Signals
Q3 2025
Global cybercrime estimated cost 2029 | Statista Global cybercrime estimated cost 2029 | Statista
1. 2.
Sign up for Mastercard Signals
Seeking advantage
Strategy
Sources
Architecture snapshot
Digital commerce is shifting from clicks to outcomes. For thirty years, people learned the grammar of software — menus, forms, checkouts. Now we are moving from manual, command-based interactions to a world in which people and businesses delegate a range of objectives to intelligent systems that can reason, plan and act on their behalf. That demands a new operating model where trust, policy and payments are built in — not bolted on. The enabling technology is still emerging, but the transformation it heralds is so profound that forward-thinking businesses must act now to stay ahead.
listen to this section · 3:50 min
Share your feedback
Checklist
FOR MERCHANTS
Agentic commerce
Agentic commerce describes machine-mediated markets where AI agents interpret users’ intent, orchestrate across digital tools and complete transactions within clear guardrails. In practical terms, a person might begin an interaction by saying, “Replace my running shoes for under $120 and have them delivered by Friday,” and an agent will verify size and fit, search inventories, compare prices and policies, apply loyalty benefits, make payment and issue a receipt — end to end.
A new operating model
Verify shoe fit
Search inventories
Compare prices and policies
Apply loyalty benefits
Make payment
Issue a receipt
Crucially, each action occurs within programmed constraints: budgets, approvals, identity and consent.
Agentic commerce is possible today because advances in reasoning enable smart systems to break complex tasks into smaller steps and fix mistakes along the way. The arrival of open protocols allows these systems to easily connect to different apps and data and use secure ways to handle identity and payments, so they can act safely without risking sensitive information. This means humans can let AI assistants handle routine tasks like reordering or solving problems while they focus on important decisions. For stores, success will depend on making products as easy for machines to find and understand as it is for people. For banks, the challenge is to build trusted AI agents that manage money and follow strict rules, so customers feel confident letting them handle financial tasks.
Why now?
The primary audience is no longer just people but also machines acting for people. Success will depend on being machine-legible — on exposing accurate product data, on real-time availability and on transparent policies via APIs.
For financial institutions
For financial institutions, the competitive frontier moves from app features to earning trust for delegated tasks — embedding policy-aware agents that manage money proactively while meeting stringent compliance and audit requirements.
Trust is at the core of this transition. Delegation requires clear controls, consent and auditable trails so users can ask AI agents, “What did you do, and why?” and get a verifiable answer. User control and transparency are not bolt-ons; they are design requirements. Organizations that pair autonomy with accountability — through graded permissions, human-in-the-loop checkpoints for high-risk actions, and robust dispute and recovery flows — will be positioned to capture the benefits while managing new classes of risk.
Autonomy and accountability
This report explores the architecture, the enabling standards and payment rails, the early use cases and the strategic choices that will determine success. The shift is more than a feature update. It is the moment we stop teaching people to think like software — and start building software that thinks like — and acts for— us.
Agents turn thinking to action, creating a new operating model for digital commerce.
What is agentic commerce?
view section
A new competitive arena
The competitive edge in agentic commerce goes to those who build and enable trusted, legible and interoperable agents.
Superpowers
Designing for the agentic economy
Agentic commerce isn't about giving tasks to machines, but about redesigning how value is created with human-guided AI agents.
Preparing for agentic commerce: A checklist
Operating in the agentic economy requires a fundamental shift in mindset: from building for people who click to building for systems that act. Here's how.
Demystifying how it works
What powers agentic commerce? A layered system of intelligence, trust and security, where payments close the loop.
Strategic questions, key challenges, emerging battlegrounds
As AI agents gain more autonomy, the risk of costly errors and data breaches grows. Find out why.
Horizon scanning
Plausible timeline
Navigating the future of agentic commerce? We explore drivers, challenges and a plausible timeline.
Conclusion
Seizing the agentic advantage
Beyond apps and menus: Discover the new era of commerce where AI agents interpret your goals and handle transactions.
next
home
Navigate issue below
Q2 2025
Securing trust
Cybercrime: New fronters
Accelerating threats, higher stakes
Emergent responses, evolving defenses
Toward a safer future
Gartner Hype Cycle Identifies Top AI Innovations in 2025 Y Combinator’s 2025 Spring Batch Reveals the Future of Agentic AI Y Combinator’s 2025 Spring Batch Reveals the Future of Agentic AI State of AI Q1’25 Report | CB Insights data Agentic Commerce: The Future of Payments | Edgar, Dunn & Company
1. 2. 3. 4. 5.
listen to this section · 2:03 min
previous
As a buzzword, agentic commerce is well-established. But as a concrete phenomenon it is still in its early stages. There are promising prototypes and pilot deployments, but widespread adoption is still limited by technical, regulatory and trust challenges. In terms of the hype cycle,1 agentic commerce is moving from the “innovation trigger” phase into early “peak of inflated expectations.” Media attention and investment are growing, but the timeline toward practical, scalable solutions is still unclear. However, the implications for how value is created and exchanged mean it is essential for leaders to engage with the topic now, preparing strategies and systems for a future where intelligent agents mediate most digital interactions.
