Deploying large language models at the edge in retail is not a silver bullet, and there can be a mismatch between expectations of the technology’s abilities and the effort required to orchestrate, manage, update, and monitor systems. In some cases, ‘traditional’ deterministic solutions are more efficient than those provided by machine learning. The degree to which machine learning is appropriate tends to be dependent on the nature of the data to be processed, and the speed at which responses have to take place.
Fraud detection is a pure classification problem with relatively clear inputs and outputs, and is often considered a suitable starting point for LLMs: a security platform’s output is binary, and its input is drawn from existing models. Buyer segmentation is also an entry point into LLM rollout projects for some, with empirical first-party data that does not change significantly over time. However, product recommendation and supply chain optimisation are more complex. These need rich datasets drawn from many sources, each of which may present rapidly-changing data in different formats, needing highly-sophisticated pipelines.
Many industry benchmarks, such as the proportion of Amazon sales attributed to recommendations, derive either from secondary analysis or are produced by vendors themselves. Similarly, consultancies’ research materials tend to stress the potential of cutting-edge technologies in retail, rather than guaranteeing outcomes. In the latter case, a certain degree of bet-hedging is perhaps acceptable, given the variation in operational detail on the ground between retailers.
Sector decision-makers are actively seeking solutions to their particular problems, and deploying ML is an option that’s presented to them daily. A large proportion of machine learning projects fail to reach production, but that may be due to underestimating the effort required to prepare data, provide infrastructure, and properly equip the workforce with the right skills. In certain cases, bias affects consumer data, and models need continuous monitoring. Are retailers really prepared for the eventualities and issues specific to implementing machine learning operations in their set of circumstances? And in large operations that span multiple stores, does a fleet-wide solution fit every establishment?
Challenges are amplified in edge settings as scale and variety of operations increase. Distributed deployments need consistent data schema, and mechanisms for monitoring performance can use hundreds, if not thousands, of metrics and data points. These aren’t challenges particular to LLM solutions; they’ll affect deterministic, ‘traditional’ platforms too.
The recommended approach to AI deployment in edge retail is incremental. Narrowly-defined systems can be expanded after the retailer gains familiarity with the new modes of operations. Evidence from enterprise deployments in other sectors suggests small, well-scoped projects are more likely to reach production, but there’s a degree of rethinking required: some problems will be solved, but LLMs need preparation and changes to current working practices. To take a single example, what are the cybersecurity ramifications of allowing an LLM access to real-time data stores so it can better inform customers? Not a simple issue to solve without specialist advice.
Machine learning is a practical tool with constraints. As a tool, it relies on timely, high-quality data and needs iterative development and monitoring. Rapid inference and the infrastructure to support this are needed at the edge, while training (and possibly wider orchestration issues) may be new territory for retailers, and seeking information should be the first step in journey.
(Image source: “Shop” by Firelknot is licensed under CC BY 2.0. To view a copy of this license, visit https://creativecommons.org/licenses/by/2.0)
Want to learn more about IoT from industry leaders? Check out IoT Tech Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and co-located with other leading technology events. Click here for more information.
IoT News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.
