Pith. sign in

REVIEW 2 major objections 4 minor 78 references

Learning Fair And Effective Points-Based Rewards Programs

T0 review · 2 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper proves fairness is cheap in rewards programs: one threshold loses at most 1.7x revenue, and devaluation-free learning stays optimal.

desk verdict Genuinely new learning result with a clever Markov-chain natural-exploration argument, but Theorem 1's proof has a real +∞ gap and Corollary 2's expectation conversion is wrong as stated; both look fixable. read the letter →

arxiv 2506.03911 v1 pith:DONOYIEW submitted 2025-06-04 cs.LG math.OC

Chamsi Hssaine , Yichun Hu , Ciara Pike-Burke This is my paper · ORCID
classification cs.LGmath.OC MSC 91B2490B5060J1062L1268W27
keywords points-basedrewardsprogramsBuyNGetOneFreepriceoffairnesstemporalgeneralizedlinearmodelregretminimizationMarkovchainnaturalexplorationonlinelearning
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 fairness constraints in 'Buy N, Get One Free' rewards programs force a significant revenue sacrifice, and whether a seller can learn a good redemption threshold without repeatedly devaluing customers' points. Its first answer is a worst-case bound: the optimal program that sets one common threshold for all customer types earns at least $1/(1+\ln 2)\approx 0.59$ of the revenue of a personalized program, with a tighter bound that is exact at $3/2$ for two equally weighted types. Its second answer is algorithmic: a greedy epoch-based learner that re-estimates an unknown generalized-linear purchase model and changes thresholds only $O(\log T)$ times over horizon $T$ achieves the optimal $\widetilde{O}(\sqrt{MT})$ regret, and a version that only ever lowers thresholds—never devaluing earned points—pays only a constant factor. The engine is that customers' redemption cycles generate enough internal variability in their remaining points to substitute for explicit exploration. A sympathetic reader would take the paper to establish that both individual and temporal fairness are cheap in this class of programs.

What carries the argument

The load-bearing object is the points-to-redemption Markov chain for a customer under a fixed threshold $N$, whose steady-state distribution has the closed form $p_k(\tau;N)=1/\sum_{\tau'=0}^{N}\phi_k(\tau)/\phi_k(\tau')$; this yields the long-run revenue formula $R_k(N)=N/\sum_{\tau=0}^{N}1/\phi_k(\tau)$. Under the generalized linear model, the Fisher information (design) matrix of an epoch is built from the observed points-to-redemption, and the proof's crux is a uniform lower bound on its minimum eigenvalue, obtained by lower-bounding the steady-state variance of the chain by $\mu_{\min}^2 N(N+2)/(12\mu_{\max}^2)$. A new Chernoff-type concentration inequality for functions of this state-dependent Markov chain—adapted from mixing-time arguments—turns that variance bound into a high-probability guarantee that greedy MLE estimates are accurate. Fair-Greedy adds nested consideration sets: a threshold survives into the next epoch only if its estimated revenue is within $2\Delta_h$ of the best in the current set, and the algorithm picks the largest survivor, which forces the threshold to move only downward.

What would settle it

Construct a single-type instance satisfying the GLM and monotonicity assumptions but with $\phi(0)=\varepsilon$ and $\phi(\tau)=1$ for $\tau\ge 1$, so the steady-state variance of points-to-redemption is $O(\varepsilon N^2)$; as $\varepsilon\to 0$, Lemma 12's variance lower bound tends to zero and the epoch design matrix's minimum eigenvalue stops growing linearly with $MT_h$, so if the greedy MLE is run without explicit exploration, the regret should fail to be $\widetilde{O}(\sqrt{MT})$ for small $\varepsilon$. That calculation would directly test the paper's natural-exploration claim.

