fr-FR
For years, artificial intelligence in commerce has primarily helped consumers find what they were looking for more quickly. Recommendation engines, intelligent search, chatbots and personalisation supported the journey, but the customer remained in control.
Agentic commerce represents a more profound shift. AI agents no longer simply provide advice. They can interpret an intent, search for products, compare several options, check availability, apply price or delivery constraints and, in some cases, complete the transaction on the customer’s behalf.
A consumer could ask: “Find me a pair of waterproof hiking shoes in my size, for less than €150, that can be delivered by Friday.”
The agent can then analyse available offers, product specifications, inventory levels, delivery times, return policies and known customer preferences before presenting a recommendation or completing the purchase directly within the permissions it has been given.
The distinction from a chatbot is critical. A chatbot responds. An agent pursues an objective, orchestrates multiple steps and takes action.
This development therefore creates more than a new conversational interface. It redistributes roles across the buying journey: the customer expresses an intent, the agent evaluates the options, the merchant’s systems provide the required information and the platform executes the transaction.
In traditional commerce, brands optimise their websites to attract, reassure and convert human visitors. They invest in search visibility, merchandising, content, navigation and checkout.
In an agentic journey, the agent can interact directly with the company’s catalogues, inventory systems, payment services and APIs. The consumer may not visit the website at all—and may never even see the checkout page.
Brands must therefore meet two requirements:
This new type of buyer will not be persuaded by a banner or a visually engaging journey. It assesses the relevance of an offer using precise information: product specifications, price, availability, delivery times, return conditions, loyalty benefits and possible restrictions.
Product data quality therefore becomes a direct driver of visibility and conversion. A poorly described product, an outdated inventory status or an ambiguous return policy may prevent an agent from selecting an otherwise relevant offer.
Beyond SEO, businesses must prepare for a new form of discoverability: making their products readable, comparable and recommendable by AI agents.
Agentic commerce can play a role throughout the sales cycle. However, not every purchase will be delegated in the same way. The appropriate level of autonomy will depend on the context, the risk involved and the emotional or financial value of the decision.
An agent can translate a request expressed in natural language into actionable criteria and identify the most relevant products. Instead of searching for keywords alone, it can account for a need, budget, occasion, known preferences and operational constraints.
For the customer, this reduces the time spent filtering through hundreds of products. For the brand, recommendations can become more accurate—provided its product information is rich, reliable and up to date.
An agent can compare prices, availability, delivery times, shipping conditions, return policies and loyalty benefits. It can also reassess its recommendation whenever the underlying information changes.
Competition is no longer based solely on the displayed price. It also depends on the company’s ability to provide a complete, reliable value proposition that can be interpreted and acted upon in real time.
Frequent, predictable and low-involvement purchases are natural candidates for automation. An agent can anticipate a renewal, monitor household inventory, recommend the best time to reorder or trigger a purchase according to predefined rules.
For businesses, these use cases can strengthen repeat purchasing and reduce friction. They also require clear permissions, including maximum amounts, frequency, authorised categories and human approval above a defined threshold.
When permitted by the customer’s mandate, an agent can proceed through to purchase. Payment must therefore work within a journey in which the user does not personally complete every click.
The challenge goes beyond simplifying checkout. Businesses must be able to authenticate the agent and its mandate, protect payment data, detect fraud, retain proof of authorisation and manage subsequent refunds, disputes or cancellations.
An agent can track an order, notify the customer of a delay, request a change, initiate a return or contact customer service. It can therefore become an active participant throughout the entire order lifecycle.
This continuity creates the potential for a more proactive customer experience. It also requires commerce, Order Management, payment, customer service and loyalty systems to share consistent information.
Agentic commerce promises less friction for customers and greater efficiency for businesses. But it also raises a strategic question: who truly owns the relationship when an agent controls discovery, comparison and checkout?
If a third-party interface captures the customer’s intent, defines the options presented and retains the key transaction signals, the brand risks being reduced to an interchangeable supplier. Decisions may then focus on the easiest criteria to compare, such as price and speed, at the expense of experience, brand identity and loyalty.
The challenge is not to choose between participating in agentic commerce and protecting the customer relationship. It is to design this new channel in a way that preserves:
In sectors where purchases remain emotional, experiential or highly considered, journeys are unlikely to become fully autonomous. Customers may want to delegate the initial search while still speaking with an advisor before making a decision.
The most credible model is therefore hybrid: agents manage low-value or high-friction stages, while people step in whenever expertise, trust or emotion makes the difference.
Agentic commerce is not simply a matter of deploying an agent. Its performance depends on the organisation’s entire technology environment being able to provide reliable data, execute the requested actions and enforce business rules.
The agent needs to understand the customer, including their preferences, history, consent and loyalty status. This information is often fragmented across the CRM, e-commerce platform, marketing tools, customer service, POS and payment systems.
A unified data layer can reconcile these signals and provide the agent with consistent context while respecting access rights and customer consent.
