Every feature, in detail.

How it works

On first feed ingestion, our AI reads the first row of your supplier’s file and produces a column mapping from vendor headers to standard fields (sku, inventory_quantity, price, product_title, product_type, size, barcode, product_category). Each mapping comes with a confidence score.

Mapping is cached per feed configuration and reused on every subsequent sync — the LLM is called only once per feed setup, even across thousands of syncs.

Header Drift Remapping

When a supplier changes column names between syncs (e.g., QTY → QTY_AVAIL), SintoData detects the mismatch automatically. In flexible mode, the AI suggests a new mapping and holds the sync for your review. In strict mode, any drift on a required field stops the sync immediately. A 1-hour cooldown between auto-remap attempts prevents runaway LLM costs.

How it works

AI extracts color, size, and material from product titles and writes them as structured Shopify metafields (custom.color, custom.size, custom.material). In automatic mode (the default), in-vocab values are written to your store immediately after each feed sync — no review step.

Vocabulary Guardrail

Extracted values are validated against a hardcoded allow-list vocabulary. Values outside the vocabulary land in a flagged_attributes field and are never auto-written — even in automatic mode, off-vocab values stay flagged and require your review. This prevents AI-generated nonsense from reaching your storefront filters.

Metafield Auto-Creation

SintoData automatically creates the metafield definitions for color, size, and material on first scan — no manual admin steps required. Your Shopify storefront filters and search work out of the box.

How it works

AI translates your product titles and descriptions into any Shopify Markets locale. In automatic mode (the default), translations are written to Shopify via the native translationsRegister GraphQL mutation after every feed sync — only for locale fields that are currently empty. If you already have a translation (whether you wrote it yourself or paid a translator), it is never overwritten. Translations are keyed by translatableContentDigest to prevent stale overwrites. An HTML-balance guardrail blocks any translation that introduces broken tags, surfacing it for your review instead.

Shopify Markets Integration

Translations ride on Shopify’s Markets infrastructure. Once written, they appear in your admin under the product’s translation interface and serve to shoppers based on their locale. Every translation is unique per product + locale pair — re-running with the same locale updates the existing entry rather than creating duplicates.

How it works

AI analyzes each product’s title and description, then suggests a Shopify product type. In automatic mode (the default), the product type is written to your store only when the field is empty — it fills what you left blank and never overwrites a product type you already set.

Two guardrails every write passes

• Gap-fill enforcement: Before writing, SintoData re-reads the product’s current product type from Shopify. If the field already has a value, the write is skipped — the AI never replaces your work.
• Confidence threshold of 0.5: Suggestions below this threshold are discarded and the row is marked “uncategorized” for manual review. No weak matches make it to your approval queue.

Approval writes one mutation

Approving a match sets the product type via productUpdate. The gap-fill check runs on the approve path too, so even an explicit approval cannot overwrite a product type you already set.

How it works

AI generates page titles and meta descriptions for every product — the metadata that search engines see, not the product listing text your customers see. Titles target 60 characters, descriptions target 160 characters, optimized for click-through from Google and Bing results. SEO Metadata defaults to automatic mode and only writes to empty seo.title and seo.description fields — if you already have SEO metadata set (manually or via another app), it is never overwritten. Your storefront copy is also never modified.

Never touches your product copy

Unlike rewriting product descriptions — which risks overwriting copy you spent time curating — SEO metadata operates on the search engine listing fields only. Your product title and description on the storefront are never modified. The metadata is written to Shopify’s metafields_global_title_tag and metafields_global_description_tag fields, which only appear in search engine results and browser tabs.

How it works

A deterministic (non-LLM) safety net that runs inside the standard price-sync path. For every price change, it compares the new price against the last-known price using a ratio-based check: if the new price is more than 3× the old price (or less than 1/3), or the absolute change exceeds $2.00, the price is flagged as anomalous and held back from being written to Shopify.

Per-feed threshold override

Each feed can override the default ratio threshold — if your supplier typically has large price swings, you can raise it. The minimal absolute delta is fixed at $2.00 to avoid flagging penny-rounding noise.

Approve or dismiss

Anomalous prices appear in your Pricing Alerts tab. Approving applies the held-back price exactly as the supplier sent it. Dismissing discards it permanently. The last-known price is never updated on anomaly, so the same suspicious price re-triggers on the next sync — you won’t miss it.

How it works

A daily shop-wide catalog scan using difflib.SequenceMatcher to find likely duplicate product listings based on title similarity. The similarity threshold is calibrated at 0.96 (raised from 0.85 after benchmarking found false positives like “Wireless Mouse” vs “Wireless Mouse Pro” at 0.875).

Pure read-only — zero risk

This feature makes zero Shopify writes, ever. It surfaces candidate duplicate pairs for your review. You fix them yourself in your Shopify admin. The only operations available are “view pending” and “dismiss.” There is no approve button — there’s nothing to write.

Shop-wide scope

Unlike other AI features that operate per feed, duplicate detection scans your entire catalog. Duplicates that span two different supplier feeds are caught. This is a catalog-wide hygiene scan, not a feed-level operation.

Feature-level toggles

Each AI feature can be set to automatic or manual mode per feed. All four generative features — Attribute Extraction, Translation, Category Matching, and SEO Metadata — default to automatic. Want manual review on your main supplier but automatic on a secondary one? Want translation automatic but attribute extraction manual? Configure it per feed, per feature, in the feed setup modal.

Guardrails run in both modes

Automatic mode doesn't mean blind trust. Every guardrail described above — the vocabulary allow-list, HTML-balance check, gap-fill enforcement, and confidence threshold — runs identically in automatic and manual mode. In automatic mode, clean output is written immediately; flagged output is held for your review. In manual mode, everything is held for your review. Switching from manual to automatic never removes a safety check.

Per-feed product attribution

Every AI feature row knows which supplier feed it came from. When you filter pending reviews by feed, you see only the products relevant to that supplier. Automatic scans run per feed after every sync — a feed only touches its own products, never another feed's catalog. Products not yet attributed to a feed are included after that feed's next sync.

Your catalog, your rules

SintoData never creates products from supplier feed data. Unmatched feed SKUs are skipped and reported as errors — no productCreate, no auto-generated listings. And it never writes your primary-language product title or description — the copy your shoppers see in your store's main language is always yours. Every AI write either fills an empty field or writes into a separate namespace (its own metafields, search-engine tags, or a non-primary locale's translation slot).