SSSimrita Singh

Simrita Singh

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Simrita Singh has a total of 7 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Feature-Driven Priority Queuing

    by Simrita Singh, Itai Gurvich, Jan A. Van Mieghem

    Rethinking Artificial Intelligence-Assisted Priority Queuing: The Cost of Locking in Classifiers When job types must be inferred from observable features, should artificial intelligence (AI) classifiers be locked before optimizing queue assignments, or should the full system be trained end to end? Singh, Gurvich, and Van Mieghem show that the modular “type-first” approach—locking a type classifier and then, optimizing queue assignment based only on its output distribution—can be fundamentally suboptimal. The authors establish that type-first achieves optimality only when the locked classifier recovers the Bayes posterior almost everywhere, a condition rarely satisfied in high-dimensional or misspecified settings. In contrast, their “direct” approach jointly optimizes feature-to-queue mappings to minimize empirical waiting costs and converges to the theoretical optimum. Using 100,000 chest X-rays from the National Institutes of Health, they show that direct optimization substantially improves waiting-cost performance and reveal mechanisms, such as statistical pooling and strategic overprioritization of medium-cost types. The work speaks to the growing debate over whether AI systems should be locked or adaptable.


  • Open Access

    Managing the Human Fallback: Skill Investment Under Improving AI and Worker Mobility

    by Simrita Singh, Naireet Ghosh, Tinglong Dai

    When firms deploy autonomous AI, they must decide how much work to leave to the system and how much to keep workers engaged. This decision affects current output and future human capital. We develop a parsimonious two-period model in which AI may outperform the worker when it functions, but may fail with positive probability. A firm chooses worker engagement; engagement lowers current output for below-benchmark workers, but changes future skill through learning and erosion. We distinguish two dimensions of AI progress: capability, the system's output when it works, and reliability, the probability that it works. In a single-firm benchmark, engagement is valuable only as fallback investment. The firm engages the least-skilled workers most, because they have the largest skill gaps and are least costly to bring toward a useful fallback level. With worker mobility, engagement also affects labor-market sorting: workers prefer jobs that build more valuable skill trajectories. This sorting motive targets higher-skill workers near the AI frontier, where skill gains are more valuable and engagement is less costly. Mobility can therefore reverse the engagement pattern, shifting investment from the least-skilled toward the most-skilled workers below the AI benchmark. Mobility also reshapes how AI progress affects engagement: greater capability raises engagement by increasing the value of the skill trajectory a firm offers, whereas greater reliability can raise or lower it because it reduces fallback need while also changing learning opportunities. Under worker mobility, human–AI work design becomes a problem of human-capital investment, in which allocating work today shapes future skill.

  • Open Access

    Algorithmic Noise in Retail Search: How Irrelevant Results Bias Demand Estimation and Inventory Decisions

    by Simrita Singh, Saravanan Kesavan

    Lost sales are a fundamental challenge in retail, particularly in settings with slow-moving inventory where retailers cannot stock all items locally and must fulfill some demand from distribution centers or nearby stores. Estimating how sensitive customers are to the resulting wait times is critical for inventory planning. Many retailers now use algorithm-mediated search systems-online search engines and in-store digital catalogs-to help customers find products, and the lookup records these systems generate offer a natural data source for demand estimation. However, these systems can retrieve items irrelevant to the customer's purchase intent, particularly in settings with keyword-based retrieval over long-tail catalogs with overlapping terminology. We term this contamination algorithmic noise and show that it biases demand sensitivity estimates. Using a stylized econometric framework, we demonstrate that algorithmic noise produces omitted variable bias whose direction depends on the correlation between noise and the attribute (e.g., price or wait time) of interest, generating two distinct distortions: phantom sensitivity, where the estimated effect of an attribute on demand is accentuated or entirely spurious, and masked sensitivity, where a genuine effect is attenuated toward zero. We classify each item in a lookup as either relevant or noise using conversion rates and co-purchase patterns from a temporally separated calibration period, and apply this classification to catalog search data for a popular vehicle model from three brick-and-mortar auto parts stores. We find that 80 of 218 parts exhibit phantom wait-time sensitivity-spurious effects driven by algorithmic noiseand that this is the dominant source of estimation bias; an additional 16 parts reveal significant sensitivity only after denoising. The denoised model achieves better predictive fit, and an inventory optimization model shows that stocking decisions based on denoised estimates improve profit by 3-5% while reducing holding costs by 9-35%, primarily by eliminating unnecessary inventory for parts with phantom wait-time sensitivity. These results replicate across a different vehicle segment and city. As search algorithms increasingly mediate retail transactions, algorithmic noise represents a pervasive source of bias in demand estimation, with direct consequences for inventory decisions.

  • Open Access

    Feature-Driven Priority Queuing

    by Simrita Singh, Jan Albert Van Mieghem, Itai Gurvich

    Traditional queuing theory assumes that customer types are perfectly observed, assigning each customer to a type-specific priority queue-this is known as type-driven priority queuing. We study feature-driven priority queuing, where types are not perfectly observed but inferred from features using a classifier. Our work introduces an explicit modeling of the classifier within the queuing system design, which is novel in this context. We analyze the optimal queue classification that minimizes queuing delay costs and compare a plug-and-play, type-first approach-where type probabilities are predicted by a classifier and then used to optimize queuing-with a direct approach that optimizes the classifier to predict queue probabilities directly from features. Our primary contribution is proving that the direct approach outperforms the type-first approach, except when the type classifier perfectly predicts the Bayes posterior. To illustrate how the direct approach achieves this improvement, we use a combination of analytical results, numerical examples, and simulations on a large real-world dataset. In our numerical study using state-of-the-art image classifiers on a dataset of medical images used in radiology triage, the direct approach reduces delay costs by 30-60% relative to the type-first approach, particularly in high-utilization scenarios.