Descriptions, attributes, prices, inventory levels, delivery times, promotions and commercial policies must be complete, current and machine-readable.
This requires close integration between the catalogue, PIM, commerce platform, OMS and logistics systems. Without this continuity, the agent may recommend an unavailable product or make a promise the company cannot fulfil.
Agentic commerce standards are still evolving. Businesses must be able to interact with multiple agents, interfaces and protocols without rebuilding their architecture for every new participant.
Reliable APIs, an integration layer and business logic decoupled from individual interfaces allow channels to evolve while the company retains control of its data and processes.
An agent must understand not only what it can do, but also what it cannot do.
Businesses need to define authorised actions, thresholds requiring human approval, pricing rules, data usage conditions and escalation mechanisms. Every significant action must be explainable and auditable.
A recommendation only creates value if it can be executed. The agent must be able to check inventory, apply a promotion, create a basket, initiate a payment, place an order, track delivery and engage customer service.
The objective is not to place an agent on top of fragmented systems. It is to provide the agent with a reliable environment in which every recommendation is based on the right data and every action complies with the company’s rules
Agentic commerce does not rely on a single solution. It requires a coherent combination of e-commerce capabilities, customer data, artificial intelligence, order execution and the systems that maintain continuity across the journey.
Within the Salesforce ecosystem, this architecture can be built around the following components:
Agentforce Commerce provides the foundation for B2C, B2B and D2C e-commerce experiences. It connects agents with catalogues, pricing, promotions and transactional journeys to support product discovery, selection and purchase.
Merchandising teams can also use AI to automate selected tasks, including product enrichment, promotion creation, offer presentation and performance analysis.
Agents need a reliable view of inventory and orders. Order Management orchestrates post-checkout execution, including allocation, tracking, amendments, cancellations, returns and refunds.
This capability is essential to prevent a seamless conversational experience from leading to a promise the business cannot fulfil.
Einstein provides recommendation, prediction and personalisation capabilities. Agentforce embeds those capabilities into agents that can pursue an objective and execute tasks according to the company’s rules.
Together, they enable businesses to move from AI that makes suggestions to AI that completes a job to be done: finding the right product, preparing an order, resolving a request or triggering the next best action.
To act effectively, an agent must understand whom it is serving. Data 360 unifies profiles, consent and signals from commerce, CRM, marketing, customer service and the point of sale.
This view provides agents with current context, including preferences, purchase history, loyalty status, recent interactions and eligibility for an offer. Personalisation therefore no longer depends on an isolated channel, but on a shared understanding of the customer.
The information an agent needs rarely sits within a single platform. MuleSoft and APIs connect Salesforce with ERP systems, PIM platforms, inventory systems, pricing engines, payment solutions and logistics applications.
This integration layer allows agents to access the right information and trigger actions within the relevant systems, without multiplying bespoke connections for every new agentic interface.
Service Cloud maintains conversational continuity when a request involves tracking, a return, a complaint or a handover to a service advisor.
Marketing Cloud can adapt journeys and communications using signals generated by agentic interactions. Slack facilitates collaboration between merchandising, e-commerce, service and operations teams. Tableau provides visibility into performance, including conversion, recommendation quality, human interventions, incidents, satisfaction and value generated.
Together, these components create a continuous value chain:
Understand intent → activate data → recommend → execute the transaction → orchestrate the order → serve the customer → measure and optimise.
Technology, however, is only part of the answer. Value comes from selecting the right use cases, defining the responsibilities of each system, setting clear rules for autonomy and identifying the moments when people should intervene.
Standards, interfaces and use cases will continue to evolve. However, waiting until they are fully established would cost businesses valuable time. The foundations required for agentic commerce already address current omnichannel performance challenges.
A pragmatic approach can begin with four steps:
Initial projects should measure more than conversion rates. Businesses should also monitor recommendation quality, the level of human intervention, average order value, satisfaction, errors prevented, operational costs and the ability to recognise and retain customers across different interfaces.
In the future, a growing share of buying journeys could begin in a conversation and continue without traditional navigation through an e-commerce website. For brands, the challenge will no longer be limited to attracting a visitor. They will also need to be selected by an agent, provide it with reliable information and enable it to act within a secure framework.
This transformation does not make the brand experience less important. On the contrary, it requires businesses to distinguish between what can be automated and what should remain relationship-led.
Agents will absorb complex searches, comparisons and selected transactional tasks. Brands will need to preserve the moments when advice, emotion, creativity and trust build lasting preference.
The businesses that gain an advantage will not necessarily be those that automate everything first. They will be those that build foundations strong enough to open their services to agents while retaining control of their promise, data and customer relationships.
Is your commerce, data and CRM architecture ready for this new channel?
BayBridgeDigital helps retailers and brands define and deploy agentic journeys connected to their data, processes and business priorities.
Speak with our experts to identify the use cases you should activate first.
.jpg)
.jpg)