Watch

Extended reading notes

Core claim

The paper's central claim is that the price of fairness for BNGO programs is uniformly bounded: for any instance, the optimal personalized long-run revenue is at most $K-(K-1)2^{-1/(K-1)}\le 1+\ln 2$ times the optimal non-personalized revenue, and for $K=2$ the worst-case ratio is exactly $3/2$. In the learning setting, it claims that greedy maximum-likelihood threshold selection is not only stable but optimal: the epoch-based Stable-Greedy algorithm attains $\widetilde{O}(\sqrt{MT})$ regret with high probability, matching the $\Omega(\sqrt{MT})$ lower bound that holds for any policy, while changing the redemption threshold only $O(\log T)$ times. The devaluation-free variant Fair-Greedy, which restricts itself to nested consideration sets and therefore never increases the threshold, retains the same order of regret with only a factor-of-two worsening. The deeper claim is that no explicit exploration is needed: the Markov chain that traces a customer's remaining points to redemption supplies the stochastic diversity that would normally require deliberate price experimentation.

Load-bearing premise

The regret guarantees depend on a known positive lower bound $\mu_{\min}$ for every purchase probability at every point in the redemption cycle, because that bound is what guarantees redemption cycles run fast enough to create the natural variation the estimator needs.

Editorial extensions

If this is right

  • A seller who intentionally keeps one redemption threshold for everyone loses at most a factor of $1/(1+\ln 2)$ of personalized long-run revenue, and with two equally weighted customer types the worst-case retained fraction is exactly $2/3$.
  • No learning policy—even one allowed to change thresholds every period—can improve on $\Omega(\sqrt{MT})$ worst-case regret, so Stable-Greedy's rate is the best possible up to polylogarithmic factors.
  • A devaluation-free policy, one that never raises the redemption threshold, achieves the same order-optimal regret, so temporal fairness can be obtained at only a constant-factor revenue cost.
  • Because thresholds change only $O(\log T)$ times, customers experience long stretches of stable goals, and increasing the population size $M$ shrinks the per-customer regret as $\widetilde{O}(\sqrt{T/M})$.

Reading between the lines

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

  • This proof strategy suggests a general recipe for greedy optimality: in any online learning problem where the 'context' is a controlled Markov chain rather than an i.i.d. draw, a uniform lower bound on the stationary variance of that chain may replace the usual positive-definiteness assumptions on exogenous contexts.
  • The worst-case price-of-fairness bound is probably conservative for real loyalty programs; based on the paper's own random instances, one would expect typical uniform-threshold policies to retain well above 80% of personalized revenue, which a field study with panel purchase data could test.
  • For customer segments with very low baseline purchase probabilities, the algorithms' practical performance degrades sharply in the paper's misspecification experiments, suggesting that deployments should pair these learners with a conservative initial threshold or a short cold-start exploration phase when $\mu_{\min}$ is likely to be tiny.
  • The bounded price of fairness implies a ceiling on the hidden value of personalization for airlines and hotels under consumer-protection scrutiny: even if differentiated thresholds are attractive, a uniform program cannot lose more than a constant fraction of revenue in the worst case.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 4 minor

Summary. The paper studies fairness in Buy-N-Get-One-Free (BNGO) rewards programs. It first proves a price-of-fairness bound, showing that the optimal personalized revenue is at most K-(K-1)2^{-1/(K-1)} <= 1+ln2 times the optimal non-personalized revenue. It then considers learning an optimal non-personalized threshold under a GLM demand model, proposes an epoch-based greedy algorithm (Stable-Greedy) that changes the threshold O(log T) times, proves a high-probability regret bound of O~(sqrt(M T)) with matching lower bound Omega(sqrt(M T)), and presents a devaluation-free variant (Fair-Greedy) with only a constant-factor regret loss. Numerical experiments illustrate the price of fairness on random instances and the practical performance of both algorithms.

Significance. If the main results hold, the paper makes several valuable contributions: a uniform, instance-independent bound on the price of fairness for BNGO programs; a demonstration that greedy, exploration-free learning is optimal in a Markov-modulated demand setting, via a new Markov-chain concentration analysis; and a devaluation-free algorithm with order-optimal regret. The proofs are detailed and the lower-bound construction is careful, using standard information-theoretic tools. The numerical study is extensive and includes misspecification experiments. The two issues raised below affect the two central theoretical claims and should be resolved before the results can be fully credited.

