A single percentage point of deviation is enough, and a retailer the size of HSE (Home Shopping Europe) ties up millions in capital that is missing elsewhere.
Returns and cancellations forecasting is an underestimated finance topic for many retailers. It feeds directly into the provisions for month-end closing, and every misjudgement ties up too much or too little capital. Both have tangible consequences for liquidity and planning reliability.
Anyone who wants to steer their direct business profitably therefore has to know the costs that arise at the customer interface. HSE brought this topic to foobar Agency as a problem, without a prescribed solution. The first question was therefore a different one from the question of the model: how can a good forecast be recognised at all?
HSE (Home Shopping Europe) is one of Europe's largest live commerce companies. Its business model combines TV commerce, social commerce and eCommerce, with a correspondingly high order volume and an equally high share of returns and cancellations, as is typical in this sector.
For financial planning, this means every month-end closing depends on forecasting as precisely as possible how many of the current orders will ultimately stay with the customer.
At HSE, the forecast for returns and cancellations relied on a rule-based procedure. These rules had grown over years, but kept pace with reality less and less.
For month-end provisions, every inaccuracy meant a real financial risk. A deviation of just a few percentage points is enough to tie up millions in capital that is missing elsewhere in the business.
Before a model could be assessed, it had to be clear what success would even be measured against.
Because behind the single question “How much will come back?” there are three different use cases at HSE, from short-term operational steering through month-end closing to longer-term analyses. Each has its own time horizon and its own demands on accuracy.
In joint working sessions with controlling and the business units, the relevant metric was defined for each of these cases. Month-end closing calls for a different yardstick than day-to-day steering. This included the definition work: which data feeds in, what counts as a return, which comparison period applies.
This jointly agreed benchmark is the transferable result of the project. It decides whether a model version goes into production, and it holds beyond this one model.
The data platform was a given. What had to be decided was where the forecast would live in future and how it could keep developing. A forecast that feeds into provisions and month-end closing has to withstand every review.
Move the existing logic first. The existing rulebook came from the world of controlling. In the first step it was rebuilt one to one in the data platform, so that it became measurable, comparable and replaceable there. As in all predictive analytics projects at foobar Agency, this simple, verifiable version served as the benchmark that every further model version had to prove itself against.
The result of this comparison was instructive for both sides: a considerably more elaborate model performed worse than the simple but well-considered benchmark. An insight that reaches beyond this project, because complexity only pays off once it proves its value.
An order for what comes next. So that the forecast stays alive after handover, it was established how versions come about and replace one another: the best version runs in production, challengers are tested alongside it against the same benchmark. That answers the question that always stands at the end of data projects, namely what happens once the external support is gone.
At HSE, month-end provisions are now based on a forecast that can be measured against reality, instead of on rules that had grown over years.
Since the end of April 2026, the new forecast has been running fully in production; the old rulebook has been retired. For financial planning this means above all: the provision follows an expectation that can be checked, and its deviation is traceable month by month.
Just as important as model quality was the nature of the handover. HSE's own data science team now improves the forecast every sprint and applies it to new questions.
We built the returns and cancellation forecast together with foobar Agency. Today we improve it every sprint with our own data science team, on a foundation that makes the figure explainable and defensible, in the month-end closing and towards management.
What tipped the scales was that consulting and implementation came from one team. The same people who worked out the definitions with the business units then built the model and anchored it in the existing platform.
Anyone who has repeatedly seen where models like this fail in practice recognises more quickly which approach holds up. That matters especially when results feed directly into provisions and month-end closing. It was precisely this experience from many comparable projects that made the difference for HSE.
Equally important was a team that could devote itself exclusively to this topic. That is how a rulebook grown over years became a learning model running in production within weeks.
During implementation it quickly became clear that the potential extends beyond the original project scope. That is an indication of how much remains undiscovered in the data of a company like this.
The forecast also lays the groundwork for further steps. Once it is foreseeable which orders will come back with what probability, more can be derived from it than a provision. Conceivable are impulses for range management, for product presentation or for the service experience at the customer interface.
How solid the collaboration is also shows in the fact that HSE and foobar Agency presented the project's learnings publicly together: Wolfgang Jarnach and Dominik Thalmeier gave a joint talk at the Commerce Impact Forum on the path from rulebook to learning forecast.
Every further stage builds on a data platform designed from the outset for exactly these kinds of questions, as part of a collaboration that has long since grown beyond this single project.
How many new customers a campaign brings was always measurable at HSE. What those customer relationships would still be worth in future could only be estimated. Today HSE steers its marketing budget by the actual value of entire customer groups.
Since 2019, foobar Agency has accompanied porta Möbel on its path to digital market leadership: from legacy analysis via incremental headless commerce to a full composable commerce architecture on MACH principles.
We look forward to your enquiry.
Please accept marketing cookies to load the registration form.