):33--51

    2020

  30. [38]

    A., Shriver, S

    Johnson, G. A., Shriver, S. K., and Goldberg, S. G. (2023). Privacy and market concentration: intended and unintended consequences of the GDPR . Management Science , 69(10):5695--5721

    2023

  31. [39]

    Kenthapadi, K., Tran, T., Dietz, M., and Koeppe, I. (2019). Privacy-preserving analytics and reporting at L inked I n. LinkedIn Engineering Blog

    2019

  32. [40]

    and Zuber, C

    Korganbekova, M. and Zuber, C. (2023). Balancing user privacy and personalization. Work in progress

    2023

  33. [41]

    Kushner, H. J. (1964). A new method of locating the maximum point of an arbitrary multipeak curve in the presence of noise

    1964

  34. [42]

    Lin, T. (2022). Valuing intrinsic and instrumental preferences for privacy. Marketing Science , 41(4):663--681

    2022

  35. [43]

    and Strulov-Shlain, A

    Lin, T. and Strulov-Shlain, A. (2025). Choice architecture, privacy valuations, and selection bias in consumer data. Marketing Science

    2025

  36. [44]

    M., and Abernethy, J

    Misra, K., Schwartz, E. M., and Abernethy, J. (2019). Dynamic online pricing with incomplete information using multiarmed bandit experiments. Marketing Science , 38(2):226--252

    2019

  37. [45]

    T., and Raju, J

    Musalem, A., Bradlow, E. T., and Raju, J. S. (2008). Who's got the coupon? E stimating consumer preferences and coupon usage from aggregate information. Journal of Marketing Research , 45(6):715--730

    2008

  38. [46]

    Pyzer-Knapp, E. O. (2018). Bayesian optimization for accelerated drug discovery. IBM Journal of Research and Development , 62(6):2--1

    2018

  39. [47]

    Rafieian, O., Kapoor, A., and Sharma, A. (2025). Multiobjective personalization of marketing interventions. Marketing Science , 44(2):457--477

    2025

  40. [48]

    and Yoganarasimhan, H

    Rafieian, O. and Yoganarasimhan, H. (2021). Targeting and privacy in mobile advertising. Marketing Science , 40(2):193--218

    2021

  41. [49]

    and Yoganarasimhan, H

    Rafieian, O. and Yoganarasimhan, H. (2023). AI and personalization. Artificial Intelligence in Marketing , pages 77--102

    2023

  42. [50]

    R., and Bickford Smith, F

    Rainforth, T., Foster, A., Ivanova, D. R., and Bickford Smith, F. (2024). Modern B ayesian experimental design. Statistical Science , 39(1):100--114

    2024

  43. [51]

    Rosenblatt, L., Allen, J., and Stoyanovich, J. (2022). Spending privacy budget fairly and wisely. arXiv preprint arXiv:2204.12903

    2022 arXiv

  44. [52]

    and Berman, R

    Shchetkina, A. and Berman, R. (2024). When is heterogeneity actionable for personalization? arXiv preprint arXiv:2411.16552

    2024

  45. [53]

    Shi, L., Bakhitov, E., Hung, K., Karrer, B., Walker, C., Bhole, M., and Schrijvers, O. (2024). Scalable analysis of bipartite experiments. arXiv preprint arXiv:2402.11070

    2024

  46. [54]

    J., Stevens, J., Li, J., Parasram, M., Damani, F., Alvarado, J

    Shields, B. J., Stevens, J., Li, J., Parasram, M., Damani, F., Alvarado, J. I. M., Janey, J. M., Adams, R. P., and Doyle, A. G. (2021). Bayesian reaction optimization as a tool for chemical synthesis. Nature , 590(7844):89--96

    2021

  47. [55]

    Simester, D., Timoshenko, A., and Zoumpoulis, S. I. (2020). Targeting prospective customers: Robustness of machine-learning methods to typical data challenges. Management Science , 66(6):2495--2522

    2020

  48. [56]

    Simester, D., Timoshenko, A., and Zoumpoulis, S. I. (2025). A sample size calculation for training and certifying targeting policies. Management Science

    2025

  49. [57]

    T., \'A lvarez, M

    Smith, M. T., \'A lvarez, M. A., and Lawrence, N. D. (2018). Gaussian process regression for binned data. arXiv preprint arXiv:1809.02010

    2018 arXiv

  50. [58]

    Snoek, J., Larochelle, H., and Adams, R. P. (2012). Practical B ayesian optimization of machine learning algorithms. Advances in neural information processing systems , 25

    2012

  51. [59]

    Tucker, C. E. (2014). Social networks, personalized advertising, and privacy controls. Journal of marketing research , 51(5):546--562

    2014

  52. [60]

    T., and Moakler, R

    Wernerfelt, N., Tuchman, A., Shapiro, B. T., and Moakler, R. (2025). Estimating the value of offsite tracking data to advertisers: Evidence from M eta. Marketing Science , 44(2):268--286

    2025

  53. [61]

    Williams, C. K. and Rasmussen, C. E. (2006). Gaussian processes for machine learning , volume 2. MIT press Cambridge, MA

    2006

  54. [62]

    and Bastani, H

    Xu, K. and Bastani, H. (2025). Multitask learning and bandits via robust statistics. Management Science

    2025

  55. [63]

    Yoganarasimhan, H., Barzegary, E., and Pani, A. (2023). Design and evaluation of optimal free trials. Management Science , 69(6):3220--3240

    2023

  56. [64]

    and Misra, S

    Zhang, W. and Misra, S. (2024). Coarse personalization. In Proceedings of the 25th ACM Conference on Economics and Computation , pages 1206--1208

    2024

  57. [65]

    Zhang, W. W. (2023). Optimal comprehensible targeting. Working Paper

    2023

  58. [66]

    J., and Kosorok, M

    Zhao, Y., Zeng, D., Rush, A. J., and Kosorok, M. R. (2012). Estimating individualized treatment rules using outcome weighted learning. Journal of the American Statistical Association , 107(499):1106--1118

    2012

  59. [67]

    Zhou, Z., Athey, S., and Wager, S. (2023). Offline multi-action policy learning: Generalization and optimization. Operations Research , 71(1):148--183

    2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.