OPINION: The algorithmic shelf: why retailers need to shape what shoppers see

James Taylor, CEO, Particular Audience, looks at how to harness algorithmic attention so that retailers can ensure product visibility and discovery.

Picture a shop assistant who remembers the behaviour of every one of the customers that visit a store each year. One that understands exactly how people search for and describe products, and leads them straight to the item they came in to find; one that instantly recognises visual style and learns continuously in real time to provide the best, most efficient, personalised shopping experience.

That’s the promise of a technology-enabled revolution in online retail. The shop assistant is seamless, subtle and virtual, able to deliver a personalised experience to millions of distinct customers. And the technology behind it is the hybrid recommendation system, which combines machine learning techniques, specifically collaborative and content-based filtering, to deliver personalised, highly accurate product suggestions.



But anyone who has attempted to search for an item with an online retailer who’s getting it wrong – and that means most of them – knows that we’re not there yet.

Perfecting the balance of retail and media

As of today, retail media accounts for one in five dollars spent on online advertising – an enormous figure. In 2021, Boston Consulting Group identified a $100 billion opportunity for retailer media. On its current trajectory, I would argue that $1 trillion might be closer to the mark.

A critical element of this is the experience retailers provide online, balancing ads, organic search and other considerations. The problem is, there is a fundamental flaw in how most retail media networks are approaching this task. Their methodology relies on serving paid ads to shoppers at the expense of the personalised organic results the consumer really wants. This means the RMNs initially sell a lot of ads – which is part of the promise of retail media, after all – but at the cost of total conversions, which quickly stagnate.

Clearly, retailers need to look beyond siloed systems, in which ad revenues cannibalise organic sales, customer trust is eroded because they’re not served high-relevance products in their search results, and short-term ad revenue takes precedence over long-term audience loyalty. 

Winning the attention of the algorithm

To grow retail media sustainably, they need to actively curate discovery; aligning merchandising control, customer experience and sponsored placements, so that what wins on site still feels like shopping.

What that means is that algorithmic attention is becoming as important as human attention – which calls for a new set of skills. To harness the algorithmic kind, retailers need hybrid recommendation systems with retailer-grade control. In this next phase of development, black-box automation isn’t going to cut it.

Hybrid machine learning recommendation systems determine a number of critical choices for retail media networks (RMNs) – including which products are surfaced, which brands dominate, which SKUs get buried, and how shoppers discover alternatives they never previously knew existed.

It does that by blending three major modelling families to understand consumers and drive retail media performance:

1. Collaborative Filtering (CF): This model identifies shopper behavioural patterns, including co-viewing, co-basket behaviour, repeat affinities and behavioural clusters. Collaborative filtering drives substitute suggestions, competitor PDP placements, and “You may also like” modules.

2. Natural Language Understanding (NLP): This model extracts meaning from various language use cases. These include product titles and descriptions, reviews and structured (and unstructured) metadata. It’s worth noting that if an attribute isn’t included in the retailer’s embedding string, the algorithm cannot “see” it.

3. Computer Vision (CV): Unlike NLP, CV understands more visual, aesthetic elements. These include colour, silhouette, texture, pattern, style and quite literally: object detection. It’s great at understanding design elements text cannot express. Photons trump language every time in AI.

The advertiser’s role

And while online retail shouldn’t sacrifice customer experience for the benefit of advertisers, they are clearly an important part of the mix. Accordingly, advertisers must ensure their ads work constructively within the system. They need to categorise their metadata carefully and accurately; provide rich-multi-angle imagery; write natural-language descriptions that feature full sentences; and ask retailers the right technical questions when uploading their ads.

Then, they need to get creative with pricing, promotions and ad deals, because getting a product seen and bought is the ultimate objective. The more an advertiser can get its product co-viewed and co-purchased with other items, the more potential it has to be displayed in personalised and contextually relevant real-time customer contexts.

This isn’t off-the-peg technology; it requires substantial investment of time and effort. Contrary to what many enterprises assume, it is not realistic to build a production-grade hybrid recommendation system from a few AWS/GCP APIs. A truly scalable, retailer-grade multi-modal recommender with a merchandising-friendly UI is not an overnight project.

Nonetheless, hybrid recommendation systems that balance human and algorithmic attention are the future of online retail. Their functionality and feature set allow RMNs to cross-sell, up-sell, product-substitute and bundle. They also enable PDP merchandising, homepage, landing and full-funnel personalisation and automated editorial curation.

If advertisers understand how these algorithms think, they can align with them – ethically, transparently and effectively, delivering on the promise of retail media to offer performance at scale. The algorithmic shelf, like its physical counterpart, rewards those who understand how it works – and the retailers who master that balance will be the ones who earn both the algorithm’s attention and the shopper’s trust.

Also published in: Internet Retailing