AI and Automation

AI Metafield Mapping: Matching Shopify Metafields to Odoo Fields

How Suggest with AI on the Field Mapping page matches Shopify metafield definitions to writable, type-compatible Odoo fields on orders, contacts, and products, and why a 60 percent floor and a Save button still stand between the suggestion and your data.

The Quick Take

Metafields are where Shopify stores the data that does not fit the standard product, order, or customer. Engraving text, a care label, a VAT number, a delivery instruction, a sales agent, a warranty period. The Field Mapping page is where you say which Odoo field each one lands in. The hard part is not the mapping screen. It is knowing Odoo's field names: a merchant does not know that "Delivery instructions" is x_studio_delivery_notes on the sales order or that the VAT number on a contact is simply vat.

Synco's "Suggest with AI" on the Field Mapping page builds the candidate list from your own Odoo fields, filtered to what the connector could actually write, and asks an AI model to pick the best field for each unmapped metafield. You review and press Save. This guide covers what gets filtered, what the model sees, and what to check.

Related reading: Shopify Odoo metafield sync page, metafield sync guide, product metafields to Odoo, customer metafields to Odoo, order metafields to Odoo, how AI helps sync Shopify and Odoo.

What A Mapping Has To Get Right

A metafield mapping is a pair: a Shopify metafield definition (namespace, key, type, owner) and an Odoo field on one model. For the sync to work, three things must hold.

  • The Odoo field must exist on the right model. Order metafields map to the sales order; customer metafields map to the contact; product and variant metafields map to the product record, which also exposes the template's columns.
  • The Odoo field must be writable. Computed fields, related fields without inverse, and fields the integration user cannot write will fail at sync time with an access or validation error.
  • The types must be compatible. A Shopify date metafield into an Odoo char field is fine; a Shopify multi-line text into an Odoo integer is not. A metaobject reference into a many2one is fine only if the target can be resolved by name.

Synco already checks all three when you pick a field by hand. The AI suggestion does the same checks first, so the model only ever chooses among fields that would pass.

What Happens When You Click Suggest

The work runs as a queued job because reading Odoo's field list for a model can take a few seconds. The page polls and fills in when it is done.

1. Collect the unmapped definitions. Up to 60 metafield definitions on the current tab (order, customer, or product) that have no mapping yet. Definitions already mapped are left alone. 2. Read Odoo's fields for the target model. Every field on sale.order, res.partner, or product.product with its label, type, help text, and whether it is writable. 3. Filter per definition. For each metafield, keep only Odoo fields that are writable, type-compatible with the metafield's type, and not already used by another mapping. Then rank by word overlap between the metafield's name, key, and description and the Odoo field's label and help text. The top 15 go forward. 4. Ask the model once per definition. TypeSafe's Jev is given the metafield and the shortlist and asked which field, or none. It returns a probability per option. 5. Fill the picker above the floor. Suggestions at or above 60 percent are placed in the Odoo field picker and labelled with the probability. Below that, the row is listed as "no match" and left for you.

Nothing is saved. The page is in the same unsaved state as a manual edit. Save or discard.

Examples Of What It Finds

  • A customer metafield custom.vat_number (single line text) suggested to the contact's VAT field.
  • An order metafield checkout.delivery_instructions (multi-line text) suggested to a Studio field labelled Delivery notes on the sales order.
  • A product metafield specs.warranty_months (integer) suggested to a custom integer field on the product; never to a char field named "Warranty text", because the type check removes it.
  • A customer metafield sales.agent (metaobject reference) suggested to the contact's Salesperson field. At sync time the metaobject's display name is matched to exactly one Odoo user by name; if there is no single match, the field is left blank and the job says why.

Where The Floor Does Its Work

Sixty percent is the line below which the model's pick is not shown at all. It stops two failure modes:

  • Generic labels. Odoo has many fields called "Description", "Notes", or "Reference". A metafield called "Notes" cannot be confidently assigned to one of them, and the model should not pretend it can.
  • Studio fields with no help text. A field named x_studio_field_a1b2 tells the model nothing. It should not be guessed.

In both cases the right move is in Odoo: give the field a proper label and help text, then click Suggest again. The connector cannot know what a field means better than its own label does.

What It Will Not Do

  • Create fields in Odoo. Suggestions only point at fields that exist.
  • Map Odoo to Shopify. The Field Mapping tabs covered here send Shopify metafields into Odoo. The reverse direction has its own configuration.
  • Change a mapping you made. Only unmapped definitions are considered.
  • Run during sync. Sync reads the saved mapping table, validates the value against the field type, and writes.

A Short Check Before Save

For each suggested row, open the Odoo record once and look at the field. If it is the one your team actually reads, keep it. If it is a lookalike nobody uses, change it. Then sync one record that carries that metafield and open it in Odoo. Seeing the value in the field is the whole test.

Frequently Asked Questions

Can the AI map a metafield to a field it cannot write? No. Candidates are filtered to writable, type-compatible fields before the AI sees them.

Which Odoo models can metafields be mapped to? Order metafields to the sales order, customer metafields to the contact, product and variant metafields to the product.

Does suggesting a mapping change existing mappings? No. Only unmapped definitions are considered, and fields already used by another mapping are excluded.

FAQ

Common questions

Can the AI map a metafield to a field it cannot write?

No. Candidates are filtered to Odoo fields that are writable for the connected user and type-compatible with the metafield before the AI sees them. A read-only or computed field never appears as an option.

Which Odoo models can metafields be mapped to?

Order metafields to the sales order, customer metafields to the contact, and product or variant metafields to the product, which also carries the template fields.

Does suggesting a mapping change existing mappings?

No. Only metafield definitions with no mapping yet are considered, and Odoo fields already used by another mapping are excluded from the candidates.

Keep reading

Related guides