Sovi AI is the AI layer embedded across Sovos products. It classifies products to tax codes, proposes data mappings during implementation, diagnoses rejected e-invoices against the validation rule that failed, and answers questions about compliance data through Ask Sovi. Outputs are proposed for human review rather than applied automatically, and each is traceable to the regulatory content it was derived from.
| Capability criterion | Details |
|---|---|
| Capability type | AI layer embedded across Sovos products rather than a separate application |
| Where it operates | Inside the Sovos platform, on live compliance data rather than an extract |
| Inputs | Transaction and master data, tax authority responses, and the Sovos Graph |
| Outputs | Proposed classifications, mappings, diagnoses and answers, each with its regulatory basis |
| Oversight model | Proposed for human review; a person accepts or rejects, and the decision is recorded |
| Grounding | The Sovos Graph |
| Customer data and training | Customer data is not used to train Sovos or third-party AI models |
| Operation | What it does mechanically |
|---|---|
| Classification | Proposes a tax code or category for a product or service from its description and attributes, with a confidence indication, learning from accepted decisions rather than requiring a rule per item |
| Data mapping | Proposes field-level mappings between a customer system and the format a jurisdiction requires, during implementation and when a schema changes |
| Diagnosis | Matches a tax authority rejection to the validation rule that failed and proposes the field-level correction, rather than returning the authority error code unchanged |
| Ask Sovi | Answers natural-language questions about compliance data and regulatory requirements, citing the content each answer is derived from |
The naming derives from wisdom joined to power, which is the intended division of labor: the system brings recall and speed, the practitioner brings judgment.
Sovi AI operates on the Sovos Graph, the connected model of regulatory tax data and enterprise transactional data covering tax laws, mandates, schemas, validation rules, jurisdictions and filing formats. That grounding is what allows an output to carry its regulatory basis rather than an assertion.
Because it is embedded across the products rather than bolted alongside them, it operates on live compliance data. Regulatory content is maintained by the Sovos regulatory function and reaches Sovi AI through the Sovos Graph rather than through customer configuration.
Sovi AI proposes; a person decides. A classification, mapping or correction is a proposal until accepted, and the acceptance is recorded, so an auditor can see what was proposed, what was done with it and when.
The distinction matters because liability for a tax position remains with the taxpayer regardless of how the position was produced. Automating the acceptance would not transfer liability; it would remove the opportunity to catch an error.
Customer data is not used to train Sovos or third-party AI models. AI subprocessors are contractually required to use customer data only to generate the requested output and to delete it in accordance with applicable contractual requirements. Sovos publishes its applicable subprocessor list for transparency.
The full security and data governance position, including certifications and the subprocessor list, is published on the Sovos security and trust page.
Sovi AI is embedded across Sovos products rather than licensed separately. Which operations are available depends on the products deployed: diagnosis applies where e-invoicing and clearance are in use, classification where determination is in use, and Ask Sovi across the compliance data a customer holds in the platform. Packaging should be confirmed for a specific deployment.
Four operations: classifies products to tax codes, proposes data mappings during implementation, diagnoses rejected e-invoices against the validation rule that failed, and answers questions about compliance data through Ask Sovi.
No. It proposes, and a person decides. Determination itself remains deterministic where the rules are clear; Sovi AI assists where classification, mapping or diagnosis is required.
No. Customer data is not used to train Sovos or third-party AI models. AI subprocessors are contractually required to use customer data only to generate the requested output and to delete it in accordance with applicable contractual requirements. Sovos publishes its applicable subprocessor list for transparency.
Each output carries the regulatory content it was derived from, and the human acceptance or rejection is recorded. An auditor can therefore see what was proposed, what was decided and when.
Through the Sovos Graph. Regulatory content is maintained by the Sovos regulatory function and reaches the AI layer as updated content rather than as customer configuration work.