LIVE 4min ago
Agentic commerce is a system that operates on a simple but powerful pattern:
What it is: A new operating system
Interpret a goal
The user begins by expressing a desired outcome in natural language, such as “Buy me a new pair of running shoes.” The agentic system is designed to understand this intent, parsing not just the literal request but also any underlying preferences, constraints or context — like preferred brands, delivery dates and budget limits.
Plan & use tools
Once the goal is understood, the system intelligently breaks down the objective into actionable steps. It determines what needs to happen — such as searching for suitable products, comparing prices, checking inventory and reviewing return policies. To accomplish these tasks, the agent leverages a range of digital tools — including apps, web browsers and APIs — and seamlessly orchestrates them to build an optimal path to the user’s goal.
Execute within guardrails
The agentic system carries out the plan within clearly defined boundaries. Execution is governed by the user’s consent, ensuring transparency and control. Secure tokens are increasingly used to manage identity and payment, protecting sensitive information. The system also operates within pre-set policies, such as spending caps or approval requirements, so users retain oversight and can intervene or revoke permissions at any time.
Agentic AI doesn’t just generate content; it interprets goals and acts to complete them.
The fundamental difference between today’s generative AI and tomorrow’s agentic AI is the shift from talking to doing. And unlike simple bots that follow pre-programmed responses or scripts, agentic systems are capable of complex reasoning and autonomous decisioning.
What it isn't: A variation on the same
It’s not just another chatbot or a basic automation script.
It's not a rigid, hard-coded workflow. Its power lies in being smart, adaptive and dynamic.
Agentic commerce streamlines complex tasks by letting AI agents handle entire processes end to end.
The ultimate shortcut
Users simply state their desired outcome and agents manage all steps — from searching to payment — within set rules.
This approach transforms difficult experiences into seamless, automated journeys by managing the entire process.
Fully autonomous agents are still nascent, but the contours of this emerging landscape are becoming visible. Today’s agentic AI is best described as automation-plus — systems that operate within defined parameters but increasingly manage complex tasks on their own. Three breakthroughs are accelerating this shift:
An emerging technology
open protocols
Open protocols that enable agents to connect with tools, and each other, to unlock seamless orchestration across platforms.
secure programmable payments
Secure programmable payments that empower agents to move money with minimal user input, turning intent into concrete outcomes.
large language models (LLM)
Advances in reasoning models and multimodal capabilities enable LLMs to make decisions and interact across diverse formats — bridging the gap between static automation and dynamic problem-solving.
platform leaders
enterprise stacks
specialized agentic companies
emerging and research-focused players
Embed agentic capabilities into their ecosystems — spanning productivity, cloud, developer tools and enterprise platforms. They also offer software development kits, APIs and orchestration frameworks for building custom agents and autonomous workflows.
Platform leaders
Build agentic AI platforms that integrate into business operations, enabling autonomous task execution, workflow orchestration and decision-making within established guardrails. These agents often operate across customer relationship management, IT, HR and data infrastructure.
Enterprise stacks
Focus on domain-specific agentic solutions in IT support, enterprise automation and multimodal task execution. These companies emphasize secure, customizable and operationally efficient agents tailored to business needs.
Specialized agentic companies
Push the boundaries of agentic AI through novel interfaces, persistent memory and autonomous interaction with web and software environments. They often pioneer new user experiences.
Emerging and research-focused players
"Agentic commerce transforms AI from advisor to actor. It's a huge leap forward. Yet with that power comes responsibility, and we must lay the groundwork for trust, accountability and resilience.”
Greg Ulrich | Chief AI and data officer at
Today, people initiate interactions with software and give explicit, one-dimensional direction. Tomorrow, agents can initiate activities and execute multi-step tasks autonomously.
schedule management
mobility
travel
shopping
financial planning
supply chain orchestration
B2B procurement
financial planning & controls
treasury management
Investor interest is surging. Key indicators include the uptick in venture capital flows to relevant startups and a flurry of acquisition activity focused on agentic innovators.
Agentic startups dominating investment
Y Combinator, the startup accelerator and VC investor, has helped launch more than 5,000 companies including Airbnb, Stripe and OpenAI. Seventy-four of the 144 companies2 selected for its accelerator program in the spring of 2025 were building agentic AI solutions.3
…and acquisitions
In Q1 2025, the three largest acquisitions in the AI sector4 were of companies offering AI agent technology for enterprises.
Big growth forecast
The agentic commerce market is forecast to grow 66% annually.5
2025
2030
$136 billion
$1.7 trillion
Q4 2025
Home
listen to this section · 10:54 min
Mastercard unveils Agent Pay Ant International’s Antom Debuts Agentic Payment Solution Powering AI commerce with the new Agent Payments Protocol (AP2) Mastercard unveils Agent Pay
6. 7. 8. 9.
Agentic commerce is enabled not by a single technology, but by a compound system — a layered architecture where intelligence, trust and interoperability work together to deliver reliable, real-world performance. To scale safely and effectively, these systems must combine advanced reasoning with a secure transactional infrastructure.
