{}El Fosodefensibility diligence
Home
← All cases

Kabilio

Barcelona, Spain · Pre-seed (Nov 2025; €4M + €200K ENISA, no Series A) · AI automation for accounting firms (accountant-client workflow)

TLDR: An AI accounting copilot on Spain's Verifactu mandate. The regulatory and integration depth is evidenced; the claimed network effect is not. A switching-cost bet with data-moat optionality. Conditional yes. The meeting is about two numbers.

What they claim, and what the evidence saysmarket network: Asserted only · data: Partially evidenced

What the company asserts, held against each type's proof bar.

market network

Asserted only

The firm-client collaboration layer is real, which is more than a pure automation tool, but it does not compound across firms. Rated Asserted only as a network effect. Its value shows up instead as switching cost (diagnostics) and as a potential data effect (below).

  • What looks like a market network is a per-firm collaboration workflow: each firm uses Kabilio with its own client base, siloed from other firms.
  • No evidence that adding the 101st firm raises value for the first 100.

data

Partially evidenced

This is the stronger of the two claims. Kabilio processes private Spanish invoice and ledger data across firms, which is genuinely access-constrained and domain-specific. That is the raw material of a real data moat. What is missing is proof that accuracy improves cross-firm with volume rather than being a fixed model plus rules. Rated Partially evidenced: the ingredients are there, the compounding loop is not yet shown.

  • 97% accuracy is company-stated; the cross-firm compounding loop is asserted by structure, not shown.
Is the network effect real? Seven tests5 Partial · 1 Weak · 1 None
DiagnosticRatingBarrier testWhat would change this
Atomic network unitNoneNo cross-firm network barrier exists. A rival can win one firm without needing any other.A shared layer across firms, for example benchmarking, a shared vendor/supplier graph, or a client-portability network, would create a real atomic unit.
Cold-start statusPartialEarly adoption compounds into references and data, not into a network others must overcome.Evidence that firm density in a region or segment raises value for other firms there.
Density and clusteringWeakDensity matters only if it compounds; today it does not visibly.A data or benchmarking effect that gets better the more firms in one market use it.
Multi-tenanting exposurePartialThe integration surface is the thing that discourages multi-tenanting; it is real but unproven at retention level.Retention data showing firms consolidate onto Kabilio rather than keep it alongside incumbents.
Disintermediation riskPartialThe barrier is whether Kabilio's Spanish compliance and bank-integration depth outpaces incumbents adding AI.An incumbent shipping Verifactu-compliant AI processing would confirm the risk; deep firm lock-in would soften it.
Value curvePartialCompounding holds only if accuracy improves with cumulative cross-firm volume.Evidence of classification accuracy rising with data volume across firms.
Switching-cost decompositionPartialPasses the barrier test if ripping Kabilio out means rebuilding bank connections, compliance, and client history; plausible but unproven.Net revenue retention above 100% and firms moving all clients onto Kabilio.
Does the moat survive AI?1 strengthens · 1 intact · 2 intact-but-thinner

Each moat component against LLM-era commoditisation. Every call carries its falsifier.

ComponentOutcomeFlips if…
Generic invoice OCR and classificationintact-but-thinnermedium confidenceThis flips to dissolves if a horizontal document-AI provider ships turnkey Spanish-compliant processing that firms accept.
Spanish regulatory compliance (Verifactu real-time invoicing, tax rules) and bank-integration depthintactmedium confidenceThis flips if Verifactu compliance and Spanish bank connectivity become a commoditised layer any vendor can buy.
Proprietary Spanish invoice and ledger dataset compounding classification across firmsstrengthenslow confidenceThis flips to dissolves if the classification turns out to be generic OCR that public models match, with no cross-firm learning.
Embedded firm workflow and switching costsintact-but-thinnermedium confidenceThis flips to intact if net retention proves high and integrations prove deep enough that switching is a multi-quarter project.
What to ask the founder, and what is missing3 questions · 4 gaps

Three questions for the founder

  1. Does invoice and ledger data from one firm measurably improve classification accuracy for others, or is each firm's model independent? Show me the mechanism and the accuracy-over-volume curve.
  2. What is net revenue retention, and within your ~100 firms, how many run all of their clients through Kabilio versus a subset?
  3. When Sage, A3 or Holded ship Verifactu-compliant AI natively, what specifically keeps a firm on Kabilio?

