Retailers use AI to personalise pages and predict demand

Retailers are replacing static websites and apps with AI systems that change a customer’s page during a live session. Models draw on active clickstreams, past purchases and inferred intent to generate page layouts, native copy and interactive elements when a page loads.

A McKinsey study found 76% of consumers are frustrated when digital experiences do not adapt to their needs. Retailers that deploy real-time page layouts report a 35% rise in purchase frequency and a 21% increase in average order value.

Marketing and analytics teams are expanding beyond text monitoring to systems that ingest video, audio and unlabelled images. Video represents roughly 82% of internet traffic and more than 60% of individual digital media time. Multi-modal platforms that process unstructured video and audio can detect unbranded mentions, product placement and spoken sentiment earlier than keyword-only tools. The market for multi-modal systems is forecast at $2.83 billion this fiscal year, and about 76% of analysts using visual platforms report measurable returns compared with under 60% for text-only operations.

Campaign testing increasingly uses synthetic user simulations built on large language models. Teams create virtual personas that combine demographic, psychometric and behavioral data to mirror group responses, navigation and content feedback. These synthetic cohorts run thousands of automated interviews and usability stress tests inside sandbox environments. High-performance setups refresh virtual users with data from human control groups to keep simulations aligned with market behaviour.

Physical automation in stores and warehouses relies on computer vision trained on spatial layouts and interaction patterns. Use cases include registerless checkout, real-time shelf tracking and automated store navigation. Warehouses run millions of virtual trials to teach robotic arms how to pick and pack irregular items. Forecasts expect physical automation platforms to exceed $370 billion by 2040. Edge computing hardware on the store or factory floor processes sensor feeds locally to reduce latency and avoid routing continuous raw video through central cloud servers.

Standardising how models access backend systems is part of deployments. The Model Context Protocol provides an open connection layer between core models and legacy databases, product catalogs and CRM systems. Operational models use modular instruction packages called skills so applications load only the functions needed for a workflow, preserving context window capacity and lowering token consumption. The Agentic AI Foundation, governed by the Linux Foundation and supported by major technology providers, promotes compatibility for the protocol.

Adopters report that investments in data pipelines, multi-modal ingestion, edge hardware and standardised integration support faster personalisation, more accurate campaign testing and better-aligned inventory. Companies also report that the shift requires substantial engineering effort and governance to manage models and data across distributed systems.