When AI Gets Cheaper, Do Prices Follow? Information Delays and Algorithmic Pricing

Krish Tadigotla

AI summary · arXiv: 2610.31693v1 · econ.GN

When AI Gets Cheaper, Do Prices Follow? Information Delays and Algorithmic Pricing

Krish Tadigotla

Original paper Submitted to arXiv on 20 Oct. 2026 12 pages, 10 figures, 2 tables License: NONEXCLUSIVE-DISTRIB 1.0

How to read this summary? Illustrated guide

The essentials

In a simulated two-seller market, lower AI operating costs do not necessarily translate into full price reductions for consumers.

Based on the abstract only: check the details in the original paper.

Question

The paper asks whether lower costs of running AI models reduce prices, and whether delays in observing rivals’ prices or limits on repricing affect consumer benefit. It also examines whether these effects vary with differences in firm quality and costs.

Method

The study simulates two differentiated sellers with logit demand and tabular Q learning. A factorial experiment varies rival-price information age, repricing frequency, cost reductions, and firm differences; it evaluates learned policies on 60 paired market seeds in each of 16 conditions, for 960 training runs, measuring expected consumer surplus after learning.

Findings

Result In the main setting with identical firms and falling costs, the estimated interaction between information delay and repricing frequency on consumer surplus was -0.00025 normalized units, with a 95% paired seed bootstrap interval spanning zero.
See the passage in the paper
In the main setting with identical firms and falling costs, the estimated interaction between information delay and repricing frequency on consumer surplus is -0.00025 normalized units, with a 95% paired seed bootstrap interval of [-0.0402, 0.0393].
Result The intervals for whether the interaction changes with the cost regime or firm differences also include zero.
See the passage in the paper
The interval includes zero, as do intervals for tests of whether the interaction changes with the cost regime or differences between firms.
Result With the shortest information delay and the ability to revise prices every period, the two firms’ average price reductions equal 55.6% and 44.0% of the imposed cost reduction.
See the passage in the paper
With the shortest information delay and the ability to revise prices every period, average price reductions equal 55.6% and 44.0% of the imposed cost reduction for the two firms.
Result Learned-policy consumer surplus is below the corresponding static grid Nash benchmark in every experimental condition.
See the passage in the paper
Mean consumer surplus under the learned policies is below the corresponding static grid Nash benchmark in all 16 conditions.
Authors’ interpretation The authors conclude that simulated cost savings need not be fully passed on to buyers, while cautioning that the study does not establish collusion or characterize actual AI providers.
See the passage in the paper
The findings show that simulated cost savings need not be fully passed on to buyers. They do not establish collusion or describe pricing by actual AI providers.

Limitations

The results come from a simulated market and do not establish how actual AI providers set prices.
See the passage in the paper
They do not establish collusion or describe pricing by actual AI providers.
The study’s scope is limited to two differentiated sellers with logit demand and tabular Q learning.
See the passage in the paper
We study a simulated market with two differentiated sellers, logit demand, and tabular Q learning.

Why it matters

The authors argue that simulated reductions in AI operating costs need not be fully passed on to consumers. Their findings do not establish collusion or describe pricing by actual AI providers.

Written by AI from the paper. Passages are checked word for word against the text we analysed; the rest may contain errors.

AI-generated keywords: algorithmic pricing information delays cost pass-through consumer surplus reinforcement learning

