Everything you want to know

Yes. First we think along with you and set up a test on your real catalogue together, up to 50 products, so you judge the quality on your real products instead of a sandbox with fake data. Nothing goes live before you give the go-ahead. If you decide not to continue, we delete your data.

In days, not months. There is no implementation project: the connectors already exist, so you connect your shop and get going straight away. A traditional PIM soon counts on around three months; here you are well under that. Exactly how long it takes depends on the size of your catalogue.

Not for the day-to-day work. Mapping, enriching, reviewing and pushing you do in the interface, without a line of code. The only technical step is a one-time install of the SyncRefine connector, a plugin or module, on your shop. Feed imports only need the right connection details.

For a shop connector you install our plugin or module once and give us the connection details; after that a discovery step does the rest. If you work with files or feeds, we need the right login details: FTP, SFTP, email or API. It helps if you know which suppliers and channels you want to connect. The setup itself we do together during the trial.

Absolutely. During the trial and onboarding we connect your shop and sources and set everything up together, and we personally sit in on the first call. You keep control, we take most of the work off your hands. The exact steps of a migration we align per situation.

We connect to the big e-commerce platforms and PIMs, from WooCommerce and Magento to Shopify, Lightspeed and Akeneo. The most-used connectors are ready to go, and if your platform works a little differently, we connect it as soon as we start together. Tell us what you run on, and we make sure your data flows both ways.

Yes, we are happy to solve that together. Some connectors are ready-made, others we finish the moment you come on board. That way you get exactly the connection you need and we extend our platform further. Win-win, and you do not have to wait until something is 'done' before you start.

Both. The shop and PIM connectors read in your product data and can send enriched data back. The reading in is read-only, so your source never changes during an import. Changes reach your shop only through the explicit push step, not somewhere along the way.

Yes. For commonly used systems like AFAS, Dynamics and Exact we arrange the connector, and if your system is not on the list yet, we connect it as soon as you become a customer. As long as your data comes out somewhere, a file, a feed or an API, we bring it in via CSV, XML or XLSX over FTP, SFTP, email or REST API.

CSV, TSV, Excel (XLS, XLSX, XLSM, XLSB and ODS), JSON and XML. Excel, JSON and XML we first convert internally to CSV before processing begins, so everything then runs through the pipeline the same way.

In all the usual ways: FTP and the secure FTPS, SFTP, email attachments, a public HTTP URL or a REST API endpoint (JSON with pagination, converted to CSV automatically). Uploading files directly in the app is also possible. If your source delivers in another way, we connect that as soon as we start together.

As many as you need. You connect your suppliers and sources one by one, and they all flow together into one central catalogue. Every source gets its own mapping and sync, so you expand calmly as you grow. Large numbers we look at together, because every source really is recurring work: mapping, normalising, syncing and maintaining.

Yes. You connect multiple webshops and they all read and push from the same central catalogue. That way you keep one source of truth instead of separate spreadsheets per shop, and your number of channels grows along without duplicate work.

No, only the changes. Using checksums the change detection sees which tables and products have changed, and compares only those. You set the schedule yourself, from every five minutes to every hour or every few hours. Starting manually is always possible.

Your mapping is fixed per source. If a supplier renames columns or builds the file differently, you adjust the mapping without redoing everything. When reading in we profile the columns, so deviations stand out immediately. Your original source data is always kept underneath, so a wrong import you roll back.

The connector only reads and, using checksums, fetches only the changes per sync, not your whole catalogue every time. Changes reach your shop only at the separate push step that you decide on. The load on your shop therefore stays limited, and during reading in nothing happens that you have not approved.

The AI does four things: translate a field into another language, clean up or rewrite text in its own language, create new text such as descriptions or SEO based on your existing attributes, and read the product photo to recognise attributes. There are no more than those four operations. It is not an open black box where anything can happen.

Into more than forty languages, including Dutch, English, German, French, Spanish, Italian, Polish, Chinese, Japanese and Arabic, and it recognises the source language itself. For large catalogues automatic translation works well, though it sometimes misses the nuance. That is why you keep control: you have it reviewed and approved per language before it goes live.

Leading models from Google (Gemini), in three tiers. A fast and economical model for text, a model with image recognition for reading photos, and a heavier model for tasks where the quality is critical. The task determines which tier runs.

It really looks at the pixels of the photo, up to three photos per product. That feature is not there for nothing: an earlier approach with text only never saw the photo and therefore made up materials and colours.

No, that is deliberately shut down. The instructions forbid making up specifications, sizes, materials or claims. If the model is not sure about something in a photo, it answers '__unknown__', nothing is written and the item stays in the queue. When translating, SKUs, numbers, units and brand names stay as they are.