At the base of the stack sit the enabling components that allow agents to interact with the digital and physical world:
APIs and connectors
to applications, databases, messaging platforms and IoT systems.
Trusted identity and payment rails
including tokenized payments and biometric authentication.
Interoperability frameworks
to ensure agents operate seamlessly across vendors, services and ecosystems.
protocols
that safeguard data and provide compliant, auditable execution.
This foundation supports an agentic stack, a coordinated set of layers, each with a distinct role:
Commerce core
The transactional engine: payment infrastructure, authentication, Know Your Customer (KYC) and Know Your Agent (KYA) compliance and risk assessments.
Reasoning & planning
Cognitive systems that break down goals, generate chains of action and validate outcomes.
Governance
System-wide guardrails covering rules, audit trails, dispute resolution and explainability.
Protocols & interoperability
Common languages for secure communication: passkeys and tokens for agent and user identity, network tokens for payments, and standardized commerce messages (offers, receipts, confirmations).
Experience
The user interface: agentic browsers and apps for consumers, merchant co-pilots for businesses.
Tools & execution
The agent’s toolkit for commerce: APIs for discovery, pricing, inventory, fulfillment, support and returns; "headless" websites where the front-end user interface is decoupled from the back-end commerce engine, enabling seamless integration with AI agents.
Underneath, multiple forms of intelligence operate in concert:
Large language models (LLMs)
for reasoning, dialogue and contextual understanding.
Planning systems
to organize tasks and select the right tools.
Perception modules
that use sensory inputs to read shipping labels, verify product images and detect shelf stock.
Action modules
that translate intent into results — initiating payments, booking tickets or dispatching deliveries.
Reinforcement learning
that enables adaptation and continuous improvement based on feedback.
Together, these elements form the intelligent core that allows agents to act autonomously, yet responsibly.
This infrastructure came a step closer to reality with the recent launch of:
Model Context Protocol (MCP)
Standardizes how agents connect to online tools, APIs and services they need to get jobs done in a standardized, programmable way. It is frequently characterized as a "USB-C for agents" with one negotiated interface and many interchangeable tools.
Agent-to-Agent Protocol (A2A)
Connects agents to each other so they can work together across platforms and providers. It is a message protocol for agents to find and coordinate with each other — across companies and platforms. Where MCP connects an agent to tools, A2A connects agents to agents to negotiate and complete multiparty tasks.
AGENTIC PAYMENTS
The rise of secure programmable payment infrastructures that empower agents to complete transactions independently – including Mastercard’s Agent Pay6, Ant International’s EasySafePay7 and Google’s Agent Payments Protocol (AP2)8. This closes the loop on agentic commerce — without payments, agents can only recommend; with it, they can deliver outcomes.
“AI agents will need to communicate across diverse ecosystems with trust and precision. Shared protocols for connectivity are essential. When we build for interoperability, we unlock a future where intelligent agents can deliver real value for consumers and businesses across borders, platforms and institutions."
Jess Turner | Global head of open finance & developer experience at
This architecture is more than technical scaffolding — it defines the conditions under which agentic commerce becomes viable at scale. Without layers for identity, payment and governance, autonomous agents risk fragmentation, fraud and consumer mistrust. With them, businesses unlock:
Why it matters
Seamless customer experience
where purchases, bookings and support requests flow with minimal friction.
Operational efficiency
because routine transactions, reconciliation and service interactions are handled autonomously.
New business models
from agent-to-agent marketplaces to loyalty ecosystems built on verifiable tokens.
Strategic advantage
for the early-adopter companies that set the standards and protocols others must adopt.
In short, agentic commerce is not simply about smarter software. It is about creating a secure, interoperable and adaptive system of digital actors that can transact, negotiate and execute on our behalf.
Without the power to pay, agents remain passive helpers. With it, they become active participants. Agentic payments transform intent into outcome. They close the loop — from planning to purchase — bringing trust, control and automation to the final mile of the journey. Mastercard recently launched Agent Pay, an advanced payment technology that supports agentic commerce.9
Scaling agentic commerce
Mastercard is helping shape the future of AI-powered payments and setting the foundation for trusted agentic transaction standards at scale.
Smarter payment experiences
Agent Pay will incorporate secure payment capabilities into conversational platforms, enabling seamless, context-aware transactions without leaving a chat interface.
Seamless integration
It integrates with leading AI platforms, including Microsoft’s Azure OpenAI Service and Copilot Studio, so AI agents can directly interface with millions of merchants supporting online commerce.
Trust, security and control
Agent Pay will require AI agents to be registered and verified, after which they will be able to make secure payments on behalf of their users. Consumers will control what the agent is allowed to purchase.
Enhanced tokenization technology
The program introduces Mastercard Agentic Tokens, which build on proven tokenization capabilities that today enable mobile contactless payments, secure card-on-file, Mastercard Payment Passkeys and programmable payments.