Read further

Comprehensive Summary
Layman's Summary
Blog article
Original abstract

Krish Tadigotla’s paper, “When AI Gets Cheaper, Do Prices Follow? Information Delays and Algorithmic Pricing,” asks whether lower AI operating costs lead sellers to lower prices, and whether the effect depends on how quickly they learn competitors’ prices and how often they can revise their own. The question matters because cost savings do not automatically benefit consumers, particularly when sellers use algorithms to set prices. The available material includes only the title, author information, and abstract. This summary is therefore limited to the methods and findings reported in the abstract. The study simulates a market with two differentiated sellers, logit demand, and tabular Q-learning. Its factorial experiment varies four features: the age of the information sellers have about a rival’s price, how frequently they can revise their own prices, whether production costs fall, and whether the firms differ in quality or cost. The researchers evaluate learned pricing policies using 60 paired market seeds in each of 16 experimental conditions, for 960 training runs in total. Their main outcome is expected consumer surplus after learning ends. In the main setting—where firms are identical and production costs fall—the estimated interaction between the delay in rival-price information and repricing frequency on consumer surplus is −0.00025 normalized units. The 95% paired-seed bootstrap interval is [−0.0402, 0.0393], which includes zero. This estimate therefore provides no clear evidence that the combination of information delays and repricing frequency affects consumer surplus in that setting. The abstract also reports that interaction-test intervals include zero when the researchers assess whether this relationship varies by cost regime or by differences between firms. The simulations nevertheless indicate that cost reductions are not always passed through fully to consumers. With the shortest information delay and the ability to revise prices every period, average price reductions equal 55.6% and 44.0% of the imposed cost reduction for the two sellers, respectively. These figures suggest incomplete pass-through even under relatively favorable conditions for monitoring and adjusting prices. Mean consumer surplus under the learned policies is also below the corresponding static grid Nash benchmark in all 16 conditions. The abstract does not explain how the benchmark is constructed or report the size of these gaps. The authors conclude that, in this simulated setting, lower costs do not guarantee proportionate price reductions or consumer gains. They emphasize that the findings do not establish collusion and should not be interpreted as a description of how actual AI providers set prices. The study’s scope and the limited available evidence call for caution. Its results come from a model with two differentiated sellers, logit demand, and a particular tabular learning method—not from observed markets or real pricing systems. The abstract also lacks enough detail to assess the model’s assumptions, robustness checks, or the economic significance of the consumer-surplus differences. The findings are best understood as results from a specific simulation, not as a general empirical claim about AI-related costs or real-world pricing.

When AI Gets Cheaper, Do Prices Follow?

If artificial intelligence becomes cheaper to run, will businesses pass those savings on to customers? Not necessarily. Sellers may use algorithms to set prices, and the effect of lower costs could depend on how quickly they learn what competitors charge and how often they can change their own prices. Krish Tadigotla’s paper, “When AI Gets Cheaper, Do Prices Follow? Information Delays and Algorithmic Pricing,” examines these questions through a simulated market.

The distinction matters: a business’s reduced costs do not automatically translate into lower prices or higher consumer welfare. But the paper’s findings should be read with care. The available information is limited to the title, author details, and abstract, so this article describes only the methods and results reported there.

How the simulation works

The study models a market with two sellers offering differentiated products. Buyers’ preferences are represented through logit demand, and sellers learn pricing policies using tabular Q-learning—a method in which agents adjust decisions based on outcomes over repeated interactions.

The researchers vary four features in a factorial experiment:

How old the information is that a seller has about its rival’s price.
How often each seller can revise its own price.
Whether production costs fall.
Whether the sellers differ in quality or cost.

They evaluate learned policies using 60 paired market seeds in each of 16 experimental conditions, for 960 training runs altogether. The main outcome is expected consumer surplus after learning ends. In plain terms, the researchers ask how much benefit consumers receive under the prices the sellers have learned to set.

What happened to prices and consumers?

No clear interaction effect in the main setting

For the central case—identical sellers facing a reduction in production costs—the estimated interaction between delays in rival-price information and repricing frequency was −0.00025 normalized units of consumer surplus. The 95% paired-seed bootstrap interval ran from −0.0402 to 0.0393, including zero.

That interval means the estimate does not provide clear evidence that the combination of information delays and repricing frequency changes consumer surplus in this setting. The abstract also reports that interaction-test intervals include zero when the researchers examine whether the relationship differs across cost regimes or between firms with different characteristics. These results do not establish that information and repricing never matter; they indicate that the reported tests do not clearly identify an interaction effect.

