Conference Presentation
Dr. Uri Yerushalmi, Co-Founder&CAIO, Fetcherr: How AI Driven Decision Maximizes Business Performance
- Speaker and Background: Uri Arushalmi, Chief AI Officer and co-founder of Fetcher, with over 30 years of experience in AI decision-making, a PhD in Computational Neuroscience, and prior tenure as Head of AI at a Wall Street algorithmic trading firm.
- Core Thesis: Existing Generative AI (Large Language Models and visual models) has not yet delivered significant autonomous business growth or operational optimization due to a lack of quantitative, market-specific understanding.
- Proposed Solution: Fetcher has spent the last six years developing "Large Market Models," which apply deep learning principles to market data rather than text or video, to output anticipated market dynamics.
- Analogy to Video Models: Unlike video models that simulate the probabilistic behavior of visual scenes (e.g., falling leaves), Large Market Models simulate the complex, interconnected, and probabilistic behavior of live, moving markets.
- Data Modality Differences:
- Video models output 4-dimensional data (height, width, time, color).
- Large Market Models output hyper-dimensional data encompassing multiple time frames (transaction time, execution time), product features, provider/buyer identifiers, and probability-based metrics like demand elasticity and competitor pricing positioning.
- Limitation of Current AI: Large Language Models are qualitative and text-based, whereas business decisions require quantitative predictions; current generative AI focuses on replacing or augmenting humans rather than autonomously driving revenue.
- Operational Shift: Large Market Models are designed to be inherently quantitative and business-oriented, enabling systems to anticipate future dynamics rather than merely reacting to lagging signals.
- Market Dynamics Impact:
- Manual or rule-based systems show rare, correlatively linked price changes.
- AI-driven systems produce immediate, gradual, and pronounced price reactions to every market "hint," acting proactively rather than reactively.
- System Architecture:
- Data Consolidation: Integrates internal business data (transactions, orders, prices) with external data (capital markets, competitor behavior).
- Simulation Engine: Runs computationally intensive simulations of decision-making policies to evaluate outcomes across timeframes (daily/hourly) regarding demand, competitor reactions, and inventory.
- Policy Selection: Selects the specific policy expected to maximize a defined business reward (revenue, market share, or profitability).
- Visualization and Analysis:
- High-dimensional model outputs are sliced into 2D representations for analysis, plotting a product's price against a competitor's price.
- Color gradients represent expected demand, where blue indicates low demand (high price relative to competitors) and red indicates high demand.
- The model identifies equilibrium lines and varying market impacts (e.g., low competitor influence vs. highly competitive environments where every cent matters).
- Empirical Results: A/B testing across various customer scenarios consistently demonstrates a statistically significant revenue uplift of more than 10% when using the AI decision-making system compared to control groups.
- Secondary Benefits: The deployment of these systems results in reduced human resource requirements alongside revenue growth.
- Competitive Implications: Success in modern markets is no longer defined by traditional attributes (size, speed, employee intellect, or IP) but by access to the underlying market structure via AI decision-making tools.
- Strategic Advantage: Entities with access to Large Market Models can shape market outcomes, gaining the ability to move before the market shifts rather than following lagging indicators.