acceptodds
Under review as a conference paper at ICLR 2027

A General Framework for Budgeted Threshold Incentives on Request

Abstract

On-demand delivery platforms pay riders through incentive activities whose tiers are set from recent completions of riders with a similar history, so that a little extra effort earns a clearly stated reward. Operators request such plans for changing periods, rider populations, payment rules and budgets, days ahead and within minutes, often for holidays or bad weather where randomized trials are scarce and take months to collect. We present a request-driven framework that composes four stages—conditional prediction, population reduction, trajectory integration and budget allocation—through seven replaceable modules that exchange conditional trajectory laws, whose award probabilities and award-marked moments give payment and uplift for any activity rule. To shorten a long randomized campaign, a response-correction step reweights trajectories from abundant no-offer history to match the moments of a short pilot. We prove that, on a fixed plan menu and given the stage errors, the end-to-end value loss is bounded by the sum of four stage terms—synthetic data, moment matching, integration and decision—with constants that cannot be improved from the final tables, and that for every stage there are instances on which omitting it leaves an error floor the others cannot remove. On 3,000 riders over 45 weekly origins, re-drawing trajectories and solving exactly for every request takes 117.8 s per week-long request against 3.07 s for the framework; including the one-off sampling pass, all 127 windows of a week are answered 11.04× faster with identical scenarios and at most 0.92% value lost by the allocation on the audited week-long tables, a gain that comes entirely from reuse, while a point forecast, independent days or an equal budget split each lose accuracy. Holding cities out of training changes plan-cost error by +0.003 [−0.018, +0.020], within a pre-set margin. On 24 new controlled response laws, with the plan, the optimum and validity all judged within 20% of the budget, the response correction with a one-week pilot lowers regret by 51.2% relative to a trial with the same nominal randomized rider-weeks and by 53.6% in the stricter pre-specified 10% band, and in the exact-summation analysis of these week-long requests a four-week pilot, with 1/4.5 of the randomized rider-weeks, comes within +0.007 [−0.012, +0.027] of an 18-week trial. In the registered primary conditions of two studies where windows, populations, rules and binding budgets change from request to request, the framework's regret is below that of a trial with the same nominal rider-weeks and below that of dose interpolation of the same pilot data at every pilot length, and reusing its one-off preparation answers 60 requests 14.1× and 2.70× faster than re-running the pipeline for each, with identical answers. When the paid window or a selected subpopulation changes behaviour, a window-aware correction lowers one-week regret by 0.195 against direct reuse. Against a nine-offer trial fitted with the framework's own dose curve, one-week regret is 0.055 lower; 73.5% of the gain over per-offer fits of that trial comes from sharing the tilt across offers. Public retail data and three payment rules confirm the transfer with identical exact outputs.

Primary area:
optimization
Keywords:
capacity planning, threshold incentives, probabilistic forecasting, budget allocation

Reviews

Among:
Shape:
1234this paper, Bottom 10%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.67 over 3 reviews, above 10% of the others.
Reviewer_MexkOpenReview ↗
Rating
2
Confidence
1
Informativeness
3
Presentation
2
Soundness
2
Read review
Paper summary

This paper proposes a request-driven framework for budgeted incentive allocation on delivery platforms. It combines per-rider-day forecasting, activity layering, entropy-based response correction, trajectory integration, and multiple-choice knapsack allocation. The system is well engineered, and the empirical study is extensive.

Critical strengths
  1. The topic studied is interesting and practical.

  2. The overall framework appears well structured.

Critical weaknesses

As an AI researcher, I am not familiar with the core method of this paper. From the perspective of AI, the innovation is very weak, with no significant module proposed. All AI techniques used are drawn from mature methods.

Questions for authors

This paper is very interesting. However, what is the relationship between this paper and ICLR? I think this paper should be submitted to transportation or management journals.

Reviewer_MS5eOpenReview ↗
Rating
2
Confidence
3
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

This paper proposes a request-driven framework for budgeted threshold incentive allocation on delivery platforms. The framework integrates four stages: distributional prediction, population reduction, trajectory integration, and budget allocation. A key component is a history-anchored response correction that uses a short randomized pilot to estimate incentive-induced behavioral changes through exponential tilting. The paper also develops a four-term error decomposition and evaluates the framework on historical rider records, controlled response environments, and public retail data.

Critical strengths
  1. The paper considers a problem involving limited experimentation, complex incentive rules, and changing planning requests. The proposed modular framework is intuitive, and reusing conditional trajectory distributions across requests is a useful engineering idea.

  2. The paper provides a large number of controlled experiments, including multiple response generators, ablation studies, matched-information comparisons, and robustness analyses. The improvements over several baselines are encouraging, and the computational savings from reusing intermediate outputs are clearly demonstrated.

Critical weaknesses
  1. This paper claims that short randomized pilots, combined with historical data, can provide accurate incentive-response estimates and substantially reduce the need for long experiments. However, the experiments supporting this claim rely entirely on synthetic response laws. Moreover, the proposed response correction relies on a low-dimensional exponential tilt, which may not capture more complicated behavioral changes. The extent to which these controlled environments represent real incentive responses remains unclear. This limitation is particularly important because accurate behavioral response estimation is central to the proposed allocation framework.

  2. The framework combines several established techniques. While the integration is reasonable, the paper could better clarify its methodological contribution beyond combining these existing approaches.

  3. From Table 10, even with a four-week pilot, the bound improves upon the trivial benchmark for only 5.8% of feasible requests. This limits its practical informativeness.

  4. The pre-specified near-lossless comparison with the long RCT does not meet its criterion, whereas the main text emphasizes a near-lossless result under a different integration and budget-band setting. The distinction should be made clearer.

Questions for authors
  1. Can the authors clarify the main methodological novelty relative to existing approaches?

  2. Given that the behavioral response experiments are synthetic, what evidence supports the generalizability of the proposed correction to real responses?

  3. Could the authors clarify the practical role of Proposition 1, given that its stage errors are unknown at planning time and the resulting bound is informative for only a small fraction of evaluated requests?

  4. Could the authors distinguish the pre-specified near-lossless results from the exact-integration analysis highlighted in the main text?

Reviewer_PYFDOpenReview ↗
Rating
1
Confidence
2
Informativeness
1
Presentation
1
Soundness
3
Read review
Paper summary

The paper studies a request-driven framework for on-demand delivery platforms. They predict demand and trajectories and allocate budget to provide incentives to riders. They compare with MIP baselines and show that latency is much lower while being comparable in performance.

Critical strengths

The paper studies a request-driven framework for on-demand delivery platforms which is a relevant problem nowadays with rampant use of such platforms.

Critical weaknesses

The problem domain is very niche and the solution proposed seems overfitted to it. The writing is very difficult to follow and might be AI generated. The techniques proposed do not seem to have a lot of originality or generalizability across domains. The baseline seems to be an MIP solved to optimality and no other heuristics or algorithms. The dataset used is not clearly presented.

Questions for authors

Could you present more information about the dataset used for evaluation? Have you considered any other intuitive baselines or algorithms for comparison?

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.