Cost savings were not fully passed through

The simulations do show that lower costs do not always lead to equivalent price reductions. Even with the shortest delay in rival-price information and the ability to revise prices every period, the sellers’ average price reductions amounted to 55.6% and 44.0% of the imposed cost reduction, respectively.

Those figures point to incomplete pass-through in the model, even under relatively favorable conditions for monitoring and adjusting prices. They do not tell us exactly why prices fell by less than costs, but they illustrate the paper’s central caution: a cost reduction does not guarantee a proportionate price cut.

Another reported result is that mean consumer surplus under the learned policies falls below the corresponding static grid Nash benchmark in all 16 conditions. The abstract does not explain how that benchmark is constructed or provide the size of the gaps. The comparison is therefore worth noting, but the information available is not enough to assess how large or economically important the differences are.

What the findings do—and do not—show

Tadigotla’s conclusion is deliberately limited: in this simulated setting, lower costs do not guarantee proportionate price reductions or consumer gains. The study does not establish that sellers are colluding, and its results should not be treated as a description of how actual AI providers set prices.

That caution is important because the analysis relies on a specific model: two differentiated sellers, logit demand, and a particular tabular learning method. It uses simulations rather than observed market prices or real-world pricing systems. The abstract also does not provide enough detail to evaluate the model’s assumptions, robustness checks, or the practical significance of the consumer-surplus results.

The takeaway

The paper offers a focused test of whether cheaper operations necessarily benefit consumers when algorithmic pricing is involved. In its simulations, the answer is no: sellers do not always pass cost savings through fully, and the reported evidence does not clearly show that information delays combined with repricing frequency determine consumer surplus.

These results are best understood as findings from a particular simulation—not as a general empirical claim about AI costs or real-world prices. They raise a useful question for further research: under what market conditions do businesses pass technology-related savings on to customers, and when do those savings remain with sellers?

Created on 31 Oct. 2026

Was this synthesis useful?

Score: 0

Ask questions about this paper to our AI assistant

How to chat with this paper? Illustrated guide

Compare with the PDF

You can also chat with multiple papers at once here.

⚠The license of the paper does not allow us to build upon its content and the AI assistant only knows about the paper metadata rather than the full article.

AI assistant instructions?

Tools for this paper

Explain it differently 3 free rewrites left today

Save to a collection Group papers in collections to compare them, chat with them and export them. Create a free account Log in

Export BibTeX · RIS (Zotero, Mendeley, EndNote) · Markdown (Notion, Obsidian)

Advanced tools Premium

For researchers who follow a field closely: citation alerts, journal-club slides, version comparison and Zotero export. Discover Premium

Zotero

Citation alerts Citation alerts guide

Who works on it? The people and institutions who publish on its topic

Weekly letter Presented in "Economics and finance this week" of 26 October 2026. Economics and finance this week: When better rules change the game

Journal club Premium+ Slides to present it

Versions What changed between versions

Talk about it in public Premium+ Press kit: release, questions, hype check Press kit guide

Disclaimer: The AI-based summarization tool and virtual assistant provided on this website may not always provide accurate and complete summaries or responses. We encourage you to carefully review and evaluate the generated content to ensure its quality and relevance to your needs.

When AI Gets Cheaper, Do Prices Follow? Information Delays and Algorithmic Pricing

The essentials

Question

Method

Findings

Limitations

Why it matters

Read further

Definitions

When AI Gets Cheaper, Do Prices Follow?

How the simulation works

What happened to prices and consumers?

No clear interaction effect in the main setting

Cost savings were not fully passed through

What the findings do—and do not—show

The takeaway

Ask questions about this paper to our AI assistant

Tools for this paper

Advanced tools Premium

Papers around this one

Closest papers

Prior work it builds on

Go further

Cite this paper (BibTeX)