Construction · service #6 · class C

AI invoice data extraction for construction in Spain

WhatsApp photos, email PDFs, a box of paper archive — and out comes a table: proveedor, NIF, base imponible, VAT 10/21%, retención, line items, total. Arithmetic is checked by code, doubtful scans get flagged, and the books stay with your gestoría.

Get invoice extraction Chat on WhatsApp

$9.40 per invoice processed by hand · on average; best-in-class — $2.78 [I — Ardent Partners, 2025]

ISP a no-VAT invoice isn't an error · inversión del sujeto pasivo, art. 84 LIVA: the note is recognised

10/21% VAT and the 1% retención · fields are checked with formulas, not «by eye»

in batches archive digitisation · hundreds of invoices overnight, −50% of the cost

What the box of invoices really costs

01 · Problem

Ardent Partners, 2025 [I]

$9.40

per manually processed invoice

The average cost of handling an invoice by hand; the cycle — 9.2 days. Rekeying fields, chasing the supplier, hunting for the lost scan.

art. 84.Uno.2º.f LIVA; art. 101 LIRPF

€0

of VAT on a subcontractor's invoice

In construction an invoice legitimately arrives without VAT (inversión del sujeto pasivo) or with a 1% retención (módulos). Whoever rekeys by hand «fixes» — and confuses — exactly that.

typical SMB process — verified during the audit

a quarter

before you learn what was inside

The box of invoices goes to the gestoría whole, and the per-site cost picture arrives after the fact — when renegotiating with the supplier is no longer on the table.

And the past years' archive sits on paper: finding a two-year-old invoice for a warranty dispute is a half-day expedition.

Invoice processing is a production process

02 · The process we close

incoming invoices (photo · PDF · email · archive) field extraction (proveedor · NIF · base · VAT · retención) arithmetic check by code table/Excel + export for the gestoría searchable archive

CLASS

Class C · production. Invoices are a flow with a unit of work: the effect is measured in time and entry errors. Secondary class — B: a per-site table turns the paper box into a cost picture you can act on before the quarter ends.

Today — by hand

Invoices pile up in the phone, the inbox and the glovebox. Some get rekeyed into Excel, some ride to the gestoría in a box. The subcontractor's no-VAT invoice sparks an argument, the retención gets lost, and per-site costs live in someone's head.

With the system

Every invoice — photo or PDF — enters the pipeline and comes back as a table row with all the fields and its scan attached. The gestoría gets an export instead of a box, you get per-site costs in real time, and the archive is searchable in seconds.

Input → AI → output. And where the person is

03 · How it works

An example: the month's batch holds a €4,800 subcontractor invoice with no VAT line. The system finds the «inversión del sujeto pasivo» note, files it in its own column and won't let anyone «fix» the zero to 21%. Hand-made fixes like that are exactly what ends in redone tax filings.

Input

Source material

Invoices as they come: WhatsApp photos, email PDFs, scans — and the old archive by the box
Your suppliers: usual invoice formats, who works under ISP, who is on módulos
Where to export: your Excel, your books or your gestoría's format

What the AI does

Extracts the fields: proveedor, NIF/CIF, número, fecha, base imponible, VAT rate and cuota, retención, line items, total
Knows the construction rules: an ISP note → zero VAT is legitimate; a 1% retención is a módulos marker, not a typo
Runs the archive in batches: overnight processing, −50% of the cost, hundreds of invoices per pass

Output

What you get

A table/Excel with every field — and the scan linked to each row
An export for the gestoría in the agreed format
A searchable archive: by supplier, date, site, amount

Human

What the person does

Spot-checks fields before posting — «on approval» level
Resolves «review» flags: handwriting, crumpled photos, odd formats
Postings and filings stay with your gestoría: we prepare data, not the books

RULE 01

Arithmetic is checked by code, not by the language model: base × rate = cuota, line items sum to total — a mismatch surfaces as a flag. Numeric slips by language models in tables are a known weakness, so the formulas are mandatory.

RULE 02

ISP and retención are rules, not exceptions: a no-VAT invoice citing art. 84 LIVA goes to its own column, not to «errors».

RULE 03

A bad scan is never guessed: an illegible field comes out flagged «review», no silent substitutions.

RULE 04

Professional boundary: this is data preparation, not accounting or tax advice — deductions and filings are your gestoría's/asesor's call.