Expanding and securing the ecosystem
Mastercard is actively collaborating with AI and commerce leaders including Microsoft, OpenAI, Stripe, Google, Checkout.com, Braintree and Ant International’s Antom to make secure agentic transactions accessible and scalable for digital merchants and platforms globally. We’re also partnering with IBM to accelerate B2B agentic use cases, and by the holiday season all U.S. Mastercard cardholders will be enabled for the Mastercard Agent Pay program, with global rollout to follow shortly thereafter. Globally, Mastercard is helping define interoperability protocols that will allow agents to transact safely and seamlessly across the open internet.
Powering agentic M2M payments
We’re also extending agentic capabilities into machine-to-machine (M2M) transactions. Through our collaboration with Pairpoint, Vodafone and Sumitomo, Mastercard is rolling out embedded payments technology that allows machines to trade with other machines on behalf of enterprise users. This will unlock new efficiencies in sectors like fleet, freight, shipping and logistics — supporting autonomous payments for fuel, port charges, handling fees, storage and more.
“Agent Pay bridges intent and action — helping give AI agents the power not just to advise, but to act. It’s a foundational shift in digital commerce, where trust, automation and execution converge to unlock entirely new consumer experiences.”
Pablo Fuerez | Chief digital officer at
6. 7. 8.
Jess Turner | Global Head of Open Finance & Developer Experience at
listen to this section · 9:42 min
The Agentic Bank - Announcing Griffin's MCP Server Implementation Journey: Bringing AI Agents into Your Bank | by Santosh Vaidya | Medium Walmart bets on AI super agents to boost e-commerce growth | Reuters Agentic AI in Retail: Real-World Examples and Case Studies Retail Rewired Report | Walmart
9. 10. 11. 12. 13.
Agentic commerce could dramatically change the arena of competition. The advantage may no longer lie with those who have the slickest apps or most prevalent ads, but with those who build agents that are trusted, legible and interoperable.
The next frontier for banks lies beyond the mobile app. Institutions that provide capable financial agents — while maintaining brand reputation, intuitive user experiences and robust partnerships — will be best positioned to maintain primary customer relationships. However, banks face a strategic decision: How much will they enable or restrict third-party agents to perform money movement and financial activity on behalf of users? This choice will shape the trust and control layer in agentic commerce, especially as open banking and A2A payments evolve. Agentic commerce in finance spans several domains: autonomous finance for personal money management, corporate automation for treasury operations, compliance and fraud detection, and agent-to-agent execution for complex transactions. Together, these illustrate how agents can transform both consumer and enterprise financial activities.
Banks: The battle for financial primacy
Autonomous finance
Once embedded in daily life, agents can monitor spending, optimize funds, manage subscriptions and flag risks before they occur.
Corporate automation
Treasury agents can execute sweeps, invoice matching, FX optimization and real-time liquidity management.
Compliance and fraud
Continuous, embedded audits will validate KYC compliance, thresholds and suitability in real time, while AI agents triage alerts and escalate only true anomalies.
Agent-to-agent execution
Mortgage applications, investment reviews and payment negotiations will increasingly occur directly between consumer and bank agents.
UK banking-as-a-service bank Griffin is testing an agentic setup that allows AI assistants to complete banking tasks, with guardrails, for consumers. Griffin has released a beta MCP server that lets agents open accounts, make payments, and read past transactions — initially in a sandbox environment.9
UK banking-as-a-service bank Griffin is testing an agentic setup that allows AI assistants to complete banking tasks, with guardrails, for consumers. Griffin has released a beta MCP server that lets agents open accounts, make payments and read past transactions — initially in a sandbox environment.10
In the U.S., Mountain Credit Union has deployed AI agents across customer service, fraud detection, lending operations and personalized financial guidance.11 As a result, it says, customer satisfaction increased 28 points and operational costs declined by 23%.
Decline in operational costs
For merchants, the competitive challenge is no longer solely about attracting eyeballs. It’s also about making storefronts machine-legible, enriching product catalogs for language model understanding and implementing agent-to-agent negotiation capabilities. Success will depend on readiness for agent-driven personalization, recommendations and loyalty programs.
Merchants: From websites to machine-legible storefronts
Agent SEO (AEO)
Success depends on publishing credible, real-time product feeds and verified reviews.
Agent-to-agent commerce
A request like “sustainable jacket under $100, delivered by Friday” will trigger merchant agents to match inventory, negotiate delivery and apply discounts — automatically redefining competition.
Real-time orchestration
Merchant agents will better adjust merchandising, reroute inventory and coordinate fulfillment across channels based on live demand.
Dynamic loyalty
Loyalty shifts from human campaigns to proactive agent logic that detects disengagement and triggers personalized offers in real time.
