Know your gaps before the mandate does
A scored report your auditor can follow, with fixes ranked by owner and effort — without a months-long consulting project. Enough to decide whether you are onboarding in weeks or planning a programme.
ClayDesk Assure · powered by GoRoute
Every invoice looks compliant until a validation engine reads it.
ClayDesk Assure reads a sample of your real invoice data — an ERP export, a spreadsheet, even plain PDF invoices — through GoRoute's standards-pinned validation engine, and returns a readiness score, the exact gaps that would block compliance, and a remediation report your auditor can follow. Minutes of processing plus a human review, not weeks of workshops.
Sample data under NDAConsultant-reviewed reportNo integration to start
Readiness report · 60-invoice sampleIllustrative · synthetic data
62/ 100
Gaps to close
- BR-CO-14VAT total does not recompute from the category amounts14 / 60
- BR-S-02Seller VAT registration missing on standard-rated invoices9 / 60
- BR-CL-04Currency code not from the ISO 4217 list6 / 60
Verdicts · GoRoute deterministic engine · EN 16931 + PINT OM 1.0.1
Narrative · AI-drafted · consultant-reviewed before release
0–100
Readiness score, with every finding traceable to the rule identifier that produced it
5 layers
Schema → business rules → official code lists → CEN/Peppol Schematron → country rules
Any format
PDF invoices, CSV or Excel exports, UBL XML — bring what you have, no template to fill
1 reviewer
A consultant reads every report before it reaches you — AI drafts, a human signs
Why readiness fails
The data looks fine on the PDF. Then a machine reads it.
Mandates arrive on fixed dates. Businesses discover their data problems at go-live — when a rejected invoice is a cash-flow event and fixing the cause is most expensive.
The usual answer is a readiness engagement: weeks of workshops, a spreadsheet checklist, and an opinion. Slow, costly, and inconsistent from one consultant to the next. It tells you what people believe about the data, not what the data does under a validator.
The uncomfortable truth for most finance teams is that their invoices look compliant on a PDF and fail structured validation: tax registrations missing, VAT that does not recompute, fields the ERP never captured. Nobody knows until a machine checks — so we start by having a machine check.
"Every invoice looks compliant until a validation engine reads it."
What actually fails first
- Missing tax registrations — the seller or buyer VAT identifier the mandate requires is simply not on the record
- VAT that does not recompute — line, category, and document totals disagree once a rule engine adds them up
- Codes that are not the official list — currencies, units, and tax categories typed by hand rather than drawn from the published code lists
- Identifiers in the wrong shape — dates, scheme prefixes, and party identifiers formatted for humans, not for the standard
- Fields the ERP never captured — buyer references, delivery details, and country-specific elements that no template ever asked for
How it works
Five steps from a messy export to a scored report.
Two things stay fixed throughout: the verdicts always come from the deterministic engine, and wherever AI helps, a human confirms before anything moves on.
- Deterministic engine
- AI assists
- ✓You confirm
-
Step 1You
Intake
A short questionnaire: ERP in use, invoice volumes, registrations, integration preference. Save and resume.
15 minutes · no IT involvement -
Step 2You
Upload anything
CSV or Excel exports, UBL XML, or ordinary PDF invoices. No template to fill, no IT project to start.
tens to a few hundred invoices -
Step 3AIYou confirm
Understand the data
Columns are mapped to one canonical invoice model. AI proposes mappings for unfamiliar headers and extracts structured data from PDFs — every mapping visible and confirmed by a person.
values copied verbatim · formats normalised -
Step 4Engine
Deterministic scan
Every invoice runs the five-layer validation gate — schema, business rules, official code lists, CEN/Peppol Schematron, country rules. Real rule identifiers, reproducible verdicts, a 0–100 score.
stateless dry run · nothing persisted -
Step 5AI draftsConsultant signs
Report
Executive summary, findings in plain business language with an owner and effort per fix, the field-mapping table, and a recommended onboarding path — as a branded PDF.
verdicts quoted from the engine only
Under the hood
Five layers, version-pinned. The same answer every time.
Whatever comes in — CSV columns, a spreadsheet, a PDF extraction — is normalised into one canonical invoice model, then rendered and pushed down through the gate. The engine is the one inside GoRoute's Peppol-certified e-invoicing platform, not a marketing replica of the rules.
InCanonical invoice → compliant UBL
-
1
XSDUBL 2.1 schema
Structure
Is the document well-formed and every element the shape the standard expects?
catches: dates not in ISO form, missing mandatory elements -
2
Business rulesarithmetic & presence
Arithmetic
Do the numbers agree with each other — line totals, VAT category amounts, document totals?
e.g. BR-CO-14 · VAT total ≠ sum of category amounts -
3
Code listsofficial eDEC lists
Codes
Are currencies, units, tax categories, and scheme identifiers drawn from the published lists?
e.g. BR-CL-04 · currency not in ISO 4217 -
4
SchematronCEN EN 16931 + Peppol BIS
Semantics
The full semantic rule set of the European standard and the Peppol business rules, executed as compiled, checksummed XSLT.
e.g. BR-S-02 · seller VAT identifier required for standard-rated supply -
5
Country rulese.g. PINT OM 1.0.1
Jurisdiction
The country specification on top — Oman's PINT OM rules and Fawtara-specific requirements today; further packs as specifications stabilise.
catches: Oman scheme 0248 identifiers, OM-prefixed VATIN, tax-data-document fields
every failure recorded, run continues
Out
Findings with rule IDs → score 0–100 →
Version-pinned · checksummed rule artefacts
Reproducible · same file, same verdict
Stateless · dry run, nothing persisted
Traceable · rule ID on every finding
The trust architecture
The AI never decides whether your invoice is compliant. The rules do.
