~/index personal site, 2026
Pavel (Pasha) Kireyev.
Machine intelligence at the edge of markets.
I worked on the first wave of machine learning entreprise adoption (in Tokyo / Boston) and channel my practical experience into research, development, and education on digital platforms, AI strategy, data and decentralized markets. I've managed talent-dense teams of experts in ML (computer vision, recommenders, pricing), data engineering, and microeconometrics on ambitious projects both in industry (entreprise / frontier startups) and academia (top business schools). I'm proud to have worked with great founders who went on to create companies like Robust Intelligence (acquired by Cisco for $400m), Taktile, and Compute Exchange.
I hold a doctorate from Harvard Business School, an MA Statistics from Yale, and BSc from LSE. I taught AI Strategy in MBA, ExecEd and Causal Machine Learning in the PhD program at INSEAD (France / Singapore), Product Analytics at LSE (UK), started the Asia office at QuantCo out of Harvard, and helped start Onaji in Tokyo. I've produced some of the earliest cases on frontier AI labs (PFN in Japan in 2019) and written about new decentralized technologies in the Harvard Business Review.
Behavioral evaluation of AI systems
-
01
writeup ↗ source ↗
PriceBench
Behavioral benchmark for LLM booking agents. Probes price sensitivity, quality preferences, and brand bias across language models on realistic hotel booking tasks, then recovers each model's purchasing personality via conditional logit analysis.
-
02
site ↗ source ↗ writeup ↗
AttributionBench
Marketing-decision simulation that pits humans against AI models on real attribution tasks, testing whether language models can correctly identify what actually drives revenue.
LLM behavior · Digital platforms · Marketplace design
-
01
paper ↗
User Preferences for Large Language Models: Implications for AI Safety and Market Structure
How user choice and refusal sensitivity across proprietary and open-source LLMs shape safety incentives and market structure.
-
02
paper ↗ substack ↗
Large Language Models Augment or Substitute Human Experts in Idea Screening
Combines LLMs with expert ratings to cut idea-screening effort, examining when models augment versus replace human evaluators.
-
03
substack ↗
Balancing Safety and Usage in Conversational AI Systems
Models how refusals and multi-turn dynamics shape engagement, finding stricter moderation barely reduces overall LLM usage.
-
04
Long Descending Auctions: Buyer Dynamics and Pricing in Digital Marketplaces
Structural model of buyer waiting and pricing in long-running descending auctions used in digital collectible marketplaces.
-
05
paper ↗
Artificially Intelligent Marketplaces
Back-tests algorithmic trading strategies on peer-to-peer marketplaces and shows low-bid spamming can earn meaningful returns.
-
06
paper ↗
Should Firms Promote Public Health to their Customers? Evidence from a Natural Experiment in Ridesharing
Field evidence that a ridesharing platform's COVID-19 vaccination nudge raised both vaccinations and trip revenue.
-
07
paper ↗ substack ↗
Denied by an (Unexplainable) Algorithm: Teleological Explanations Enhance Customer Satisfaction
Field and lab experiments show goal-based explanations for algorithmic denials lift satisfaction even without revealing mechanisms.
-
08
Management Science paper ↗ substack ↗
Don't Fake It If You Can't Make It: Driver Misconduct in Last Mile Delivery
Uses last-mile delivery data to quantify fake-remarked deliveries and their spillover drag on next-day driver productivity.
-
09
RAND J. of Economics paper ↗ substack ↗
Markets for Ideas: Prize Structure, Entry Limits, and the Design of Ideation Contests
Empirical contest model showing more prizes plus per-participant submission caps improve quality outcomes in ideation platforms.
-
10
Marketing Science paper ↗
Match Your Own Price? Self-Matching as a Retailer's Multichannel Pricing Strategy
Game-theoretic analysis of when retailers should match their own online and in-store prices across channels.
-
11
Int. J. Research in Marketing paper ↗
Do Display Ads Influence Search? Attribution and Dynamics in Online Advertising
Time-series model showing display ads lift search conversions, with substantial implications for ad attribution and budget allocation.
Frontier technology adoption and experimentation
-
01
INSEAD popular case ↗ substack ↗
Preferred Networks: A Deep Learning Startup Powers the Internet of Things
Traces a Japanese deep-learning startup partnering with industrial giants to apply AI to autonomous driving, factories, and healthcare.
-
02
INSEAD case ↗
Sorare: Bringing Fantasy Sports into the Metaverse with Blockchain
Follows a Paris startup building blockchain-based fantasy football through league partnerships amid volatile crypto markets.
-
03
INSEAD popular case ↗
ObEN PAI: Building a World of Personal AI Avatars
Follows a Los Angeles startup building consumer and celebrity AI avatars across entertainment and healthcare use cases.
-
04
HBS case ↗
Zalora: Data-Driven Pricing
How a Southeast Asian fashion retailer used customer and pricing data to guide markdown and pricing decisions across its catalog.
MBA · ExecEd · MiM · PhD
-
01
INSEAD · MBA, ExecEd, MiM substack ↗
Artificial Intelligence Strategy
Strategic frameworks for AI adoption: build, buy, or partner. Covers data, talent, and competitive positioning across industries.
-
02
INSEAD · PhD substack ↗
Machine Learning, Causality, and Management
Supervised, unsupervised, and causal ML methods for empirical management research at the doctoral level.
-
03
LSE · MiM
Analytics II
Advanced regression, discrete choice modeling, and causal inference via A/B testing for product, brand, and innovation decisions.
-
04
INSEAD · MBA
Future of Digital
Emerging digital frontiers (AI, platforms, decentralized technologies) and the strategic shifts they trigger.
~/elsewhere continue here