Market & Match
Market & Match #7: AI on Courts and Banks
In today’s edition of Market & Match, AI asserts itself from the tennis court to banks, from financial markets to the workplace, with the same conclusion: value is created mainly when it is anchored in concrete uses, reinvented workflows, and solid governance.
- Infosys and Alcaraz bet on on-court AI
- PwC shows AI value is highly concentrated
- Oracle deploys AI agents in banking
- Vision-Language Models still struggle in sideways markets
- Generative AI boosts jobs, wages and productivity
1. Infosys teams up with Alcaraz for tennis AI
Infosys partners with Carlos Alcaraz in a global partnership that will make AI a tool for preparation, analysis and strategy at the heart of high-level tennis.
Key takeaway: Infosys announced a multi-year global partnership with Carlos Alcaraz, named global brand ambassador, to develop with Infosys Topaz AI-powered match analysis tools and a personalized performance app for match preparation and in-match strategy. The initiative continues more than ten years of Infosys investments in tennis-focused technology and also includes social impact projects with the Carlos Alcaraz Foundation.
In practice: Concretely, the partnership goes beyond a branding deal: Infosys will work with Alcaraz and his team to build match-analysis tools and a personalized app designed to improve preparation and tactical adjustments during matches. Carlos Alcaraz has said he wants to rely on data and AI to better understand his game and raise his performance. For Infosys, this visible use of Topaz demonstrates how its generative and agentic AI capabilities can be deployed in an environment of extreme demands. The agreement also extends to joint “tech-for-good” initiatives via the Carlos Alcaraz Foundation.
Analysis: In market terms, this announcement fits into a broader Infosys strategy to align its “AI-first” corporate positioning with concrete, high-visibility use cases. The Economic Times notes that the group has already built, over the years, digital products for organizations like the ATP, the Australian Open and Roland-Garros, showing that the Alcaraz collaboration is not a one-off but an extension of a lasting presence in data, performance and tennis fan engagement. Moneycontrol adds that Infosys presents this partnership as a demonstration of AI applied to elite sport, in line with its broader strategy. More broadly, sports partnerships are becoming a testing ground for enterprise AI vendors seeking differentiating uses beyond back-office automation.
The stakes: The potential winners are first Infosys, strengthening the commercial credibility of Topaz through a premium use case, and Carlos Alcaraz, gaining privileged access to bespoke preparation and analysis tools. Professional tennis could also gain if these innovations feed Infosys’s tennis platform and extend advanced insights to players, coaches and fans. Potential losers are technology vendors less embedded in the tennis ecosystem, as collaborations with elite athletes give Infosys access to proprietary data and a hard-to-reproduce credibility. The central stakes are thus twofold: turning a sports partnership into a credible enterprise AI showcase, while proving that these tools deliver real on-field and overall sports experience advantages.
Verdict: Infosys’s partnership with Carlos Alcaraz is a smart strategic move, because it turns AI from a corporate slogan into a tangible competitive advantage in one of the world’s most demanding sports environments. But its true value will hinge on evidence: if these tools truly improve decision-making and performance without reducing tennis to a mere algorithmic war, Infosys will have found more than an ambassador—a credible demonstration of what applied AI can change in the economics of sport.
Sources :
2. PwC: 20% of companies capture AI value
According to PwC, a handful of companies are already turning AI into a growth engine and automated decision-making, while the majority remain stuck in experiments with no real payoff.
Key takeaway: PwC’s 2026 study concludes that a minority of companies capture the bulk of AI’s economic value: 20% of organizations account for 74% of gains, while the majority remains stuck at the pilot stage. According to PwC, leaders do not just add AI tools: they use them to drive growth, rethink their business model, and automate more decisions with governance guardrails.
In practice: In practice, the study shows that firms achieving the best financial returns use AI as a engine for reinvention rather than as a simple productivity lever. They are two to three times more likely to leverage AI to identify new growth opportunities and almost twice as likely to redesign workflows around AI. PwC also notes that they significantly increase the number of decisions made without human intervention, while being more advanced on Responsible AI frameworks and cross-functional governance bodies. The Irish Examiner echoes this reading, noting that only a minority truly transforms AI activity into measurable financial returns.
Analysis: The market context suggests that this concentration of gains is not accidental, but tied to structural advantages already present in the economy. The Wall Street Journal reports that companies seeing meaningful returns often own proprietary data, have a large installed software base, strong integration capabilities, and the power to redraw processes rather than run isolated experiments. This helps explain why access to models alone is not enough: data preparation, governance and employee adoption remain bottlenecks. In this frame, PwC’s results fit into a broader dynamic where AI tends to amplify scale advantages and reinforce a winner-take-most market logic. PwC adds that the ability to capture growth opportunities from sector convergence is the most determining factor of AI-related financial performance, ahead of mere efficiency gains.
