New arrivals: 14 Scout case studies

Case study · Scout

Scout: finding and comparing products from several suppliers with the help of AI

Scout is a procurement platform that connects property organisations with many product suppliers and service providers. Finding the right item across all these catalogues requires knowing each one’s vocabulary. We built a search that understands a freely worded request, finds an exact reference, shows only accessible offers and helps compare them using up-to-date information.

AIProduct searchB2B marketplaceProperty procurement

Scout is aimed at property organisations that manage properties made up of flats and other units. To maintain these homes, their teams need parts, equipment, supplies and services. The platform, built with Medusa and Mercur, brings together the offers of several suppliers in a single purchasing space. This consolidation opens up access to a large number of references, but it makes searching harder. Each supplier describes its items in its own way, with its own references and its own warehouses. The person searching, meanwhile, often starts from a need on the ground: replacing a part, fitting out a home, finding a model already in use. Our work consisted of connecting that need to the offers in the catalogue, then helping to choose the one that suits a specific property, taking into account price and available stock.

Describing a need the way it is put on the ground

A keyword search assumes knowing the term the supplier uses. Yet a real request looks more like a sentence, with a product type, a characteristic and sometimes a constraint. By way of illustration, someone may search for a kitchen tap to fit a single-hole sink, rather than typing a model’s trade name.

When a request is worded this way, AI turns it into search criteria. These criteria are not invented: the AI keeps those that actually exist in the catalogue, such as an existing category or a characteristic described in the product pages. The search thus stays anchored in the suppliers’ data, instead of producing filters that would not match any offer.

If AI assistance is not available at the time of the request, the search continues to work without it. The person still gets usable results and can carry on with their purchase.

Searching for an idea or an exact reference

The search can also understand a concept rather than a precise spelling. A product can be found from what it does or what it is, even if the supplier’s product page uses a different word. This search by meaning is useful when neither the brand nor the exact name is known.

Conversely, a technician who copies the reference printed on a worn part wants exactly that part, not items that resemble it. When a search is for a reference, its meaning is preserved: it is not diluted in approximate results. The references and descriptions of all accessible suppliers are queried at the same time, so there is no need to guess in advance which supplier offers the item.

As soon as typing begins, suggestions offer products, categories or suppliers. They distinguish a complete piece of equipment from its parts and accessories, to limit confusion when the same word refers to the appliance and its spare parts. Finally, each person can find their recent searches, without duplicates, to pick up where they left off without retyping everything. They can delete a recent term or clear their entire history if they no longer wish to see these suggestions.

Seeing only the offers accessible to one’s organisation

Not all products on the platform are intended for everyone. An organisation does not necessarily have access to all suppliers, and some offers are not made available to it. The search, including search by meaning, respects these access rules: results are limited to the offers the organisation is authorised to view. Whether they can be ordered then depends on each offer’s stock, delivery coverage and purchasing terms.

This precaution avoids discovering an attractive offer only to find out afterwards that it is not open to the organisation. The choice is made among real possibilities.

Ranking results according to what serves the property

Two offers for the same item do not have the same value depending on where it is needed. On the supplier side, each warehouse is recorded with its address, its stock quantities and the geographical areas it serves. A supplier can specify particular terms attached to the products of a warehouse.

This information makes it possible to rank results taking stock, price and distance into account. An offer that is available and close to the property concerned can thus rank ahead of a distant or out-of-stock offer. Availability is checked for the target site and taking the current basket into account, so as not to show as purchasable a quantity that no longer is.

The real cost of a purchase is not limited to the item’s price. For a given product, it is possible to find out whether it may qualify for free delivery, depending on the warehouse’s delivery coverage and the available stock. This information helps to decide between two similar offers.

Comparing side by side with updated prices

Once a few products have been identified, the person can compare them on a single screen, whatever their supplier. This comparison does not rely solely on the information stored during the search: prices are recalculated at the time of comparison, based on up-to-date data.

Products that have become unavailable or are no longer visible to the organisation are removed from the comparison. The choice is thus made between offers that can still be purchased, on current terms, and not on an outdated state of the catalogue.

A summary of the characteristics that matter for purchasing

Product pages vary widely from one supplier to another. Some are long, others scatter the important information. To make them easier to read, AI writes a summary of the characteristics that matter for purchasing, based on the product data.

This summary remains tied to the product page it comes from. An existing summary is only reused if the product information has not changed; when a supplier updates its product page, a new summary is produced. This prevents a summary from describing a version of the product that no longer exists.

On Scout, a team can start from its need, expressed in its own words or from a reference noted on site, and arrive at a selection of offers that are genuinely accessible to its organisation. The ranking takes into account stock, price and how close the warehouses are, then the comparison relies on recalculated prices and sets aside anything that can no longer be purchased. AI helps to word the search and to read the product pages, but it does not buy anything: the decision and the order remain in the hands of the person comparing. The benefit comes from this way of working: less time spent guessing each supplier’s vocabulary, and a choice made on up-to-date information.

All case studies →

Something to dig into? Add this article to your office hours discussion.

See my discussion