Walmart is consolidating numerous bots into four AI “super agents”— for shoppers; employees; developers; and suppliers, sellers and advertisers. For customers, the shopping agent Sparky moves beyond product recommendations to do things like reorders, party planning and even recipe ideas based on fridge photos. Employees and partners will get agents to handle leave requests, pull sales data, onboard sellers, manage orders and build and test new tools. The goal is to reduce friction across shopping and operations and push e-commerce to 50% of sales within five years.12
Amazon and Shopify use agentic AI to equip shopping assistants to learn from browsing history, cart behavior and even abandoned checkouts. Adapting to each customer, they suggest complementary products, predict reorder needs and adjust suggestions based on real-time feedback. According to Shopify’s 2025 Retail Report, stores that use AI-driven personalization see 25% higher average order values and 19% lower return rates.13
Average order values
+25%
Reduction in return rates
-19%
Nearly half of consumers say that they are willing to let AI agents shop within budget rules. Among time-pressed parents, interest rises above 60%. Trust is conditional on clarity — rules, limits and transparency.14
Consumer trust is rising
Agentic commerce compresses decision-making and rewrites the rules of engagement:
Banks
Banks that fail to deploy financial agents risk losing customer primacy. Customers may choose third-party agents or platforms that offer more convenience, automation and proactive financial management. This means banks risk losing their position as the primary interface for customer relationships, leading to reduced engagement, lower brand loyalty, and diminished influence over customer choices and data flows. In short, banks that lag in agentic capabilities risk becoming less relevant in the eyes of customers whose financial lives are increasingly managed by intelligent systems.
Merchants
Merchants who do not adapt — remaining reliant on human-centric websites and manual workflows — risk becoming less visible to consumer agents. Their offerings will not be surfaced, compared or selected by autonomous agents acting on behalf of customers. This could mean a dramatic drop in engagement: fewer agent-driven purchases, less visibility in automated shopping funnels and ultimately exclusion from the new agentic commerce ecosystem. Merchants must transition to machine-legible storefronts or risk being left out of the decision-making process entirely.
Platforms and networks
Platforms and networks that define the standards and protocols are likely to become the facilitators of the agentic era, helping agents to participate, exchanging data within the ecosystem and executing transactions. Engagement will likely be concentrated around these facilitators as businesses strive for visibility and relevance.
10. 11. 12. 13. 14.
listen to this section · 7:10 min
As agents gain autonomy over financial decisions, scheduling and communications, the risk of costly errors grows — especially when agents act without sufficient oversight or context. A misconfigured agent could overdraw an account, book incorrect travel or share sensitive data with unauthorized parties. These failures aren’t just technical — they can have reputational, legal and financial consequences.
The fallibility trap
Containment strategies are essential
Sandboxed environments, rate limits, human-in-the-loop checkpoints and rollback mechanisms must be built into agentic systems. The challenge is balancing autonomy with control.
In an agent-mediated transaction, who (or what) is to blame in a dispute? Enterprises that fail to implement robust containment protocols may find themselves liable for decisions made by their agents, even if those decisions were unintended. As agents act with increasing autonomy, attributing liability when things go wrong becomes a first-order question. Current tools for dispute resolution, such as chargeback processes and arbitration mechanisms, need to adapt to maintain trust, meet regulatory obligations and prevent systemic risk.
Liability and disputes
As agents begin to transact, negotiate and represent users or organizations, verifying their identity and intent becomes critical. Without robust Know Your Agent (KYA) protocols, it becomes difficult to distinguish between legitimate agents, malicious bots and impersonators. This opens the door to fraud, misinformation and unauthorized access. Just as financial institutions rely on KYC (Know Your Customer) to prevent criminal activity, agentic ecosystems will require standardized identity frameworks, cryptographic signatures and behavioral authentication to ensure agents are who they claim to be, across multiple platforms, and are acting within authorized bounds. The development of agent registries, confidence scores and interoperable credentials will be key to building a secure and trustworthy agentic economy. Without these safeguards, the agentic web risks becoming a chaotic and exploitable environment.
Know Your Agent and identity protocols
Users must feel confident that their agents will act in their best interests, respect their privacy and operate within clear boundaries. This means agent providers and platforms must invest in auditable systems, graded permissions and human-in-the-loop checkpoints for high-risk actions. The ability to provide clear answers to questions like “What did my agent do, and why?” will be essential for earning and maintaining delegation.
Winning customer delegation
Who sets the agent's budget, ethical guidelines and preferred vendors? The user? The agent provider? The platform it runs on? The entity that controls the defaults holds immense power. Defaults shape agent behavior, influence purchasing decisions and determine which merchants, products and services are prioritized. For example, an agent pre-configured to favor certain brands or payment methods can steer billions in transaction volume without explicit user input. This battleground extends to ethical and regulatory dimensions. Who decides the agent’s risk tolerance, data sharing policies or dispute resolution protocols? If defaults are set by platforms or financial institutions, users may unknowingly cede control over critical choices. The struggle for default settings will play out in negotiations between users, service providers and regulators, with significant implications for competition, consumer protection and market fairness. Companies that offer transparent, customizable defaults — and empower users to easily review and change them — will be better positioned.
The battle for default settings
This technology creates a new layer between businesses and their customers. As agents become an important interface for transactions, merchants and banks may lose direct access to customer data, preferences and feedback. The agent could become the new “middleman,” controlling the flow of information and value and potentially commoditizing brands that fail to differentiate themselves in agentic ecosystems.
