Case study · Scout
Scout: querying your purchases in natural language and finding the data behind the answer
Scout is a B2B procurement platform. It connects property organisations, which manage properties made up of apartments, with product suppliers and service providers. Every order adds to a purchasing history, but querying it often means knowing which filters to combine. In the platform, built with Medusa and Mercur, we developed a purchasing analysis that can be queried in everyday language. It respects each account’s scope, presents charts and filters to explore, and makes it possible to go back to the purchases on which the answer is based.
The organisations that use Scout buy for properties they manage: products ordered from suppliers, and services entrusted to contractors. These purchases go through a single platform, which, order after order, builds up a history shared across the organisation. This history contains what is needed to track spending, compare periods or identify where amounts come from. It still has to be possible to query it. A question that is simple to put into words, for example about an expense, a supplier or a period, often requires a conventional tool to pick the right criteria, combine them and check that none has been forgotten. We wanted the person responsible for tracking purchases to be able to ask their question the way they think of it. The answer also needed to be verifiable against the original data.
Asking your question in your own words
The starting point is a question written in natural language. A person who tracks purchases does not have to translate their request into a series of filters themselves. They describe what they are looking for, and the analysis takes care of relating it to the purchasing history.
As an illustration, and not answers obtained at a customer, questions might look like these: “What did we spend this quarter on maintenance supplies?”, “Which suppliers come up most often in our orders?” or “How have our purchases changed from one month to the next?”. All of them stem from a need to keep track. They are phrased the way someone who manages purchasing would phrase them, not the way a query tool would.
This approach suits one-off questions. There is not always time to prepare a report for a question that comes up in the course of a meeting or an exchange with a supplier. You simply ask it directly, then refine it if the first wording does not target exactly what you were looking for.
An answer limited to what each account can view
In a property organisation, not all team members have the same role or the same visibility. The person who buys for a property, the person who supervises several teams and the person who handles accounting do not need the same view of spending. Scout already manages these roles within each customer organisation.
The purchasing analysis respects this organisation. It queries the history within the scope to which the logged-in person is entitled. It does not become an indirect way of accessing purchases from another part of the organisation. A question asked in everyday language covers only the properties to which the person has access.
Several profiles can thus ask their questions without widening what they can see: each person who uses the analysis gets an answer that matches what they are authorised to view.
Charts and filters to explore the answer
A one-sentence answer is rarely enough when it comes to spending. When the question lends itself to it, the analysis provides charts that make the breakdown or the change in amounts visible. It is easier to read a trend or spot an unusual value on a chart than in a list of orders.
The answer also comes with explorable filters, which reflect criteria from the question, for example a period, a supplier or a property. The person can use these filters to continue exploring. They do not have to start again from a blank page or rephrase their entire request.
The natural-language question thus serves as an entry point, not a black box. It prepares the selection work the user would otherwise have had to do alone, then hands control back to them to examine the data from another angle.
Going back to the purchases behind the figure
An amount is only useful if it can be trusted. Before using it in a discussion with a supplier, a restocking decision or a meeting with management, a person who manages purchasing wants to know what it is made up of. The analysis was designed for this move from the answer to the detail.
From the filters associated with the answer, the user finds the data that supports it. They can check that an expected purchase is indeed included in the calculation, see that a supplier accounts for more than they thought, or notice that a period was interpreted differently from what they intended. The answer is no longer an isolated statement: anyone can check it against the purchasing history.
This check also helps correct the question. If the detail shows that the analysis did not apply the intended scope, the user refines their request or adjusts their exploration. They always reason from the same data, not from an estimate produced separately.
An aid to analysis, not an automatic buyer
The role of this feature is clearly bounded. It reads and presents the purchasing history. It places no orders, changes no baskets and makes no purchasing decisions on the teams’ behalf. Orders continue to follow Scout’s usual flow, with the permissions and approvals specific to each organisation.
This separation makes it possible to ask questions freely. Querying your spending commits you to nothing. The decisions that follow from an analysis, such as renegotiating, consolidating orders or changing supplier, are still made by the people concerned, using the tools designed for that purpose.
Property organisations that buy on Scout thus have a direct way of querying their spending. They ask the question in their own words and receive an answer limited to their scope, illustrated with charts and accompanied by filters to explore. They can then trace it back to the purchases it is made up of. The purchasing history becomes easier to consult. Verification remains possible at every step, and no purchasing decision is handed over to the tool.