FAQ

On one perimeter. A priority range, a strategic supplier, an important marketplace, a price and availability feed, or a recurring duplicate problem. That scope is small enough to measure the errors, settle the rules and see the gain, and it is what tells you whether to extend the pipeline. We would rather run one of your real files than present a methodology.

The first session works on one of your files rather than a demo dataset: we map it, run it through Fuse and walk you through the exceptions it found. After that, the timeline depends on how many sources, rules and destinations are in scope, so we size it with you rather than quote a number here.

Yes. Imports run in TEST or in PROD, and configuration changes are committed in TEST first. The normal path is to validate a slice of the catalogue in TEST, check the exceptions and the exports, then promote the workflow to PROD where it runs against live catalogue data.

No. Fuse is the quality layer in front of one. It prepares, controls and enriches product data before the PIM takes it in, and again when you publish to the shop, the marketplaces and other applications. Where an organisation has no suitable PIM, Fuse can cover part of the product-data job, but its place is between the sources and the systems rather than instead of them.

It keeps your data fit for the business, on every change. Fuse brings each change in, fills what is missing within the rules you set, and checks the result before anything moves on. What passes is ready for the shop, the PIM, the ERP and your other channels. The work is highly automated and runs continuously, the way a data steward would if one could work every hour of the day. When a change fails a rule or needs a judgement, it waits for a person.

It manages the references, not the files. Fuse checks that image, notice and datasheet URLs are present and mapped, and flags the ones that are missing, so a product does not go live without its media. It does not store or version binary files, so it does not replace a DAM.

Any data, from any system, through a connector. Fuse reaches the systems you already run, and the files and APIs they already produce. Where a connector does not exist yet, one is added around the core, and the rest of the platform stays as it is.

Near real-time, or on a schedule. What passes your rules goes out on the timing you set. Data that fails a rule is parked for review, and it does not reach your systems or your publication channels until a person clears it.

In the European Union, at rest and in transit. That includes the large and small language models the platform provides: they run in the EU, so your data does not leave it when an agent works on a record. Data Punks is based in Brussels.

You do, fully. The data you bring into Fuse, and the records Fuse builds from it, stay yours. You can export them at any time, as files, over REST or through a connector, so leaving Fuse never means leaving your data behind.

Both. Fuse is delivered as SaaS, hosted in the European Union, and it can also run on Kubernetes in your own data centre. In that case your data stays on your own infrastructure.

We keep it as low as we can. The Fuse SaaS runs with a hosting provider powered by 100% renewable energy. And we optimise every interaction with our language models to use as few resources as possible. Much of the work does not need a model at all: formats, completeness, allowed values and your business rules are checked by deterministic rules.

Yes. Any language, and the currencies you sell in.

Both, and the difference is the point. Word-for-word translation produces content that is technically correct and commercially flat, especially when units and conventions differ between Germany, France and Italy. Fuse translates titles, descriptions, attributes and categories, keeps local units and applies the brand vocabulary from your glossary. Fields it could not translate are flagged rather than shipped empty, and anything that needs a commercial judgement goes to a person.

Yes. A search engine or an AI assistant can only answer as well as the product data it reads. Fuse produces that trusted data layer: it reconciles the sources, checks each field against the rules you set, and publishes a projection each channel can read.

You configure the rules, and the AI works according to them. The checks, the priorities between sources and the strictness are yours. The AI applies that setup to each record. It does not set the rules, and it does not clear a record that fails them.

Two kinds. One kind acts as your data consultants and Fuse experts: they configure the platform with you and manoeuvre it, from the model and the connections to the workflows. The other kind works inside the platform. Each has a specific role, and a limit on the action it may take on the data itself. Not everything is done by an agent. Deterministic rules and checks run alongside them: formats, completeness, allowed values and the business rules you set. An agent does not override those checks.

Yes. You can swap the models the platform provides for your own enterprise models, and the agents then work with the models you choose, under the same rules and checks. Where your model runs, and so where it processes your data, is then your choice.

Teams whose actual project is to replace a PIM. Catalogues of a few dozen references still maintained by hand, where a spreadsheet genuinely is cheaper. And teams looking for a DAM, an order management system or a stock allocation engine, none of which Fuse is. Fuse is the link between your sources and your systems, and it improves the quality of the data on the way through. That is where it fits: in front of the PIM, the ERP, the shop and the channels.

The platform is broader. The same pipeline — ingest, map, reconcile, control, enrich, govern, distribute — applies to supplier, customer and operational records, and that is what Fuse Platform describes. Product data for e-commerce is the configuration we sell today, because that is where the model, the locales and the identity rules already exist.