The stakes : The concrete stakes are a widening gap between firms that industrialize AI and those that accumulate pilots with no demonstrable financial impact. Potential winners are groups with already solid data assets, teams capable of transforming workflows, and large-scale trust mechanisms; losers are likely to be companies less mature in data, governance, and internal adoption. For leadership, the message is that deploying AI solely to cut costs could miss the most value-creating uses. Without a change in approach, PwC estimates the gap would widen as leaders learn faster, expand validated use cases, and automate decisions more safely at scale.
Verdict : PwC’s study confirms a troubling truth: AI does not spontaneously democratize value creation; it rewards mainly those firms already able to transform their data, processes and governance into a strategic advantage. The real risk for laggards is no longer missing a tech fad, but entering a market where AI durably widens a winner-take-most gap between those who reinvent their business model and those who get stuck in pilots without returns.
Sources :
3. Oracle deploys AI agents for corporate banking
Oracle extends agentic AI to corporate banking with AI-enriched applications and preconfigured agents for treasury, trade finance, credit and loans. The offering aims to automate critical processes, accelerate decision-making and move banks from fragmented manual workflows to a unified, real-time, data-driven system, while maintaining human supervision and AI governance.
Key takeaway: Oracle Financial Services is expanding its agentic AI platform to corporate banking with AI-enhanced applications and preconfigured agents for treasury, trade finance, credit and loans. The offering aims to automate critical processes, accelerate decision-making and transition banks from fragmented manual workflows to a unified, real-time, data-driven system, while maintaining human supervision and AI governance.
In practice: Specifically, Oracle introduces agents capable of extracting data from long and complex loan contracts, structuring financial data, validating information, monitoring external news sources to detect risk signals, and then generating a first usable draft of a credit memo. On the trade and supply chain finance side, the platform can also validate banking collateral files, spot non-standard clauses, and configure supply chain financing programs from commercial contracts. Oracle notes that bankers remain in a human-in-the-loop mindset, with review of anomalies and final approval. Stock Titan adds that Oracle plans to deploy hundreds of corporate and retail banking agents in the next 12 months.
Analysis: This announcement fits into a broader banking sector movement where technology vendors move from generic generative AI pilots to agents integrated into workflows such as credit, compliance, servicing, or treasury. The market context provided by American Banker notes that competitive battles are played out mainly within core banking systems and regulated processes, where established players like Oracle, FIS, Fiserv, Temenos and cloud partners can distribute AI via software already used by banks. This clarifies Oracle’s strategy: prebuilt agents serve to strengthen the platform’s relevance and capture automation budgets ahead of specialized startups. The same context also reminds that data quality, model governance, and explainability remain major barriers to rapid large-scale deployment. On markets, Stock Titan notes a positive reaction in Oracle’s stock on the day of the announcement, signaling that investors saw this expansion as a credible use case for AI applied to critical workflows.
The stakes: The potential winners are Oracle, which can strengthen its foothold in strategic banking software, and client banks, which can reduce manual work, accelerate loan processing, and further standardize controls on sensitive operations. Bankers do not disappear from the process, but their role shifts to supervision, validation, and exception management, which can improve productivity without eliminating accountability. Potential losers are providers of isolated AI tools that do not have access to banks’ core systems, as well as institutions that are slow to modernize their data and governance. The central stake is whether Oracle can transform this road map of hundreds of agents into real adoption and measurable gains in an environment where compliance, explainability, and trust govern every deployment.
Verdict: Oracle is right to bet that the next real wave of AI in banking will not play out in generic demonstrations, but in the automation of regulated workflows where speed, traceability, and human oversight create immediate value. But the bet will only pay off if banks prove that these agents truly improve credit, treasury, and trade finance without compromising explainability or accountability, because in finance, powerful AI without governance is not innovation, it’s a systemic risk.
Sources :
4. Vision-Language Models still struggle to read candlesticks
A new study shows that vision-language models still struggle to read candlestick charts where markets spend most of their time: in sideways and ambiguous phases.
Key takeaway: This arXiv paper proposes a new multi-scale benchmark to test whether vision-language models truly read candlestick charts, using daily/weekly image pairs of HS300 and S&P 500 stocks from 2015 to 2025. The results show that most VLMs perform mainly in persistent uptrends or downtrends, but remain weak in more ordinary sideways markets, with pronounced prediction biases and low sensitivity to the forecast horizon requested.