Dealing with disintermediation
sOLUTION
The solution is an open, standards-based ecosystem (like the agentic stack) that ensures participants can still connect directly and compete on value, not on who owns the agent. Interoperable protocols, transparent APIs and shared governance frameworks are essential to prevent lock-in and fragmentation. By participating in open agentic networks, businesses can maintain visibility, offer personalized experiences and compete on the merits of their products and services. The issue of disintermediation is ultimately a fight for relevance and resilience in a world where agents mediate most interactions.
Modernizing legacy systems and shaping future regulations are key challenges in enabling agentic commerce. Many enterprises still rely on siloed, proprietary infrastructure that will hamper agent interoperability. Upgrading these systems means cleaning data, exposing APIs and embedding agentic protocols — requiring cross-functional coordination and sustained investment. Meanwhile, regulatory frameworks will need to update to resolve issues around consent, auditability, liability and cross-border data flows, which require policy-aware routing to ensure compliance with diverse regional regulations and standards.
Enterprise integrations and regulatory gaps
eIDAS 2.0
For example, agents operating in the EU must navigate AI Act risk categories, respect eIDAS 2.0 wallet requirements for identity and trust services and adhere to data residency rules that restrict where personal data can be stored or processed. Embedding policy logic into agent workflows enables safe, scalable and compliant global transactions. Enterprises should collaborate with regulators, industry bodies and tech partners to help define policies that support innovation while protecting consumers. An exemplar of this is the way that payment systems are rethinking chargeback and arbitration processes to handle disputes from machine-mediated transactions.
listen to this section · 6:54 min
Agentic commerce is not about simply handing tasks to machines — it’s about redesigning how value is created, delivered and governed when AI agents become intermediaries. In this new landscape, companies will likely build hybrid systems where agents orchestrate actions and people retain oversight.
The app-first paradigm — where companies build end-to-end human interfaces — is being supplemented by an API-first model. Agents don’t need screens or brand-boosting content. They need atomic services: the modular building blocks that agents can call upon to complete tasks, such as a service that verifies a user’s identity, another that sends an email, or yet another that checks a bank balance. Price, policy, availability and even terms and conditions must be machine-readable.
Traditional model engaging people
App-first experiences
Businesses build apps & websites designed for human browsing and inputs.
UX as differentiator
Value comes from intuitive design, copywriting and visual interfaces.
Persuasion funnel
Designed to influence human decision- making: Awareness → consideration → purchase → loyalty. Focused on the business acquiring the custom of the buyer.
Brand emotion & attention
Loyalty built through advertising, design and emotional resonance.
Siloed systems
Platforms hold data in proprietary formats. Integration is manual and patchy.
Trust as add-on
Identity, permissions and fraud checks are bolted on after core experience.
Human in control
People drive every step. Automation is narrow and rule-based.
Payments as final step
Checkout is the last action in a user journey.
Agentic model engaging machines
API-first infrastructure
Businesses expose structured, machine-readable data and atomic services for agents to consume.
Orchestration as differentiator
Value comes from reliable workflows, exception handling and cross-system coordination.
Execution funnel
Designed to facilitate agents in the completion of task: Intent → orchestration → verification → payment → fulfillment → aftercare. Focused on the buyer acquiring the service of the business.
Trust & performance
Loyalty built on transparency, reliability, policy enforcement and secure execution.
Interoperable protocols
Shared standards enable multi-agent collaboration across platforms.
Trust by design
Control, auditability and tokenized credentials are embedded from the start.
Hybrid oversight
Agents execute autonomously with humans supervising, setting policies and intervening at critical points.
Embedded payments
Agentic payments close the loop instantly within agent-to-agent interactions.
Excellence in core capabilities will serve as a competitive differentiator for agentic systems. Click below to explore these superpowers:
Working a plan
Proficiency with tools
Learning from the past
Following rules
Closing deals
Transparency
Multitasking
what is it?
The agent’s ability to operate browsers, apps, terminals and APIs like a power user — filling out forms, checking inventory, comparing terms, applying coupons and points and completing payments — with user control and within spending caps.
Why it differentiates
The more environments in which an agent can reliably operate (even when APIs are missing), the more end-to-end journeys it can complete — which directly lifts conversion and revenue.
Coverage = conversion
Strong “computer use” lets agents adapt when UIs, captchas or flows change, reducing brittle failures and abandonment.
Resilience to change
Broad tool competence shortens integration timelines and opens access to a wider range of merchants and platforms.
Speed to value
Impact metrics
These include journey completion rate, the range of sites and APIs covered, successful checkout rates, average time-to-task, % of flows recovered after a change to a site’s user-interface.
The agent books a replacement flight by navigating the airline’s site (no API), applies a travel credit and completes payment via a network token — without human input.
Segmenting high-level goals into steps, selecting tools, sequencing actions, handling branches and self-checking against the objective.
Better planning reduces retries and escalations, cutting costs and user friction.
Fewer dead ends
Strong reasoning allows fast extension to new categories and edge cases.
Generalization
Self-checks and correction prevent silent failures — crucial for regulated or high-value actions.