major comments (2)
  1. [Section 3 / Appendix B.1 (Theorem 1)] The proof of Theorem 1 is not defined when an optimal personalized threshold is N_k^* = +infinity. The proof defines R_k^pers = rho_k * N_k^* / sum_{tau=0}^{N_k^*} 1/phi_k(tau) and then applies Proposition 1 to the uniform vector (N_k^*, ..., N_k^*); both operations are undefined for N_k^* = +infinity. This is not a corner case: the feasible set is [N_max] union {+infinity}, and the tightness example in Lemma 8 uses a type with N_2^* = +infinity. Equations (27)-(29) and Lemma 6 implicitly treat every N_k^* as a finite integer, and the manuscript provides no separate limiting argument (e.g., through Corollary 1) for types where no-loyalty is optimal. The result may be salvageable by treating +infinity as the limit a_k = N_k^*/(N_k^*+1) -> 1 with R_k^pers -> rho_k * phi_bar_k, but this case is absent from the proof.
  2. [Section 5.2 / Corollary 2] The conversion from the high-probability bound in Theorem 3 to the expected regret bound in Corollary 2 omits the contribution of the failure event. Theorem 3 holds with probability at least 1 - 7 K H(T) delta; with delta = O(1/sqrt(T)), the failure probability is O(log T / sqrt(T)). On the failure event the regret can be as large as O(M T), so the failure contribution to the expectation is O(M sqrt(T) log T). This term is not dominated by the good-event bound O~(sqrt(M T)) when M grows with T, and it is not captured by the claimed O~(sqrt(M T) + M/sqrt(T)) in Corollary 2. The corollary needs either a smaller delta (e.g., delta = 1/T) or an explicit accounting of the failure term; as stated, the expected-regret claim and the resulting optimality statement are not justified.
minor comments (4)
  1. [Section 6 / Theorem 4] The statement of Theorem 4 says "Algorithm 1 guarantees" but the theorem concerns Algorithm 2 (Fair-Greedy); this appears to be a typo and should be corrected.
  2. [Appendix C.1 / Proof of Theorem 2] The proof states that Assumption 1 holds for both constructed instances for any Delta in (0, 1/2], but at Delta = 1/2 the purchase probability at tau = 2 is zero, violating Assumption 1(a) which requires a strictly positive mu_min. The Delta used in the optimized bound is small, so the lower bound itself is not affected, but the verification statement is inaccurate.
  3. [Appendix B.1.3 / Lemma 8] In the final sentence of Lemma 8, "P oF = 3/2" contains a typo ("P oF" should be "PoF").
  4. [Abstract / Section 1.1] The abstract states that the algorithm achieves optimal ~O(sqrt(T)) regret in expectation, while Theorem 2 gives a lower bound of Omega(sqrt(M T)). The claims are consistent when M is treated as a fixed constant, but the dependence on M should be made explicit in the abstract to avoid ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the theoretical bounds are derived from explicit model assumptions, concentration inequalities, and standard external results; the simulations do not feed back into the proofs.

full rationale

The paper's derivation chain is self-contained with respect to its own claims. Theorem 1 follows from Proposition 1, which gives a closed-form expression for long-run average revenue derived from a finite-state Markov chain, and the subsequent algebraic bounds in Lemmas 6-7 do not presuppose the price-of-fairness bound. The regret guarantees in Theorems 3 and 4 rely on MLE concentration, design-matrix lower bounds, and Markov-chain mixing arguments; the lower bound in Theorem 2 constructs two explicit instances with known parameters and applies the Bretagnolle-Huber inequality, so no fitted quantity is renamed as a prediction. The only self-citation, to Freund and Hssaine (2025), appears in the related-work discussion and is not load-bearing for any theorem. The proof of Theorem 1 has an omitted edge case concerning +infinity thresholds, but that is a completeness issue rather than circularity, since the argument is not presupposing its conclusion. Overall, no derivation reduces to its own inputs.