In practice: In practice, the study isolates the visual contribution by building a dataset centered on candlesticks as the primary input, rather than multimodal datasets where text and tables muddy attribution of performance. The authors generate 193,524 candlestick samples from the HS300 and S&P 500 constituents, then evaluate several commercial VLMs against an XGBoost baseline. The protocol combines confusion matrices, IC and Rank IC to measure both direction and ranking quality of forecasts. The authors conclude that certain models show targeted potential — notably Claude-sonnet-4-5(thinking) for direction and ranking, and Gemini2.5-pro on the S&P 500 for mean IC and Rank IC — but that current VLMs resemble more short-term signal aids than robust long-horizon forecasting systems.
Analysis: The market context points in the same direction: interest in multimodal AI in finance extends from textual assistance to chart reading, document analysis, and signal generation, but reliability requirements remain high. Risk.net’s context notes that fund managers and risk teams stay cautious, since visually convincing outputs can fail when market regimes shift, data becomes noisier, or edge cases arise. This is precisely what this benchmark seeks to test under more realistic conditions, checking whether models extract trading-worthy information from charts or if they succeed mainly in simplified conditions. The paper also shows that 68.1% of days in the dataset are in a sideways market, the most common regime, and that is precisely where performance remains weak. In this frame, the growing call for benchmark transparency, explainability, and stress tests aligns with the empirical limits highlighted by the authors.
The stakes : The stakes are concrete for funds, fintechs and quantitative teams exploring VLMs to transform charts into actionable signals. Potential winners are researchers and product teams able to build short-term decision-support tools, well-bounded, tested across regimes and used with supervision, rather than sold as autonomous forecasting engines. Potential losers are players who would confuse a convincing chart-reading capability with robust production forecasting, especially in sideways and volatile markets. For the market, this work reinforces the idea that before any real deployment, financial multimodal models must prove not only average performance, but also stability, calibration and stress-tested behavior.
Verdict : This study serves a valuable function for markets: it reminds us that reading a chart convincingly is not predicting a market, and that Vision-Language Models remain fragile exactly where real-world market conditions are most common—sideways and ambiguous phases. The proper use of these models is not to market them as trading oracles, but to confine them to tightly tested tactical aids, because in finance, impressive AI in obvious trends can become dangerous as noise replaces direction.
Sources :
5. Generative AI Linked to More Jobs in the US
According to a new study, the US sectors most exposed to generative AI have accumulated productivity gains, more jobs and rising wages, countering the scenario of mechanical job destruction.
Key takeaway: A new study by Christos Makridis and Andrew Johnston concludes that, between 2017 and 2024, the US sectors most exposed to generative AI have not only recorded productivity gains but also more jobs and stronger wage growth. In 2024, a level of AI exposure one standard deviation higher was associated with about 10% more productivity, 3.9% more jobs and 4.8% higher wage growth compared to comparable sectors in the same state.
In practice: The study measures AI exposure based on tasks dominated by language, code or data processing, then compares it to employment, wage and output trajectories. Results suggest that when AI complements human work — for example in marketing, writing or financial analysis — employment rises, with about a 3.6% increase per standard deviation of exposure. Conversely, in roles where AI can act more autonomously, the authors find no significant job rise, and wage growth is slower. The Conversation also notes that gains began as early as 2021, driven by enterprise tools already embedded in workflows, such as GitHub Copilot, Jasper and GPT-3-based business applications.
Analysis: The broader frame is a labor market where AI effects are not distributed evenly. The IMF context indicates results vary by country, sector and company, depending on digital infrastructure, worker retraining, management quality, and the balance between increasing human capacity and substituting work. This aligns with the paper: positive effects appear mainly where companies use AI to boost production and reorganize tasks, rather than eliminate work. The Conversation adds that observed benefits were concentrated in states with more efficient labor markets, reinforcing the idea that local institutions matter as much as technology itself. Phys.org highlights this point in public discourse on employment: the result is not mechanical job destruction, but depends on how AI is deployed and the level of trust created in organizations.
The stakes : The winners are the companies able to use AI as a productivity-boosting tool, employees whose tasks can be augmented rather than automated, and regions where labor markets allow rapid redeployment of skills. Potential losers are workers in more standardized and easily automatable roles, especially if firms favor substitution over task redesign. For leaders and policymakers, the issue is less whether AI eliminates tasks than whether the economy can translate this efficiency gain into new demand, new roles and higher wages. The authors also stress a concrete lever: without clear strategy, managerial trust and psychological safety to experiment, AI adoption remains weaker and its economic benefits risk concentrating among a small number of players.
Verdict : This study weakens the lazy narrative that generative AI mechanically destroys jobs: when deployed to augment human work rather than replace it, it can simultaneously boost productivity, hiring and wages. But the political and economic message is more demanding than optimistic: without fluid labor markets, retraining, and company strategies centered on complementarity, AI gains will mostly benefit a few sectors and leave automatable roles to bear the cost of the transition.
Sources :