Trustworthy autonomy
First‑attempt success, plan‑repair rate, exception resolution rate, tokens/steps per successful outcome (efficiency)
“Replace my running shoes for under $120 by Friday” triggers a plan that crosschecks purchase history, shoe size, stock, delivery agreements, return policy and loyalty redemption before proposing three options.
Secure portable memory for preferences, histories and constraints — accessed via consented data and identity tokens.
Accurate context produces better recommendations and higher acceptance of agent actions.
Personalization that pays
Continuity
If memory travels with the user (not the platform), the agent remains useful across ecosystems — building loyalty to the agent that earned delegation.
Portability
Repeat use rate, personalization uplift (average order value and conversion rate), “nudge” acceptance; cross-merchant success using the same preferences.
The agent knows the user prefers “sustainably produced, regular fit, U.S. 9” and auto-filters results accordingly.
Enforcing budgets, corporate policies, legal and brand rules, and risk thresholds with graded autonomy: Ask → suggest → act within bounds.
Robust policy engines unlock B2B and regulated use cases (T&E limits, merchant allow lists, category locks).
Enterprise grade trust
Autonomous, policy-safe actions reduce approval overhead while keeping exposure in check and maintaining user confidence.
Fewer escalations
Jurisdictional and brand rule compliance make agents practical at scale.
Global readiness
Indicators would include the proportion of tasks completed autonomously within policy, violations prevented, spend within budget bands and human approval deflection rate.
A corporate agent orders items within policy, auto-upgrading the shipping only when the risk of breaching a service-level agreement (SLA) exceeds a threshold.
Comparing offers, coordinating delivery windows, bundling items and sequencing multiparty steps (customer ↔ merchant ↔ carrier ↔ bank) to achieve the user’s objective.
Negotiation on price, shipping and bundles extracts tangible savings or value‑adds.
Better economics
Orchestration wins complex journeys (split shipments, returns & exchanges, multi‑stop travel) that simpler agents can’t handle.
Frictionless complexity
Merchants and issuers will likely prioritize agents that communicate cleanly and close deals reliably.
Preferred vendor status
Indicators would include the proportion of tasks completed autonomously within policy, violations prevented, spend within budget bands and humanapproval deflection rate.
Given “sustainable jacket under $100 by Friday,” the agent proposes in-store pickup to meet the deadline and secures a loyalty discount to land under budget.
End-to-end visibility with immutable audit trails, conversational explanations (“why this, why now?”) and one-tap control.
People and compliance teams delegate more when they can inspect, assess and overrule.
Earning delegation
Clear logs simplify disputes, chargebacks and support, cutting operating expenses.
Faster resolution
Explainability and provenance reduce model-risk concerns.
Regulatory confidence
Customer satisfaction score, dispute resolution time, % actions with human-readable rationale, time-to-revoke/revoke success rate.
A receipt shows the agent’s decision path (inventory, return policy, best available shipping) with a tap-to-revoke token if something looks off.
Executing multiple steps simultaneously — price checks, inventory calls, delivery quotes — then consolidating results into a single decision.
Parallel work compresses response times from minutes to seconds, which lifts conversion and net promoter score.
Cycle‑time advantage
The agent can consider more markets and suppliers without slowing down — improving price and availability outcomes.
Broader search, same time
Parallel, asynchronous design handles load spikes and failures gracefully.
Peak resilience
Time‑to‑first‑answer: how long the slowest sessions take to get a final answer, number of options evaluated in parallel per second.
The agent queries five retailers and three carriers concurrently, returning the top three fully costed options in under two seconds.
listen to this section · 11:11 min
Agentic AI in Retail: Real-World Examples and Case Studies Agentic AI in Retail: Real-World Examples and Case Studies
16. 17.
listen to this section · 5:47 min
Agentic commerce can be expected to unfold in stages — progressing from narrow experiments to more autonomous orchestration. Companies can best prepare by modernizing systems, adopting open standards and embedding governance and user controls. The innovative tailwinds behind agentic commerce include the exponential pace of technological development and the multiplicity of potential use cases. The headwinds include policy and regulatory gaps, slow enterprise integration and consumer trepidation. A plausible timeline may look like this:
Today – 2027
An initial phase of experimentation to find product-market fit for early agentic systems. This period is marked by intense experimentation across startups, tech giants and enterprise labs. Developers are testing agentic interfaces in domains like customer support, scheduling and financial planning. Key challenges include aligning agents with user intent, managing hallucinations and integrating with legacy systems. Regulatory bodies begin drafting early frameworks, while standards groups explore interoperability protocols.
Prototyping
2027 – 2030
Winning applications and platforms rise to the top and interoperable agent ecosystems begin to gain utility and scale. Early enterprise adoption accelerates, especially in finance, logistics and healthcare. Agent marketplaces emerge, offering plug-and-play capabilities. Governance models, identity frameworks and Know Your Agent (KYA) protocols become critical to ensure security and accountability.