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

The central claims rest on the structural GLM and bounded-probability assumptions, plus standard stochastic and statistical tools. There are no invented physical or economic entities, and no free parameters fitted to data in the theoretical results.

assumptions (5)
  • domain assumption Customer purchase probabilities follow a Generalized Linear Model φ_k(τ)=μ_k(β_{k,1}+β_{k,2}τ), with μ_k known and β unknown.
    Stated in Section 4.1 as the behavioral model for the learning setting.
  • domain assumption There exist known constants μ_min, μ_max with μ_min ≤ φ_k(τ) ≤ μ_max for all k and τ.
    Assumption 1(a), used to bound the steady-state variance and mixing time.
  • domain assumption Customers are non-strategic and make independent Bernoulli purchase decisions; no stockpiling; no cash purchase once at the redemption threshold.
    Modeling assumptions in Section 2, explicitly acknowledged by the authors.
  • standard math Standard concentration inequalities and MLE bounds for generalized linear models hold.
    The proofs rely on Paulin (2015), Hoeffding bounds, and Li et al. (2017), cited in Appendix D and F.
  • standard math The redemption Markov chain is finite, irreducible, and has a stationary distribution.
    Used in Proposition 1 and the mixing time analysis.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning Fair And Effective Points-Based Rewards Programs." pith.science (2026). https://pith.science/paper/DONOYIEW

@misc{pith2026250603911,
  author       = {Pith},
  title        = {Pith review of: Learning Fair And Effective Points-Based Rewards Programs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DONOYIEW}},
  note         = {Machine review of arXiv:2506.03911}
}
abstract

Points-based rewards programs are a prevalent way to incentivize customer loyalty; in these programs, customers who make repeated purchases from a seller accumulate points, working toward eventual redemption of a free reward. These programs have recently come under scrutiny due to accusations of unfair practices in their implementation. Motivated by these concerns, we study the problem of fairly designing points-based rewards programs, with a focus on two obstacles that put fairness at odds with their effectiveness. First, due to customer heterogeneity, the seller should set different redemption thresholds for different customers to generate high revenue. Second, the relationship between customer behavior and the number of accumulated points is typically unknown; this requires experimentation which may unfairly devalue customers' previously earned points. We first show that an individually fair rewards program that uses the same redemption threshold for all customers suffers a loss in revenue of at most a factor of $1+\ln 2$, compared to the optimal personalized strategy that differentiates between customers. We then tackle the problem of designing temporally fair learning algorithms in the presence of demand uncertainty. Toward this goal, we design a learning algorithm that limits the risk of point devaluation due to experimentation by only changing the redemption threshold $O(\log T)$ times, over a horizon of length $T$. This algorithm achieves the optimal (up to polylogarithmic factors) $\widetilde{O}(\sqrt{T})$ regret in expectation. We then modify this algorithm to only ever decrease redemption thresholds, leading to improved fairness at a cost of only a constant factor in regret. Extensive numerical experiments show the limited value of personalization in average-case settings, in addition to demonstrating the strong practical performance of our proposed learning algorithms.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

78 extracted references · 68 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address author booktitle chapter doi edition editor eid howpublished institution isbn issn journal key month note number organization pages publisher school series title type url volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.all := #1...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in "" FUNCTION format.date year ...

  3. [3]

    Management Science 67(9):5549--5568

    Ban GY, Keskin NB (2021) Personalized dynamic pricing with machine learning: High-dimensional features and heterogeneous elasticity. Management Science 67(9):5549--5568

    work page 2021

  4. [4]

    ACM SIGMETRICS Performance Evaluation Review 51(1):55--56

    Banerjee S, Hssaine C, Sinclair SR (2023) Online fair allocation of perishable resources. ACM SIGMETRICS Performance Evaluation Review 51(1):55--56

    work page 2023

  5. [5]

    Management Science 67(3):1329--1349

    Bastani H, Bayati M, Khosravi K (2021) Mostly exploration-free algorithms for contextual bandits. Management Science 67(3):1329--1349

    work page 2021

  6. [6]

    Manufacturing & Service Operations Management 24(4):2010--2028

    Bastani H, Harsha P, Perakis G, Singhvi D (2022) Learning personalized product recommendations with customer disengagement. Manufacturing & Service Operations Management 24(4):2010--2028

    2022

  7. [7]

    Management Science 61(4):723--739

    Besbes O, Zeevi A (2015) On the (surprising) sufficiency of linear models for dynamic pricing with demand learning. Management Science 61(4):723--739

    work page 2015

  8. [8]

    Boucheron S, Lugosi G, Massart P (2013) Concentration Inequalities: A Nonasymptotic Theory of Independence (Oxford University Press)

    work page 2013

Show all 78 references
  1. [9]

    Broder J (2011) Online algorithms for revenue management

    2011

  2. [10]

