用生成式人工智能改变商品交易

© Oliver WymanCommodity trading firms have come through some turbulent times.As we pointed out in an earlier analysis, the industry is rebalancing fromsevere disruption that occurred in 2022. And as the industry settlesinto a new normal, we are seeing leading commodity traders embracegenerative artificial intelligence (AI) to expand their competitive edge.Firms are finding generative AI to be an invaluable tool in tasks rangingfrom data pre-processing to augmenting long-term strategic planning andshort-term portfolio optimization. In parallel, the cost and time to deploygenerative AI models has dramatically decreased since 2022, loweringthe barriers to experimentation and impact. Just as with the leap fromlandlines to smartphones, generative AI has the potential to revolutionizethe commodity trading industry, potentially leaving behind firms stilltethered to more traditionalprocesses. © Oliver WymanGENERATIVE AI MOMENTUM IN COMMODITY TRADINGGenerative AI is the latest chapter of a decades-long evolution in AI and machine learning(ML) techniques used by the commodity trading industry as early as the 1990s. The practiceevolved in specialist firms and commodity trading hedge funds, and has gradually gainedtraction among the large independent traders in recent years. Traders use predictiveanalytics to create different scenarios in supply and demand to discover trading factorsby analyzing large market datasets. Additionally, traders can use supercomputers withAI algorithms to predict weather patterns to determine solar and wind energy supply andlocal energy demand, or to process satellite images of loading docks to understand oiltanker logistics. Such cutting-edge market intelligence can then be monetized, puttingthe trader ahead ofcompetitors.Several commodity trading firms are already using AI for simple efficiency and productivitygains. From 2018 to 2023, total IT costs and investments across all organizations increasedby 47%, according to Oliver Wyman proprietary data on leading commodity traders.These investments are primarily going into firms’ data management platforms, cloudcomputing, and talent budgets — the key ingredients to build a generative AI organization.IT headcount increased 15%, and total IT spending per full-time IT role rose 28% duringthat same timeframe.Exhibit 1: IT investment levels in commodity trading organizations47%Increase in IT investments28%Increase in IT spendper full-time IT role15%Increase in IT full-timeemployment headcountSource: Oliver Wyman proprietary data and analysisBARRIERS TO AI EXPERIMENTATION HAVE DECREASEDAI is more accessible to industry players than ever before. The cost of deploying AI modelshas fallen by 60-fold since 2020, mainly driven by better chipsets, algorithms, and energyefficiency. Similarly, the time from idea to deployment of AI models has decreased from12 months to 12 weeks, largely due to the proliferation of ready-made AI developmenttools. © Oliver WymanTime to deploy a generative AI modelIn months2.52022202430.50.51.5Cost to train a GPT-3 modelIn US$450,00075,000202020222024Design and planningModel training developmentSource: Unite.AI, ARK Invest, American Enterprise Institute, and Oliver Wyman analysisNot all trading operations are ready for this level of AI adoption. In fact, our view is thatcommodity trading firms will diverge into two camps: a camp of AI visionaries, which can puttheir existing talent, capabilities, and infrastructure to work building a lasting competitiveadvantage; and a camp of AI challengers, which find themselves squeezed between theAI visionaries and a new generation of challenger firms that are native to generativeAI.AI VISIONARIES WILL CHART THE FUTUREFirms that develop leadership capabilities, organizational structures, and human skillsetsto support incorporating AI into their operations are in the best position to harness thepotential of generative AI. They added the infrastructure and processes to collect and curateproprietary data. For example, electricity traders use algorithm tools to spot and executetrades, with little human intervention. Regional utilities use AI for short-range local weatherforecasting to predict gas and power demand. Some commodity traders took a page fromthe logistics arms of tech giants such as Amazon and Alibaba to explore using generative AIin pre-processing massive amounts of complexdata. © Oliver WymanThe incredible performance of commodity trading firms in 2022 and 2023 prompted muchof this activity. As shared in Oliver Wyman’s annual commodity trading report, the industryearned record margins in 2022 of $150 billion due to intense volatility triggered by the war inUkraine, then margins declined last year to $100 billion, still the second-highest year for theindustry. Hedge funds, which traditionally focus on big data and quantitative analysis, wereattracted to the sector. Energy traders subsequently took notice. While traders traditionallyrely on asset networks and existing relationsh