O.D.E.S.S.A. Layer 1: Origination AI for M&A — Workflow in 4 Hours

Updated on 1 June 2026

The first layer of the O.D.E.S.S.A. framework focuses on target Origination: identifying acquisition targets that match strategic and financial criteria. Traditional origination requires weeks of manual research. AI origination compresses this to hours while maintaining accuracy through human senior review. This guide explains the practical workflow in 4 hours, the technology stack, real cases, and governance pitfalls.

Origination AI vs traditional origination

Traditional origination: senior associate spends 2-3 weeks building long-list (sector mapping, database queries, manual filtering, initial outreach). Cost: EUR 25-40k of professional time per origination cycle. Output: 30-50 initial targets, refined to 10-15 worth deeper analysis.

AI-powered origination: 4 hours of structured workflow combining database queries, AI scoring, senior review. Cost: EUR 3-5k per cycle (license fees + senior time). Output: 30-50 initial targets with multi-dimensional scoring, ready for senior qualitative review. 6-8x acceleration at equivalent quality.

The technology stack

• Data sources: Mergermarket, S&P Capital IQ, Pitchbook (premium databases) + Cerved/Crif (Italian-specific) + Crunchbase (startup ecosystem)
• LLM layer: GPT-4 or Claude for natural language brief processing and target scoring
• RAG infrastructure: vector database (Pinecone, Weaviate) for target-specific document retrieval
• Workflow orchestration: custom Python/JavaScript or no-code (Zapier, Make) for cross-tool integration
• Output platform: structured database (Notion, Airtable) for senior review interface

The concrete 4-hour workflow

Step 1 — Brief in natural language (15 minutes)

Senior partner dictates investment thesis: sector, sub-sector, geography, EV range, EBITDA threshold, ownership characteristics (family vs PE-owned), specific operational characteristics (technology, brand strength, supply-chain position). LLM processes brief into structured search parameters.

Step 2 — Long-list generation (1 hour)

Automated database queries across data sources, deduplication, initial filtering on hard criteria (size, geography, sector classification). Output: 100-200 candidate targets matching basic criteria. Human review removes obvious mismatches.

Step 3 — Multi-dimensional scoring (1 hour)

For each candidate, AI analysis on multiple dimensions: strategic fit, financial profile, operational characteristics, ownership transition probability, valuation potential. Scoring weighted by investment thesis priorities. Output: ranked long-list with 30-50 prioritised targets.

Step 4 — Senior qualitative review (1.5 hours)

Senior partner reviews top 30-50 targets, applies qualitative judgment (relational considerations, sector dynamics knowledge, competitive sensitivities, founder profiles), refines to 10-15 priority targets. Strategic considerations AI cannot capture remain in human domain.

Anonymised real case — buy-side mid-market industrial

Context: Italian industrial group seeking bolt-on acquisitions, EUR 5-25M EV range, precision machinery sector, Northern Italy preference. Traditional origination cycle: 3 weeks, EUR 40k cost. AI-powered approach: 4 hours active work, EUR 4k cost. Quality comparison: AI approach identified 12 priority targets, traditional approach 8 priority targets — both included same top-3 candidates. AI approach surfaced 4 additional candidates not on traditional radar (smaller specialised players). Conversion: 2 acquisitions closed from AI-generated long-list vs typical 0-1 from traditional approach.

Pitfalls and governance

Pitfall 1 — Garbage in, garbage out

AI quality depends on data source quality. Italian mid-market data often incomplete in international databases. Pattern: combine multiple sources (international + Italian-specific) for complete coverage. Validate AI output against manual cross-check for 5-10 targets to confirm accuracy.

Pitfall 2 — Confirmation bias

AI scoring can reinforce existing assumptions. Senior reviewer must apply contrarian thinking: are we missing targets that don’t fit obvious patterns? Pattern: structured devil’s advocate review of low-scored targets often reveals overlooked opportunities.

Pitfall 3 — Data privacy

Target information may include confidential data. Confidentiality protocols: ensure data sources have appropriate licensing, AI processing within secure environments, no inadvertent sharing through public LLM APIs of confidential client information. Italian and EU GDPR compliance critical.

When AI origination works best

• Clear investment thesis with structured criteria
• Active buy-side mandate (vs opportunistic deal pursuit)
• Sectors with sufficient database coverage
• Mid-market focus (EUR 5-100M EV typically)
• Geographic concentration enabling local database use

When AI origination has limited value

• Highly relationship-driven sectors (private banking, specific niches)
• Pre-revenue targets (insufficient data)
• Stealth-mode companies not in public databases
• Cross-border with limited database coverage in target geography