What is included — and what is not

04 · What is included

Included

Field extraction into a table/Excel: proveedor, NIF/CIF, número, fecha, base, VAT 10/21%, retención, line items, total
Construction logic: ISP (art. 84 LIVA), the 1% retención (módulos), formula-based arithmetic checks
Templates for your suppliers' formats; new ones added as they appear
One-off archive digitisation — a separate project: hundreds of invoices in batches
Export for the gestoría or your books — format to agree («to confirm»)
A DPA before real CIF/NIF data is uploaded; data stored in the EU
A guide + recorded walkthrough + short consultation

Not included

Accounting entries and filings (modelos 303, 111, 130) — the gestoría/asesor
Auto-posting without a person: a spot-check comes before the books
Invoice issuing and VeriFactu software: VeriFactu governs issuing and becomes mandatory from 2027; our terrain is intake and archive
Handwriting and illegible scans «on trust» — only with a «review» flag

Integrations & data

WhatsApp · Telegram · email — invoice intake · to confirmExcel — table and export · nativeYour gestoría's format — A3/Sage and others · to confirmYour books (Holded, Quipu…) — custom · to confirmArchive — EU storage — included

The stack: invoice2data (MIT) as the template base + PaddleOCR (Apache-2.0) for scans + Claude for fields and line structure; volume runs in batches (Batch, −50% of API cost). Sparrow — reference only: GPL.

Five steps — launch in a week

05 · Implementation

Audit

0.5–1 day · Your suppliers' invoice formats, who is on ISP and módulos, the export format for the gestoría.

Setup

3–5 days · Field templates + check formulas + intake channels.

Test

30–50 invoices · Real invoices, every field checked by hand — discrepancies down to zero.

Launch

~1 week from start · The flow goes live; the archive — as a separate project pass.

Support

by subscription · New supplier formats as they appear, SLA-backed support.

SLA: An invoice lands in the table in minutes; litcore support — reply ≤ 1 business day · If it does not fit the window, the flow is too wide: we cut the scope, not the quality.

The economics, counted conservatively

06 · Economic effect

An ROI canvas for a contractor with 60 supplier invoices a month. We count only entry time; VAT/retención errors and the archive digitisation sit separately and on top.

Parameter	Value	Source / assumption
Process and class	invoice processing — class C (secondary B)	litcore framework, block 2 · lever: time + errors
Volume	60 invoices/month	[estimate] — refined during the audit
Entry «before»	~10 min per invoice: rekeying, chasing, searching	[estimate]; market: $9.40/invoice (Ardent Partners, 2025 [I])
Entry «after»	minutes of spot-checking per batch	[estimate] — fixed in the pilot
Time lever	60 × 10 min ≈ 10 h/month × €28/h ≈ €280/month	[estimate]; the rate — your real one
Entry errors	a wrong cuota, a lost retención → fixes at the gestoría — reserve	not in the conservative scenario
Archive digitisation	a one-off project from €490 — costed separately	the «Archivo» plan; price — approved by the owner
Costs	Start from €590 + €0.15/doc or €39/mo	«from» price — approved by the owner

PAYBACK

≈ 4 months. Start at €590 with 60 invoices a month and a €28/h rate, on a 40/70/100% adoption curve over 3 months [estimate]. Avoided refilings and the invoice found in time for a dispute come on top and stay out of the maths.

Adoption: every invoice — WhatsApp photo included — goes through the pipelineTime: minutes per invoice before/after, hours per monthQuality: arithmetic discrepancies in the test — zero; ISP and retención in their own columnsConvenience: the gestoría gets an export with scans linked to rows, not a box

Not financial advice: the calculation is a proposal built on market reference points; final figures are fixed after the audit and a pilot on your invoices.

Want this calculated on your numbers? Request a proposal →

Price — from €590

07 · Price

Model: one-off setup + pay-per-volume (€0.15/doc) or a subscription. The archive is a separate one-off project from €490: hundreds of invoices in batches. The final price — after the format audit.

Start

from 590 €

+ €0.15/doc or €39/mo

field extraction into a table/Excel
templates for your suppliers’ formats
DPA (GDPR) check before real data

Request a proposal →

Pro

from 1 490 €

+ volume package

everything in Start
integration for supplier formats
optional integration with bookkeeping/gestor

Request a proposal →

Data, taxes and boundaries — by the rules

08 · Risks & control

Data

Invoices are commercial and personal data: NIF, addresses, prices. Only what's needed is transferred, a DPA before real CIF/NIF data, storage in the EU; the EU/US transfer for the analysis is disclosed (DPF + SCC), and the non-Zero-Data-Retention mode is stated openly.

Human oversight

«On approval» level: a spot-check before posting; anything doubtful comes out flagged «review» rather than posted silently.

Not accounting advice

Deductions, entries and filings are your gestoría's/asesor's responsibility. We deliver verifiable data with every field linked to its scan.

AI Act art. 50 and VeriFactu

The use of AI is disclosed. VeriFactu governs invoice issuing and becomes mandatory from 2027 (companies — 01.01; autónomos — 01.07; RDL 15/2025) — not in force as of July 2026. Our terrain is intake and archive: it prepares the transition without waving deadlines.

AI-assisted — human-supervised · AI Act art. 50

09 · Custom

A more complex document flow?

Invoices arrive through a client portal, EDI or Facturae? Need albaranes, contracts or acceptance acts processed alongside the invoices? Describe the flow — we'll build the pipeline around it and calculate the payback.

Describe my document flow

We'll demo it on your data — 15 minutes, no obligation.