    Operations Research 60(4):965--980

    Broder J, Rusmevichientong P (2012) Dynamic pricing under a general parametric choice model. Operations Research 60(4):965--980

    2012

  3. [11]

    Brown LD (1986) Fundamentals of statistical exponential families: with applications in statistical decision theory (Ims)

    1986

  4. [12]

    Advances in Neural Information Processing Systems 26

    Cesa-Bianchi N, Dekel O, Shamir O (2013) Online learning with switching costs and other adaptive adversaries. Advances in Neural Information Processing Systems 26

    2013

  5. [13]

    Operations Research 68(5):1445--1456

    Chen B, Chao X, Wang Y (2020) Data-based dynamic pricing and inventory control with censored demand and limited price changes. Operations Research 68(5):1445--1456

    2020

  6. [14]

    arXiv preprint arXiv:2111.08221

    Chen X, Lyu J, Zhang X, Zhou Y (2021 a ) Fairness-aware online price discrimination with nonparametric demand models. arXiv preprint arXiv:2111.08221

    2021 arXiv

  7. [15]

    arXiv preprint arXiv:2311.16528

    Chen X, Simchi-Levi D, Wang Y (2023) Utility fairness in contextual dynamic pricing with demand learning. arXiv preprint arXiv:2311.16528

    2023 arXiv

  8. [16]

    Journal of Business Research 124:179--197

    Chen Y, Mandler T, Meyer-Waarden L (2021 b ) Three decades of research on loyalty programs: A literature review and future research agenda. Journal of Business Research 124:179--197

    2021

  9. [17]

    Operations Research 65(6):1722--1731

    Cheung WC, Simchi-Levi D, Wang H (2017) Dynamic pricing and demand learning with limited price experimentation. Operations Research 65(6):1722--1731

    2017

  10. [18]

    Manufacturing & Service Operations Management 22(2):257--272

    Chun SY, Iancu DA, Trichakis N (2020) Loyalty program liabilities and point values. Manufacturing & Service Operations Management 22(2):257--272

    2020

  11. [19]

    Management Science 65(9):3969--3987

    Chun SY, Ovchinnikov A (2019) Strategic consumers, revenue management, and the design of loyalty programs. Management Science 65(9):3969--3987

    2019

  12. [20]

    Manufacturing & Service Operations Management 24(4):2134--2149

    Chung H, Ahn HS, Chun SY (2022) Dynamic pricing with point redemption. Manufacturing & Service Operations Management 24(4):2134--2149

    2022

  13. [21]

    Then came the backlash

    CNN (2023) Best Buy, Dunkin’ and Starbucks changed their rewards programs. Then came the backlash . https://www.cnn.com/2023/01/14/business/best-buy-rewards-dunkin-starbucks-ctpr/index.html, Accessed: 2025-01-17

    2023

  14. [22]

    Management Science 68(12):8536--8552

    Cohen MC, Elmachtoub AN, Lei X (2022) Price discrimination with fairness constraints. Management Science 68(12):8536--8552

    2022

  15. [23]

    Operations Research

    Cohen MC, Miao S, Wang Y (2025) Dynamic pricing with fairness constraints. Operations Research

    2025

  16. [24]

    IEEE Transactions on Information theory 52(3):1007--1016

    Csisz \'a r I, Talata Z (2006) Context tree estimation for not necessarily finite memory processes, via bic and mdl. IEEE Transactions on Information theory 52(3):1007--1016

    2006

  17. [25]

    Proceedings of the Forty-Sixth Annual ACM Symposium on Theory of Computing , 459--467

    Dekel O, Ding J, Koren T, Peres Y (2014) Bandits with switching costs: T^ 2/3 regret . Proceedings of the Forty-Sixth Annual ACM Symposium on Theory of Computing , 459--467

    2014

  18. [26]

    Surveys in operations research and Management Science 20(1):1--18

    den Boer AV (2015) Dynamic pricing and learning: historical origins, current research, and new directions. Surveys in operations research and Management Science 20(1):1--18

    2015

  19. [27]

    Management Science 60(3):770--783

    den Boer AV, Zwart B (2014) Simultaneously learning and optimizing using controlled variance pricing. Management Science 60(3):770--783

    2014

  20. [28]

