Globally, holiday digital sales in 2025 are projected to reach around US$1.25 trillion, with AI-assisted or agent-influenced orders expected to account for US$263 billion, according to Salesforce. Over half of Singaporeans (57%) plan to increase their holiday spending this year, reflecting strong consumer confidence, and 56% are starting their shopping early, some as soon as August, based on findings from Integral Ad Science (IAS).
As spending increases, new consumer shopping habits emerge
One thing that has changed significantly in this year’s IAS report is how consumers are looking for products. The findings show AI-powered shopping is expected to increase this year, with AI-driven traffic projected to rise by 520% year over year and peak in the 10 days leading up to Thanksgiving. Much of this activity is expected to focus on research rather than completed purchases.
In Singapore, the same IAS report found that 74% of Singaporeans are turning to AI tools for gift inspiration, with 27% using them to research product or gift ideas online. Top items for AI-driven research include toys, electronics, jewellery, and personal care, as consumers compare options and look for value.
Retailers responding to AI with AI
Retailers are also adopting AI-enabled tools to manage higher traffic volumes, increased order complexity, and regulatory requirements linked to evolving trade policies. Deloitte has surveyed retail buyers to understand how they are approaching the holiday season, noting: “Newer forms of AI and advanced analytics, barely on the radar for retail buyers in 2020, may be helping to build resilience for the 2025 holiday rush.”
The report found that 78% of surveyed retail buyers use AI-enabled tools to support buying activities, while 74% use AI to address challenges linked to trade policy changes. Respondents using AI reported improvements in several areas, including:
- Better supply chain management, using analytics to anticipate potential disruptions and optimise logistics.
- Pricing optimisation, where algorithms analyse market trends and consumer behaviour to adjust prices based on demand.
- Product assortment optimisation, helping streamline inventory management so products are available where and when needed.
- Demand forecasting, using predictive models to reduce overstock and stockouts.
AI implementations are easier said than done
While these benefits highlight the potential of AI during the holiday season, implementing AI at scale across national or international retail operations is complex.
Out of the box, foundational large language models lack the contextual awareness needed to collect, analyse, and act on enterprise data. Retailers require an underlying technology architecture that allows AI systems to access contextual, real-time data across operations spanning suppliers, warehouses, physical stores, and online fulfilment, often across multiple regions.
Why architecture matters for AI
This becomes more relevant with the emergence of agentic AI applications. Unlike traditional systems that rely on predefined instructions, agentic AI applications can make decisions and adapt to changing conditions.
Supporting this requires real-time access to business-critical data and a continuous flow of contextual information. From an architectural perspective, this places new demands on how retail technology teams integrate systems and data sources.
Event-driven architecture (EDA) is one architectural approach used to address these challenges. By decoupling systems through an event broker, or a network of brokers known as an event mesh, EDA reduces tight dependencies between components. This can help address some of the scalability and maintenance challenges seen in earlier system designs, including some early microservices implementations.
Using an event broker allows systems and applications to communicate asynchronously. This loose coupling supports independent development and deployment, reducing the need for teams to coordinate tightly across systems.
Event-driven data movement and retail AI
An event mesh can support retail AI initiatives by enabling real-time data sharing across systems. By creating a unified data layer, an event mesh can improve visibility across supply chains and help retailers respond more quickly to disruptions. It can also support analysis of customer interactions across channels, which may assist retailers in adjusting strategies as demand patterns change. In this way, event-driven approaches can contribute to broader omnichannel capabilities.
AI-driven retail use cases
Below are several examples of how AI, combined with event-driven data flows, is being applied in retail environments today.
Inventory and demand prediction
AI-powered inventory management extends beyond basic tracking to support demand prediction and automated restocking. By integrating event data from sources such as shelf sensors or RFID tags, AI systems can analyse stock levels and anticipate future needs, factoring in variables such as weather. This can help reduce waste, limit stockouts, and improve operational efficiency.
The ‘silent shopping’ experience
The ‘silent shopping’ model applies AI to address in-store operational issues with minimal customer disruption. Events triggered by factors such as temperature changes or detected spills can be routed to staff devices. AI systems can help prioritise alerts based on urgency and location, supporting timely responses while minimising visible disruption.
Real-time loss prevention
Some retailers are integrating multiple data streams to support real-time loss prevention. Video feeds, point-of-sale data, and inventory information can be analysed together within an event-driven architecture. AI can assist in identifying potential loss events and notifying security teams with relevant context, supporting more targeted interventions.
Omnichannel personalisation
Omnichannel personalisation aims to provide a consistent customer experience across digital and physical touchpoints. Using an event backbone, AI systems can capture interactions such as website activity or in-store signals to maintain an up-to-date customer profile. This can help ensure customer context is retained when switching channels, supporting more relevant engagement.
The AI store manager concept
Some retailers are exploring AI-driven systems that coordinate store operations by processing real-time events from sources such as foot traffic sensors, smart shelves, staff devices, and weather data. These systems can support decisions related to staffing, layout adjustments, or restocking, contributing to more responsive store operations.
AI is not just for Christmas
The effectiveness of AI in retail depends heavily on the quality, timeliness, and context of the data feeding those systems. To support use cases such as pricing analysis, supply chain coordination, and customer engagement, AI systems require continuous access to data from across the enterprise, including warehouses, supplier platforms, physical stores, and online channels.
Traditional tightly coupled architectures can make this level of real-time data movement difficult to achieve. Event-driven approaches, including event meshes, offer a more flexible and decoupled option for distributing data across systems. Used appropriately, these architectures can help provide the contextual data AI systems rely on, both during peak retail periods and beyond.
