Weita Supplier Onboarding Workflow: Wholesale Digital Transformation

Weita AG is a Swiss wholesaler in healthcare, cleaning, safety, hygiene and non-food. Supplier input arrives as PDFs, pasted email threads, product images and Excel spreadsheets, and one product often spans all four formats at once. The output has to be a fully attributed, compliant PIM product whose category-required labels are demonstrably satisfied — a real swiss wholesale pim implementation problem, not a content-tagging exercise.

The previous flow leaned on email and manual entry across Content Management, Marketing and Support. Provenance was thin, compliance evidence lived in a spreadsheet next to the system of record, and a missed label only surfaced downstream. The brief was to replace that with a deterministic master data management workflow that an auditor can read.

A job queue tells you what ran. A state machine tells you where a product is, who owns the next move and which transitions are legal from here. For a regulated wholesale digital transformation that distinction is the audit trail.

Every product walks an explicit state list: Initiation, CategorizationProcessing, CategorizationReady, ExtractionProcessing, CM, Marketing, Support, Finalized — with dedicated Failed states for both AI phases. Each state knows its legal next transitions and the role that owns it. Nothing slips sideways because the workflow refuses any transition that is not on the list.

Two deployables. The backend is a Laravel 12 modular monolith with nine business modules — Auth, User, Category, Attribute, Label, DocumentType, Workflow, AI and Products — on PostgreSQL 18, Redis 7 and Horizon for queue supervisors. The frontend is Next.js 13 App Router with PrimeReact and i18next.

AI never blocks an HTTP request. Each supplier file runs through async jobs — ProcessWorkflowFileWithMistralJob, RunWorkflowCategorizationJob, RunWorkflowExtractionJob — with Mistral OCR called in two passes per file (raw plus structured) and OpenAI in JSON mode for both categorization and extraction. Prompts live in a database-backed prompt registry keyed by a PromptKey enum, so a prompt change ships without a redeploy. Sanctum handles auth, Flysystem talks to three S3 disks for supplier docs, product docs and product images. Local development runs on Laravel Sail.

Compliant document automation is not a matter of one good model call — it is a chain of human decisions logged against the same record. The workflow enforces who can do what at each step.

• Content Manager owns the CM state: confirms category, fills attributes the AI did not resolve.
• Marketing owns the Marketing state: tone, multilingual descriptions, channel-ready copy.
• Support owns the Support state: final checks before publication.
• Finalized is terminal: the product is live in the PIM and the workflow row is the audit record.

Both AI phases have their own Failed states, with retry-ready details columns instead of corrupt records. An auditor reading the workflow row sees the full path, the role on every transition and the AI evidence pack that informed each call.

Swiss wholesale categories carry required labels — safety pictograms, hazard statements, hygiene markings, regulatory references. The Label module declares which labels each category requires; the compliance engine derives live satisfaction from the actual product attributes for that category.

If a category demands a label and the supplier pack does not evidence it, the product cannot advance to Finalized. The AI context builder also assembles the evidence pack for each model call, so categorization and extraction always see the same attribute and label context a reviewer would. This is what makes the compliance automation defensible rather than decorative.

What changed: ad-hoc email and manual entry replaced by a deterministic, auditable workflow; AI-assisted categorization and attribute extraction with prompt-registry governance; enforced role ownership at every step; live compliance evidence tracked against category-required labels; AI failures contained in dedicated Failed states instead of polluting good records.

This is the second SAPIENTROQ engagement for Weita. We previously built the API-gateway Interim-CIO engagement for the same client; this workflow project is a separate, later scope. If you are a Head of Digitalisation or Head of Master Data at a Swiss wholesaler and you recognise the supplier-onboarding pain, read the FAQ below or book a discovery call to plan your supplier onboarding workflow.