    International Journal of Management Reviews 14(3):217--237

    Dorotic M, Bijmolt TH, Verhoef PC (2012) Loyalty programmes: Current knowledge and research directions. International Journal of Management Reviews 14(3):217--237

    2012

  21. [29]

    International Journal of Research in Marketing 15(5):459--471

    Dr \`e ze X, Hoch SJ (1998) Exploiting the installed base using cross-merchandising and category destination programs. International Journal of Research in Marketing 15(5):459--471

    1998

  22. [30]

    Management Science 67(10):6055--6070

    Elmachtoub AN, Gupta V, Hamilton ML (2021) The value of personalized pricing. Management Science 67(10):6055--6070

    2021

  23. [31]

    Operations Research 73(2):775--797

    Feng Q, Zhu R, Jasin S (2025) Temporal fairness in learning and earning: Price protection guarantee and phase transitions. Operations Research 73(2):775--797

    2025

  24. [32]

    Advances in Neural Information Processing Systems 23

    Filippi S, Cappe O, Garivier A, Szepesv \'a ri C (2010) Parametric bandits: The generalized linear case. Advances in Neural Information Processing Systems 23

    2010

  25. [33]

    Production and Operations Management 34(1):16--29

    Freund D, Hssaine C (2025) Fair incentives for repeated engagement. Production and Operations Management 34(1):16--29

    2025

  26. [34]

    Available at SSRN 3871915

    Hamilton M, Singal R (2023) Churning while experimenting: Maximizing user engagement in recommendation platforms. Available at SSRN 3871915

    2023

  27. [35]

    Quantitative Marketing and Economics 6:109--137

    Hartmann WR, Viard VB (2008) Do frequency reward programs create switching costs? a dynamic structural analysis of demand in a reward program. Quantitative Marketing and Economics 6:109--137

    2008

  28. [36]

    Journal of Comparative Psychology 17(3):393

    Hull CL (1934) The rat's speed-of-locomotion gradient in the approach to food. Journal of Comparative Psychology 17(3):393

    1934

  29. [37]

    arXiv preprint arXiv:2402.08533

    Jaillet P, Podimata C, Zhou Z (2024) Grace period is all you need: Individual fairness without revenue loss in revenue management. arXiv preprint arXiv:2402.08533

    2024 arXiv

  30. [38]

    Journal of Machine Learning Research 20(9):1--49

    Javanmard A, Nazerzadeh H (2019) Dynamic pricing in high-dimensions. Journal of Machine Learning Research 20(9):1--49

    2019

  31. [39]

    Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 296--314

    Kallus N, Zhou A (2021) Fairness, welfare, and equity in personalized pricing. Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 296--314

    2021

  32. [40]

    Mathematics of Operations Research 49(2):1192--1222

    Kanoria Y, Lobel I, Lu J (2024) Managing customer churn via service mode control. Mathematics of Operations Research 49(2):1192--1222

    2024

  33. [41]

    Operations Research 62(5):1142--1167

    Keskin NB, Zeevi A (2014) Dynamic pricing with an unknown demand model: Asymptotically optimal semi-myopic policies. Operations Research 62(5):1142--1167

    2014

  34. [42]

    Marketing Science 20(2):99--120

    Kim BD, Shi M, Srinivasan K (2001) Reward programs and tacit collusion. Marketing Science 20(2):99--120

    2001

  35. [43]

    Management Science 50(4):503--520

    Kim BD, Shi M, Srinivasan K (2004) Managing capacity through reward programs. Management Science 50(4):503--520

    2004

  36. [44]

    Journal of Marketing Research 43(1):39--58

    Kivetz R, Urminsky O, Zheng Y (2006) The goal-gradient hypothesis resurrected: Purchase acceleration, illusionary goal progress, and customer retention. Journal of Marketing Research 43(1):39--58

    2006

  37. [45]

    The Quarterly Journal of Economics 102(2):375--394

    Klemperer P (1987) Markets with consumer switching costs. The Quarterly Journal of Economics 102(2):375--394

    1987

  38. [46]

    Review of Marketing Science 1(1):0000102202154656161002

    Kopalle PK, Neslin SA (2003) The economic viability of frequency reward programs in a strategic competitive environment. Review of Marketing Science 1(1):0000102202154656161002

