AI invoicing software in Spain: a 2026 buyer's guide that doesn't name names
18 months ago, “AI” on an accounting-software website was a differentiator. Today it’s noise: every platform on the market puts it on the first screen, in the nav, and in the chat pop-ups. The buyer’s question is no longer “does it have AI?”. It’s “what does its AI do, exactly?”.
This article gives you the framework: four product categories that exist today in Spain, eight questions that separate real AI from marketing, and five demo red flags you need to spot before signing.
The 4 product categories that coexist in the Spanish market
1. ERP-style generalists
Platforms that cover invoicing, accounting, CRM, inventory, HR, everything. Origin: SMB ERP. Their AI is typically improved OCR plus some categorization rules. Real value sits in integration (everything in one place), not the AI. Pricing: 25-80 €/month for the base, extra modules for SII, per-store inventory, phone support, etc.
2. Freemium for self-employed
The segment that dominates “free invoicing” search. Free plan with 50-100 invoices/year, paid plans 8-15 €/month. AI here is usually nonexistent or very basic (OCR for expenses on the paid plan). Excellent for one-line-of-business autónomos with low volume, problematic when you grow.
3. Vertical AP / firm-first
Products specialized in accounts payable and/or accounting firms, with extraction and categorization AI as the core. Per-seat or per-document pricing. They’re the newest in the market and they understand the multi-client firm flow best.
4. AP AI-native specialist
Products designed from day one with AI as the central engine, not a bolt-on feature. Calitem belongs here. The difference is architectural: the model learns from each customer’s history, not from global rules. Transparent pricing, no upsells for legal compliance.
The 8 questions that define “real AI”
Print this list and use it in every demo. If an answer is vague, ask for screen-share.
1. Is data extraction rule-based or model-based?
Rule-based: the system looks for predefined patterns (“whenever ‘TOTAL’ appears in line X, extract the number to the right”). Works on standard invoices, fails the moment the vendor changes format. Model-based: the AI understands the document as a whole, no matter the format. Ground truth: ask them to process a quirky invoice of yours, not their demo one.
2. Does it learn from your vendor history?
Concrete question: “if I upload 5 invoices from Iberdrola and categorize them all as 628, will the sixth Iberdrola invoice categorize automatically without me doing anything?”. If the answer is “yes but you have to activate a rule”, it’s not AI, it’s a rules engine with a UI.
3. How is confidence measured per field?
Serious AI returns a score per extracted field. The user should see “invoice number: 99% confidence, total: 95%, date: 87%”. Without scores, there’s no traceability or auditability. The auditor will ask.
4. What happens when the model gets it wrong?
Two correct answers: “it lets you correct and learns from your correction” or “it flags low confidence and queues to human review”. Wrong answer: “it never fails”. Every model fails. The point is the feedback loop.
5. Is there automatic categorization against my chart of accounts?
Specific question: “can I upload my custom chart of accounts and the AI categorizes against it, not against a generic one?”. SMEs and firms customize accounts. AI that only understands the base chart forces you to re-categorize manually.
6. Is 3-way match automated?
PO + delivery note + invoice automatic match. Almost nobody in the SMB segment offers it. If they do, ask to see a demo with raw data. This is the mark of a serious product.
7. Does it cover Verifactu without an upsell?
Direct question: “is Verifactu compliance in my base plan or do I pay extra?”. If you pay extra, the “AI value” is diluted because you’re paying for legal compliance.
8. Is there a public API with webhooks?
If you want to plug into your ERP, your own CRM, your data warehouse, you need an API. Ask if it’s publicly documented (not behind an email to sales). Ask for the docs URL. If there are no public docs, the product is locked in its silo.
Comparative matrix by category
| Question | ERP-style | Freemium | Vertical AP | AP AI-native |
|---|---|---|---|---|
| 1. Model-based extraction | Partial | Almost no | Yes | Yes |
| 2. Learns from history | No | No | Partial | Yes |
| 3. Per-field confidence | Rare | No | Common | Yes |
| 4. Feedback loop | No | No | Yes | Yes |
| 5. Custom chart of accounts | Yes | Limited | Yes | Yes |
| 6. 3-way match | Expensive add-on | No | Some | Yes |
| 7. Verifactu in base | Add-on | Higher plan | Common | Yes |
| 8. Public API | Limited | No | Yes | Yes |
This table isn’t to sell you a category. It’s so your next demo asks the 8 questions in order and you compare answers with data.
5 demo red flags to spot
- “AI” that’s only OCR + if-then rules. If the answer to “how does it learn?” is “you create rules”, it’s not AI.
- Fixed categorization that doesn’t learn. Concrete test: correct an invoice’s category three times, upload the fourth, see if it self-categorizes. If not, no learning.
- No confidence measurement. The system shows you the extracted data but doesn’t tell you how sure it is. Without this, you can’t prioritize review.
- Verifactu as a paid add-on. Compliance with the law shouldn’t be an upsell. If it is, there’s a pricing or architecture problem.
- No trust center / SOC 2 / security page. You’re giving access to your invoices, your IBAN, your vendor base. If the website doesn’t have a
/securitypage detailing how they protect it, red flag.
Honest 2026 price ranges
- ERP-style generalist: 25-80 €/month base + 20-50 €/month in usual add-ons (SII, inventory, support)
- Freemium for self-employed: 0-15 €/month
- Vertical AP / firm: 50-150 €/month per user, or by volume (0.10-0.30 €/invoice)
- AP AI-native specialist: 30-100 €/month base + 0.06-0.10 € per extra document over included
If you pay more, ask exactly what it includes that the others don’t.
The hidden price: ERP integrations
A3, Contasol, Sage usually sell the connectors as a separate module, not as a feature of your plan. Ask before signing: “is the integration with [my ERP] included or extra?”. A bad integration = export to Excel + manual import = you still have manual work.
How Calitem positions itself honestly
In the matrix above, Calitem sits in the AP AI-native column. We don’t compete in ERP-style (no CRM, no inventory), nor in freemium (not our customer profile). Where we compete: extraction that learns, categorization that remembers, 3-way match included, Verifactu without upsell, public API with webhooks.
The honest part: if you also need CRM and inventory in one tool, Calitem isn’t for you. If you need the best AI possible processing your accounts payable, yes.
Related reading
- OCR vs AI in accounts payable: the seven criteria for telling real AI from rebranded OCR.
- Verifactu in 2026: the plain-English compliance guide.
- AI in accounting: the pillar guide on what AI actually automates and what it doesn’t.
- Smart capture: the model that extracts data from any format.
- Touchless AP: the endpoint of the cycle when everything is automated.