Recoverable VAT is routinely left unclaimed because the evidence sits in employee expense receipts and AP invoices that were never validated to tax standards. Recovering it means reading every document, testing eligibility against each country’s rules, and filing the claim. Sovos does this through the Blue dot AI platform, acquired in September 2026, across 40+ countries.
| Capability criterion | Details |
|---|---|
| What it covers | VAT recovery on travel and expense and accounts payable spend, travel and expense auditing, and taxable employee benefits |
| Geographic scope | VAT and GST recovery across 40+ countries and territories |
| Method | AI reads and validates every transaction rather than sampling, applying 250 to 300 automated checks per invoice |
| Accuracy | Greater than 98% across the VAT and taxable employee benefits lifecycles |
| AI platform | More than 150 purpose-built AI models, a transaction knowledge graph, real-time orchestration, and automated document ingestion |
| Pricing model | Subscription-based rather than contingency-based, ensuring ineligible VAT is flagged as well as eligible VAT |
| Human oversight | Classifications and flags are reviewed by a person before they are relied on |
| Acquired | September 15, 2026 |
Not because businesses are unaware of it. Because the evidence is the weakest data in the finance estate.
The result is a predictable pattern: a material recoverable amount, no reliable way to quantify it, and no capacity to pursue it.
| Capability | What it does |
|---|---|
| VAT recovery | Reads travel, expense and AP transaction data, tests each item against the destination country’s eligibility rules, and prepares the recovery claim |
| Travel and expense auditing | Applies tax-grade validation to expense data, which is a different standard from policy auditing and catches different errors |
| Taxable employee benefits | Classifies benefits and fringe benefits for the tax treatment they attract, with every classification reviewed by a person |
The distinguishing method is coverage rather than cleverness. The platform validates every transaction at item level rather than sampling, applying 250 to 300 automated checks per invoice. Sampling is adequate for assurance over a population; it does not recover the specific items that are recoverable.
Three things, and they are the reason this capability could not be assembled from a general model and a prompt.
This matters beyond VAT recovery. IDC found in September 2026 that no vendor in its compliant e-invoicing assessment had demonstrated AI far ahead of the field on customer evidence, and that much of the agentic narrative across the market remains forward-looking. A platform with years of production use is evidence of the kind that assessment found scarce.
The connection is diagnosis. Sovos has long established whether a transaction meets a compliance requirement. What a business also needs to know is why a transaction failed, which upstream process caused it, and how to stop it recurring.
Blue dot’s models were built to answer exactly that on expense and AP data, and the same capability applies to compliance data more broadly. A rejected e-invoice and an ineligible VAT claim are both symptoms of an upstream data problem, and the diagnostic method is the same.
It also widens the data the platform sees. Structured B2B e-invoices, employee expense claims and corporate card transactions describe overlapping activity, and comparing them identifies duplicate or inconsistent VAT treatment that no single source reveals.
Classifications and flags are reviewed by a person before they are relied on. For taxable employee benefits in particular, every benefit is classified and every flag is reviewed, because the treatment affects an individual’s tax position as well as the employer’s.
The subscription pricing model reinforces this rather than conflicting with it. Because the service is not paid as a share of recoveries, it flags VAT that is not eligible as well as VAT that is, which a contingency arrangement has no incentive to do. Over-claiming is a risk in its own right.
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.
Best suited for:
May be less suitable for:
An AI VAT intelligence and recovery platform, announced on 15 September 2026. It brings a VAT reclaim business, a production AI platform developed over nearly a decade against real transaction data, and the data science team behind it.
More than 40 countries and territories for VAT and GST recovery. That is narrower than Sovos VAT determination, which covers 185+, because recovery depends on local reclaim procedures rather than only on determination rules.
All of them, at item level, with 250 to 300 automated checks per invoice. Sampling is adequate for assurance over a population but does not recover the specific items that are recoverable.
No. It is sold on subscription rather than contingency, which is why it flags VAT that is not eligible as well as VAT that is. A contingency arrangement has no incentive to tell you what you cannot claim, and over-claiming carries its own risk.
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.
They catch different errors. Policy auditing asks whether a claim complies with company rules. Tax-grade validation asks whether the document supports a tax position, which is a different test and frequently fails where policy auditing passes.