You choose when the AI runs: once on save, continuously on changes, or only when you approve a review. Manual edits and locked fields always win over the AI and stay in place on a new import. So if you correct an AI value, it is not quietly overwritten.

Yes. You run an enrichment on a small selection first and see the result on your own products before you let it loose on the whole catalogue. That way you judge the quality yourself, and you decide when it may go live.

Yes. A no-code rule engine compiles operations to SQL: transformations on text, numbers, dates and arrays, text extraction (also with regex) and category assignment. You assemble automations from fifteen ready-made building blocks, without a line of code. This uses no AI consumption.

Automations run at four moments: continuously on changes, on approving a review, once on save, or manually. A free cron schedule per automation is not there, so we do not promise that. Reading in sources does run on a schedule that you set, from every five minutes to several hours.

Access is role-based (Owner, Admin, Editor, Viewer) and per user you switch specific permissions on or off, such as publishing automations. That way you decide who may run enrichment and who only watches along. Every action goes into the audit log per company.

Instructions you give the AI yourself first pass through a filter. That blocks attempts like 'forget previous instructions' and hidden code or script tags, so such an instruction does not reach the model, your data or the billing. The filter applies to all four AI operations.

By an exact match on one of five identifier columns: SKU, EAN, MPN, GTIN or barcode. A product often has several, so which key takes priority you set per supplier. Empty or nonsense identifiers such as zeros, 'null' or 'n/a' we throw out, otherwise separate products wrongly end up on one pile. We match on the identifier itself, not on how a name happens to look.

They simply stay separate. Merging happens on an exact match of SKU, EAN, MPN, GTIN or barcode, and empty or nonsense identifiers (zeros, 'null', 'n/a') we deliberately exclude. If a product has no usable identifier, we do not merge it automatically. It waits until you link it yourself.

The incoming supplier value, with the existing value as fallback. That way an empty value never wipes out something good that was already there. If two rows point to the same product, we pick a fixed winner and record the rest as a conflict; doubtful cases can go to a review queue. It works on priority, not on a weighted trust score.

Yes. Per product we record which source is the owner, and every staging row keeps the source id plus the identifier used, so you always see where a value comes from. A full log per individual field, showing exactly who wrote which field when, we build out further as customers need it.

Yes. Alongside the dozens of fixed fields you set up unlimited custom fields for your own attributes. That way the structure follows your assortment, and not the other way around.

We do. Per field we work out the fill rate, that is the share of non-empty values, straight in the database, and the average total score we keep. Note what this does and does not say: it measures presence and coverage, not whether the content is actually correct.

Certainly. New items and doubtful cases can go into a review queue where you approve, adjust or reject. Only what you approve goes into your central record. Rejected items we block, and those cannot displace an already approved item, so a mistake does not leak to your shop unnoticed.

No. A hard quality gate stops the sync if required fields like SKU, name and price are not complete enough; for recommended fields you only get a warning. Empty catalogues we block by default. If you deliberately make an exception, we flag that and record it in an audit log.

When pushing we look at changes that are unusually large in proportion: mass deletions, or suddenly a lot of new or changed products. Just before the push we also check the types and formats, such as a numeric price, SKU and name within the length, and choice fields within the allowed list. What happens then depends on the guard level: block until your approval, warn, or let through. That way we intercept bulk deviations before they hit your shop.

Yes. Every photo that comes in we convert to WebP at quality 85, we scale it back to a maximum of 2048 pixels (keeping the ratio, never enlarged) and rotate it right automatically. Your pages get lighter for it, and a fast page counts towards Google's page-experience signals. It does no more than count, though: it contributes, but does not guarantee a higher spot.

That is your choice. In the forwarding mode the original photos stay at your source and our image proxy (imgproxy, with signed URLs) sizes them on the way; that way even a large push runs in seconds. If you choose the storage mode, we fetch the photos, convert them to WebP and keep them in our own EU storage. Your shop then gets pre-prepared WebP files in fixed sizes (300, 600, 1200 and 2048 pixels), so the resizing and the bandwidth no longer weigh on your shop server.

A failed photo never makes the push of the product crash. The product is simply created or updated, and we note that the image was skipped, with the reason. Junk photos that are too small, almost flat, broken or larger than 50MB we already filter out at the check.

Yes. Every normalised photo gets a content hash (SHA-256), and the storage key comes from that hash. Before we upload we check whether the file is already there, so identical bytes we keep only once, even if they recur across hundreds of products.

The image processing now focuses on product photos: converting to WebP, sizing and deduplicating. A full media library for videos, manuals or other documents is not standard in it at the moment. If you need that, for instance for Bol, let us know; then we look together at what is possible.