    2003

  39. [47]

    Marketing Science 31(2):216--235

    Kopalle PK, Sun Y, Neslin SA, Sun B, Swaminathan V (2012) The joint sales impact of frequency reward and customer tier components of loyalty programs. Marketing Science 31(2):216--235

    2012

  40. [48]

    Quantitative Marketing and Economics 1:179--202

    Lal R, Bell DE (2003) The impact of frequent shopper programs in grocery retailing. Quantitative Marketing and Economics 1:179--202

    2003

  41. [49]

    Lattimore T, Szepesv \'a ri C (2020) Bandit algorithms (Cambridge University Press)

    2020

  42. [50]

    Levin DA, Peres Y (2017) Markov chains and mixing times, volume 107 (American Mathematical Soc.)

    2017

  43. [51]

    Journal of Marketing Research 41(3):281--292

    Lewis M (2004) The influence of loyalty programs and short-term promotions on customer retention. Journal of Marketing Research 41(3):281--292

    2004

  44. [52]

    International Conference on Machine Learning, 2071--2080 (PMLR)

    Li L, Lu Y, Zhou D (2017) Provably optimal algorithms for generalized linear contextual bandits. International Conference on Machine Learning, 2071--2080 (PMLR)

    2017

  45. [53]

    Journal of Marketing 71(4):19--35

    Liu Y (2007) The long-term impact of loyalty programs on consumer purchase behavior and loyalty. Journal of Marketing 71(4):19--35

    2007

  46. [54]

    buy x, get one free

    Liu Y, Sun Y, Zhang D (2021) An analysis of “buy x, get one free” reward programs. Operations Research 69(6):1823--1841

    2021

  47. [55]

    Journal of Machine Learning Research 24(328):1--44

    Lugosi G, Pike-Burke C, Savalle PA (2023) Bandit problems with fidelity rewards. Journal of Machine Learning Research 24(328):1--44

    2023

  48. [56]

    Available at SSRN

    Lyu C, Zhang D (2024) Customer reward programs for two-sided markets. Available at SSRN

    2024

  49. [57]

    https://www.mcdonalds.com/ca/en-ca/getmoremcds/mymcdonaldsrewards.html, Accessed: 2025-01-17

    McDonald's (2025) MyMcDonald's Rewards . https://www.mcdonalds.com/ca/en-ca/getmoremcds/mymcdonaldsrewards.html, Accessed: 2025-01-17

    2025

  50. [58]

    https://www.nj.com/business/2013/11/made_in_jersey_sh_green_stamps.html, Accessed: 2025-01-17

    NJcom (2013) Made in Jersey: S&H Green Stamps - in the sixties, Americans were stuck on them . https://www.nj.com/business/2013/11/made_in_jersey_sh_green_stamps.html, Accessed: 2025-01-17

    2013

  51. [59]

    Electronic Journal of Probability 20(79):1--32

    Paulin D (2015) Concentration inequalities for markov chains by marton couplings and spectral methods. Electronic Journal of Probability 20(79):1--32

    2015

  52. [60]

    https://www.pcworld.com/article/2160414/microsoft-guts-microsoft-rewards-points-and-its-fans-are-outraged.html, Accessed: 2025-01-17

    PCWorld (2023) Microsoft guts microsoft rewards points, and its fans are outraged. https://www.pcworld.com/article/2160414/microsoft-guts-microsoft-rewards-points-and-its-fans-are-outraged.html, Accessed: 2025-01-17

    2023

  53. [61]

    Operations Research 72(6):2726--2744

    Perakis G, Singhvi D (2024) Dynamic pricing with unknown nonparametric demand and limited price changes. Operations Research 72(6):2726--2744

    2024

  54. [62]

    https://www.pymnts.com/news/loyalty-and-rewards-news/2023/chick-fil-a-joins-qsrs-watering-down-rewards-programs-amid-inflation/, Accessed: 2025-01-17

    PYMNTS (2023) Chick-fil-a waters down rewards and hopes customers stick around. https://www.pymnts.com/news/loyalty-and-rewards-news/2023/chick-fil-a-joins-qsrs-watering-down-rewards-programs-amid-inflation/, Accessed: 2025-01-17