Evidence missing

  • Evidence that classification accuracy improves with cross-firm data volume. This is the difference between a real, strengthening data moat and a fixed model plus rules. It is the decisive question for Kabilio.
  • Net revenue retention and share-of-clients penetration within the ~100 firms. Switching costs are the best barrier candidate, and only retention plus penetration data show whether they are real.
  • Competitive posture versus incumbent accounting platforms (Sage, A3, Holded). Kabilio's wedge is a feature incumbents could add; win/loss against them is the key defensibility signal.
  • Disclosed revenue and valuation (figures circulating publicly are estimates, not company-disclosed). Unit economics and ACV determine whether a fragmented, low-price Spanish market can support a venture outcome.
What must be true by Series A4 conditions, each with proof and kill

Derived from evidenced types only.

The proprietary Spanish data compounds into a classification advantage rather than staying a fixed model plus rules.

Proof: Classification accuracy improving with cumulative cross-firm volume, and engagement cohorts improving vertically: newer firms onboarding to higher accuracy than earlier ones reached.

Kill: Accuracy is generic and plateaus, matchable by horizontal document-AI.

Switching costs implied by the integration surface are real, not pre-seed optimism.

Proof: Net revenue retention above 100% with rising share of clients per firm (the share-of-wallet read), and CAC payback shortening as longer lifetimes lift LTV:CAC.

Kill: Firms keep Kabilio alongside incumbents on a slice of clients and churn without expansion.

The Spanish regulatory and integration moat holds against incumbent accounting platforms adding AI.

Proof: Verifactu and bank-integration depth staying ahead of Sage, A3, Holded and similar, with pricing power intact rather than discounting to close, evidenced by wins against them.

Kill: An incumbent ships equivalent Spanish-compliant AI natively and firms default to it.

The Spain-specificity that protects the moat does not permanently cap the market.

Proof: Acquisition staying efficient, with rising organic and referred share among firms, extending into a second regulated geography without rebuilding the whole compliance stack.

Kill: The moat is non-portable and the company caps out at a fragmented, low-ACV Spanish market.

What would kill the thesis3 ways this dies
  • The thesis breaks if incumbent accounting platforms embed equivalent Spanish-compliant AI, since Kabilio's wedge is a feature they can add.
  • The thesis breaks if the classification accuracy is generic document-AI that horizontal models match, removing the data barrier.
  • The thesis breaks if the Spanish regulatory specificity that defends the moat also makes it non-portable, capping the company in a fragmented low-ACV market.
Every source, cited9 items, cited

Written before the report. If it is not cited, it does not exist.

SupportsSourceConfidence
Funding €4M pre-seed (Nov 2025), Visionaries Club + Picus Capital lead, +€200K ENISATech.euhigh
Founded 2024, Barcelona, José Ojeda and Álex VallsPulse 2high
Founder backgrounds: McKinsey, Rocket Internet, 011h, Social Point, ExoticcaEU-Startupsmedium
Three tools: invoice processing 97% accuracy; bank reconciliation across 99% of Spanish banks; Verifactu-compliant invoicing with real-time syncTech.euhigh
Automates firm-client information exchange; entries integrate with accounting softwareTech.euhigh
Testing conversational assistant 'Kabi' for NL queries and admin actionsPulse 2high
Nearly 100 firms; company-stated productivity up to 50% in quarterly tax periodsTech.eumedium
Market: ~65,000 accounting/tax advisory firms in SpainTech.euhigh
Strategy: near-term Spain focus; regulatory + fragmented advisory fit automationPulse 2high

{ "verdict": null }

I would take the meeting. The regulatory depth and bank-integration are real and evidenced, and the Spanish-data ingredients are the right raw material for a moat AI does not erode. What I am being asked to pay for is the cross-firm compounding loop, and that is the one thing still unproven. So the meeting is about two facts, not a story: net revenue retention, and whether classification accuracy measurably improves with cross-firm volume. Show me either and this becomes a priced bet on the other; show me neither and it is a compliance wrapper in a fragmented, low-ACV market. Conditional yes.

El Abogado del Diablo red-teams the bull case on this evidence.