Emerging ecosystems
Beyond 2030
New software designs are natively agentic. Applications are built with agents at the core, capable of autonomous decision-making, negotiation and orchestration. Productivity surges as agents handle complex workflows across systems, but new risks emerge — such as cascading errors, opaque decision chains and ethical dilemmas. Containment strategies, audit trails and agent licensing become standard. The agentic web transforms commerce, governance and daily life, demanding continuous oversight and adaptive regulation.
Going native
Berlin-based fashion retailer Zalando uses agentic AI to manage dynamic pricing.15 Its agents track sales, competitor prices and inventory — then adjust prices automatically to stay competitive while protecting profit margins.
result
+12%
Manage replenishment and discovery.
E-commerce agents
Automate treasury, risk and personalized offers.
Banking agents
Handle door-to-door itineraries.
Travel agents
Proactively manage wellness, finance and schedules.
Personal agents
Dynamically personalize gameplay.
Gaming agents
Revenue increase per SKU
Swedish retailer H&M implemented agentic AI to test in-store merchandising.16 Each store’s AI system tracks how customers move, how long they stay in different sections and what they buy. This data informs product placements.
Basket size increase
+17%
listen to this section · 3:34 min
Operating in the agentic economy requires a fundamental shift in mindset: from building for people who click to also building for systems that act. The challenge for leaders is not only to retrofit today’s operations but also to anticipate the next decade of intelligent, autonomous commerce.
Take a strategic view
Think beyond pilots
Treat agentic commerce as an inevitable operating paradigm, not just an internal experiment. Develop a roadmap for deploying agents that act as your company’s ambassadors — handling customer inquiries, executing transactions and building relationships autonomously.
Scenario planning
Plan for how external agents will acquire, serve and retain customers — considering how they’ll discover your offerings, compare you to competitors and create new engagement and loyalty dynamics. Anticipate market reactions, regulatory challenges and new competition as your agents operate visibly and autonomously in open digital ecosystems.
Redefine competitive advantage
Redefine competitive advantage around the capabilities, trustworthiness and discoverability of your external agents, investing in their visibility and attractiveness to customer agents through agentic SEO, interoperability and transparent policies. Focus on building agents that can negotiate, personalize and deliver outcomes in real time across platforms, ensuring they comply with external standards and adapt to evolving market protocols.
Build for machine visibility
CONTENT AS CURRENCY
Prioritize user-friendly, standardized, machine-readable data. In the agentic era, legibility is visibility — if your systems are opaque to agents, your brand may effectively vanish from consideration.
Adaptive architecture
Build modular, API-first systems that can evolve as protocols change. Futureproof by designing for interoperability rather than platform lock-in.
Machine-to-machine readiness
Prepare for transactions, negotiations and supply-chain orchestration that happen autonomously between enterprise agents.
Design for trust and containment
Make explainability, control and reversibility native features, not compliance afterthoughts. Assume regulators will require real-time auditability and proof of guardrails.
Resilience planning
Expect systemic shocks such as agent-to-agent fraud and cascading errors. Embed detection, fail-safes and human oversight layers to contain risk.
Dynamic risk frameworks
Move from static fraud models to adaptive, continuous trust systems that can operate at machine speed.
Craft parallel journeys
Design for dual audiences
Build parallel experiences that serve both the human and their proxy agent. This means creating customer journeys where oversight is lightweight, transparent and confidence-building.
Skill shifts
Equip your workforce to manage and collaborate with AI agents — from “agent whisperers” who tune preferences to compliance teams that monitor autonomous behaviors.
Automated loyalty
Consider how brand differentiation will survive when agents — not people — filter the funnel. Think about embedding your value proposition in the logic agents use to decide, not just the stories humans see.
listen to this section · 3:47 min
The shift from manual, click-driven commerce to agentic, outcome-driven delegation marks a watershed moment in digital markets. We are moving beyond the grammar of software — menus, forms and checkouts — toward a future where intelligent agents interpret our goals, orchestrate across platforms and transact within trusted guardrails. This is not a binary leap, but a blended evolution: human and agent will increasingly collaborate, with autonomy and oversight flexing to fit the context.
Agentic commerce is already reshaping how we think of value being created and exchanged. Success will depend on making products and services machine-legible, architecting for orchestration and embedding trust and transparency at every layer. The competitive frontier is shifting: Brands must earn the trust required for delegation, financial institutions must build policy-aware agents, and all players must participate in open, interoperable ecosystems to avoid fragmentation and lock-in. With new power comes new responsibility. The risks — costly errors, opaque decision chains and identity challenges — demand robust containment, auditability and Know Your Agent protocols. Trust is not a bolt-on; it is the foundation of delegation and resilience.
"The agentic future is being written today. The organizations that embrace bold experimentation, establish governance early and collaborate to set new standards are defining the rules of tomorrow’s commerce. The time to act is now."
Ken Moore | Chief innovation officer at
Mastercard powers economies and empowers people in 200+ countries and territories worldwide. Together with our customers, we’re building a resilient economy where everyone can prosper. We support a wide range of digital payments choices, making transactions secure, simple, smart and accessible. Our technology and innovation, partnerships and networks combine to deliver a unique set of products and services that help people, businesses and governments realize their greatest potential.
listen to this section · 1:40 min