    2023

  55. [63]

    arXiv preprint arXiv:1604.07463

    Qiang S, Bayati M (2016) Dynamic pricing with demand covariates. arXiv preprint arXiv:1604.07463

    2016 arXiv

  56. [64]

    https://www.reddit.com/r/Chipotle/comments/xsydih/reward_points_required_for_free_entree_raised_to/?rdt=64941, Accessed: 2025-01-17

    Reddit (2023) Reward points required for free entree raised to 1625 from 1400. https://www.reddit.com/r/Chipotle/comments/xsydih/reward_points_required_for_free_entree_raised_to/?rdt=64941, Accessed: 2025-01-17

    2023

  57. [65]

    ACM SIGMETRICS Performance Evaluation Review 50(1):95--96

    Sinclair SR, Banerjee S, Yu CL (2022) Sequential fair allocation: Achieving the optimal envy-efficiency tradeoff curve. ACM SIGMETRICS Performance Evaluation Review 50(1):95--96

    2022

  58. [66]

    Singh SS, Jain DC, Krishnan TV (2008) Research note—customer loyalty programs: Are they profitable? Management Science 54(6):1205--1211

    2008

  59. [67]

    https://www.starbucks.com/rewards, Accessed: 2025-01-17

    Starbucks (2025) Starbucks rewards. https://www.starbucks.com/rewards, Accessed: 2025-01-17

    2025

  60. [68]

    Available at SSRN 4537925

    Sumida M, Zhou A (2023) Optimizing and learning assortment decisions in the presence of platform disengagement. Available at SSRN 4537925

    2023

  61. [69]

    Management Science 65(8):3889--3903

    Sun Y, Zhang D (2019) A model of customer reward programs with finite expiration terms. Management Science 65(8):3889--3903

    2019

  62. [70]

    https://www.tacobell.com/rewards, Accessed: 2025-01-17

    Taco Bell (2025) Taco Bell Rewards . https://www.tacobell.com/rewards, Accessed: 2025-01-17

    2025

  63. [71]

    Journal of Retailing 81(4):293--305

    Taylor GA, Neslin SA (2005) The current and future sales impact of a retail frequency reward program. Journal of Retailing 81(4):293--305

    2005

  64. [72]

    https://www.theguardian.com/business/2018/jan/17/tesco-delays-clubcard-changes-customer-backlash-reward-loyalty-scheme#: :text=Tesco Accessed: 2025-01-17

    The Guardian (2018) Tesco delays Clubcard changes after customer backlash . https://www.theguardian.com/business/2018/jan/17/tesco-delays-clubcard-changes-customer-backlash-reward-loyalty-scheme#: :text=Tesco Accessed: 2025-01-17

    2018

  65. [73]

    Airlines' Rewards Practices

    US Department of Transportation (2024) USDOT Seeks to Protect Consumers' Airline Rewards in Probe of Four Largest U.S. Airlines' Rewards Practices . https://www.transportation.gov/briefing-room/usdot-seeks-protect-consumers-airline-rewards-probe-four-largest-us-airlines-reward...

    2024

  66. [74]

    https://www.vox.com/money/354191/loyalty-rewards-programs-sephora-vib-amazon, Accessed: 2025-01-17

    Vox (2024) The golden age of retail loyalty programs is here . https://www.vox.com/money/354191/loyalty-rewards-programs-sephora-vib-amazon, Accessed: 2025-01-17

    2024

  67. [75]

    Wainwright MJ (2019) High-dimensional statistics: A non-asymptotic viewpoint, volume 48 (Cambridge university press)

    2019

  68. [76]

    https://www.wendys.com/rewards, Accessed: 2025-01-17

    Wendy's (2025) Wendy's Rewards . https://www.wendys.com/rewards, Accessed: 2025-01-17

    2025

  69. [77]

    International Conference on Artificial Intelligence and Statistics, 9941--9975 (PMLR)

    Xu J, Qiao D, Wang YX (2023) Doubly fair dynamic pricing. International Conference on Artificial Intelligence and Statistics, 9941--9975 (PMLR)

    2023

  70. [78]

    Available at SSRN

    Yang Z, Lei X, Gao P (2023) Regulating discriminatory pricing in the presence of tacit collusion. Available at SSRN

    2023

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.