For marketplaces and comparison sites we have ready-made feed templates, for example Bol.com (NL/BE), Google Shopping / Merchant Center, Amazon EU (DE/NL/FR/IT/ES) and Vergelijk.nl. That is just a selection of the channels; new ones are added all the time, and we build a feed for any other channel on request. If you go straight to a webshop, there are push connectors with WooCommerce, Magento and Akeneo.

You get a secure feed URL with a token that cannot be guessed. The marketplace or engine fetches it itself, in CSV, JSON, XML or TSV. Separator, encoding and header rows you set yourself, and the feed always shows your last enriched values.

Yes. Per channel you set field mappings: you rename a source field to the output field the channel expects. The push connectors in turn map from your central catalogue to the shop fields, so you decide the exact output schema per destination. SEO fields too, like meta title, meta description and URL slug, land in the right place per platform this way.

That can be done per channel. You choose whether everything goes along or only a selection on conditions that you set yourself. If you want to send, for instance, only products with enough margin or with complete attributes to a marketplace, you filter on that. Each channel gets its own selection, field mapping and format this way, all from the same source.

Multilingualism you arrange with the translate operation. It fills a field per language that you map to the right channel. For multiple currencies or separate price structures per channel it depends on your setup; a fixed multi-currency feature we do not promise without having checked it. Let us know which channels and currencies you need, and we tell you honestly what does and does not work.

Bol only lets an offer go live with a valid EAN and the required content included. For a brand-new product a GPC classification code is added to that. With a clean catalogue where the attributes are complete and linked, you prevent offers from getting stuck.

The moment a feed is requested, it lays your manual edits and the AI enrichment over the central catalogue. That way your change is immediately in the feed, without you having to wait for a full push to your shop.

Yes, in the EU. Your files sit on Hetzner Object Storage in European data centres, by default Falkenstein or Nuremberg in Germany, or Helsinki in Finland. The databases we set up separately per customer. If you want a firm commitment about the location of every database, we lay that down in the agreement.

Yours. Your data sits in your own, per-customer isolated databases and storage; SyncRefine sits as an in-between layer between your suppliers and your channels. If you cancel, you export your complete enriched catalogue and your shop keeps running on the last pushed data. How long we keep your data after that we agree together.

Every customer gets its own PostgreSQL database, its own analytical database, its own Redis database and its own subdomain. In storage all your data falls under its own key prefix, enforced through a strict key structure. And within the database every record is linked to your company.

Yes. We store login details encrypted with AES-256-GCM. Per record we use a unique IV and authentication tag, with a 32-byte server key and a version field for key rotation.

You secure your account with two-factor authentication (TOTP with hashed backup codes) or with passkeys and WebAuthn. That security setting is mandatory, and after too many failed attempts the account locks; we also record the login IP. Access is role-based, with Owner, Admin, Editor and Viewer, and the permissions you adjust per user. Every action goes into an audit log per company.

The data sits in the EU, is isolated per customer, and the storage and database relationships are set up so that we can delete a complete customer environment. SyncRefine itself is based in the EU, as SyncRefine OÜ in Tallinn, Estonia. A formal deletion process and the data processing agreement (DPA) we arrange together. Formal certifications like ISO 27001 or SOC 2 we do not have at the moment; we would rather tell you exactly how we protect your data than show a certificate we do not have yet.

For file storage we use Hetzner in EU data centres, for the AI enrichment Google Gemini via business API calls. The full list of sub-processors and the associated agreements is in the data processing agreement. A party we do not use, we would rather not name here.

Only the field values you have enriched go into the call, for instance the text that gets translated or the photo that gets read. A full database or account information we do not send along. The enrichment runs via business API calls to Google Gemini. What exactly happens with that data and whether it is used for model training we lay down in the data processing agreement; we promise nothing here that we cannot back up.

Storage and databases are isolated per customer, and back-ups and management run in a separate system space. The exact frequency, retention period and recovery times we lay down per agreement, instead of promising something here that we cannot show. A formal uptime guarantee belongs with Enterprise.

There is a ready-made template to record EUDR data, with raw material and geolocation as validated GeoJSON, plus the GPSR fields. Think of it as help with collecting and structuring the required attributes, not as an automatic declaration to the regulator. For large traders EUDR applies from 30 December 2026.

Every plan includes a set number of hours in which we think and build along with you personally, and ordinary help with questions and updates is always included. The team helps you in Dutch. A formal SLA on uptime or response time is not standard in the plans; that we agree with you upfront at Enterprise.

Yes, the team is Dutch-speaking. In the first call someone from our team sits in themselves and helps you get going in a tailored way. No obligation, and without a sales pitch.

We are open about it: SyncRefine is a young, founder-led platform that we are hardening in practice. Our pilot runs live with over 70,000 products from 40 suppliers in sync, and tens of thousands of review decisions have already gone through it. Inflated customer figures or ROI promises we do not do. There are still a few partner spots to build out together.