Tax authorities and auditors do not accept probabilistic verdicts, and neither do we. So the AI has a defined, bounded job — and one line it is not allowed to cross.
- AI assists
- Deterministic engine
- ✓Human confirms
Allowed · AI assists
Understands and explains
- Runs the intake conversation and profiles the files you actually sentbounded to the questionnaire
- Proposes how messy column headers map to the canonical invoice modelyou confirm every mapping
- Extracts structured invoice data from PDFs, with a confidence record per sectionvalues copied verbatim · labelled as extracted
- Turns rule-engine output into plain-language findings, fixes, owners, and effortconsultant-reviewed before release
Never judges compliance · never "repairs" a value under test · never invents a rule
Decides · deterministic engine
Validates and scores
- Renders the confirmed data to UBL and runs the five-layer gateXSD · rules · code lists · Schematron · country
- Issues every pass or fail with the identifier of the rule that produced itEN 16931 BR-* · Peppol · PINT OM
- Computes the 0–100 readiness score and the bandfrom the findings alone
- Runs version-pinned, checksummed rule artefacts — the same file gives the same answerreproducible on demand
The engine inside GoRoute's Peppol-certified platform · PINT OM 1.0.1 conformance · not a replica of the rules
✓A consultant reviews every report before it ships. AI-drafted, human-signed — and any AI-extracted value is labelled so it can be checked against the source document.
The release gate01 · Verdicts
Deterministic, or not at all
Pass or fail comes from a version-pinned, checksummed rule engine implementing the published standards — EN 16931, Peppol BIS, country packs. Run it twice, get the same answer.
02 · AI
A defined, bounded job
It interviews, maps, extracts, and writes. It is expressly forbidden from judging compliance, and every value it extracts is labelled so a person can confirm it against the source.
03 · Engine
The real thing
The scan uses the validation engine inside GoRoute's Peppol-certified platform (Access Point POP000991), whose Oman service is OTA-accredited via Union Digital Technologies — not a marketing replica of the rules.
04 · Data
Handled honestly
Sample data — tens to a couple of hundred invoices, not full ledgers — under NDA, processed in an isolated ClayDesk-controlled environment, and permanently deleted when the engagement closes.
Who it is for
Three people who need the same answer for different reasons.
No integration required to start
Export a sample and get back the exact field-level gaps between your ERP and the mandate's data model — the mapping table you would otherwise discover halfway through the implementation.
A readiness instrument for your clients
Deterministic findings with rule identifiers you can stand behind, packaged inside your own engagement. Ask us about running assessments across a client portfolio.
Start where you are
Three levels of readiness, in order of effort.
Each level answers a sharper question than the one before it. The first is what you believe about your data. The second is what your data actually does under the rules. The third is going live.
-
Level 1 · 5 minutes · free · self-service
Self-assessment
Ten questions on mandate coverage, data custody, and master-data readiness — an immediate scored readout of where you stand and who controls your invoice data.
-
Level 2 · sample data under NDA · consultant-reviewed
ClayDesk Assure data scan
Your real invoice data through the validation engine: a 0–100 score, ranked issue classes with rule identifiers, a field-mapping table, and a remediation plan with owners and effort. This page.
-
Level 3 · fixed-fee engagement
Remediation and go-live
Master-data fixes with named owners, ERP field mapping, and onboarding to a certified e-invoicing service — scoped from the report, so nothing is discovered twice.
Questions we actually get
Asked before the sample is sent.
What data do you need from us?
A sample, not your ledger: typically tens to a couple of hundred invoices. A CSV or Excel export from your ERP, UBL XML if you already produce it, or plain PDF invoices — whatever you can produce today. There is no template to fill and no integration to build before the assessment starts.
How much of this is AI?
It is AI-assisted, deliberately. AI maps unfamiliar column headers, extracts structured data from PDFs, and drafts the narrative of the report. It is expressly forbidden from judging compliance. Every pass or fail comes from a deterministic, version-pinned rule engine implementing the published standards, so the same file always produces the same verdict. If a vendor tells you their AI decides compliance, ask which model, and what happens when it changes.
Is this a compliance certificate?
No. An assessment is not certification and no tax authority endorses it. What it gives you is reproducible: every finding carries the identifier of the rule that produced it, so your auditor or advisor can trace it back to the standard and re-run the check.
Which mandates and rule sets are covered?
Oman Fawtara (PINT OM 1.0.1, including the Oman-specific rules) and the EN 16931 / Peppol BIS rule set that underpins Peppol-based mandates. Further country rule packs are added as their specifications stabilise; ask us about your jurisdiction and we will tell you plainly what is covered today rather than what is on a roadmap.
How is our data handled?
Sample data is shared under NDA. It is processed in an isolated, ClayDesk-controlled environment — never uploaded to a public tool — and permanently deleted when the engagement closes. The report states which processing mode was used and labels any value that was AI-extracted so it can be confirmed against the source document.
How long does it take?
The scan itself runs in minutes. The report is drafted, then reviewed and released by a consultant — typically within a few business days of receiving the sample. A traditional readiness engagement takes weeks; the difference is that a machine, not a workshop, reads the data first.
Know your gaps before the mandate does.
Tell us your country, your ERP, and roughly how many invoices you issue. We tell you what sample to send, what the report will contain, and how long it takes — with a written fixed-fee scope within 24 hours.