SFV Thesis · AI-native financial infrastructureIC-grade research,
IC-grade research,
agent-native.
An operator-led thesis on the companies rebuilding back-office and middle-office finance with LLMs at the core — scored, sourced, and ready for the partner meeting.
Demo▶ Expand
Thesis positioning
AI centrality × Workflow depth · bubble = total scorePursueWatchPassTop-right · build conviction
Market map
9 companies · ⭐ scored 22+| Company | Subsegment | Stage | Core | Score | One-line take |
|---|---|---|---|---|---|
| Sardine⭐ | Compliance | Series C | yes | 27 | AI-native fraud and compliance platform with strongest data loop (Sonar consortium, 5.4B devices) and founder-workflow fit in the category; $660M valuation, 130% YoY growth, elite logos; co-founder departure and credit underwriting distraction risk warrant diligence. |
| Casca⭐ | Underwriting | Series A | yes | 26 | AI-native LOS deployed at the #1 and #2 SBA 7(a) lenders; strongest logo quality at Series A in the market; expansion beyond SBA is the key question. |
| Bretton AI (fka Greenlite)⭐ | Compliance | Series B | yes | 25 | AI-native compliance agents deployed at regulated banks and Tier 1 fintechs; 8x ACV expansion since seed; UiPath/WorkFusion combination and zero disclosed ARR are the key questions. |
| Ramp⭐ | FinOps | Late-stage (Series E+) | no | 24 | The benchmark: $1B+ ARR spend-management platform with the most credible AI agent layer in FinOps; at $32B pre-IPO, track for thesis intelligence not venture allocation. |
| Rillet⭐ | Back-office | Series B | yes | 24 | AI-native ERP replacing NetSuite for venture-backed SaaS; 200+ customers, $108.5M raised in <12 months, 5x revenue growth; NetSuite's AI roadmap and ICP concentration in venture-backed companies are the key risks. |
| Taktile⭐ | Underwriting | Series B | yes | 24 | AI-native decisioning platform for FIs with 3.5x ARR growth and 18+ customer logos spanning fintech to traditional FIs; strongest breadth play at this stage; FICO's AI push is the key risk. |
Laurel⭐ | Back-office | Series C | yes | 23 | AI-core timekeeping platform with 300% ARR growth, Big Four logos, and strong founder-workflow fit; thesis-adjacent — this is professional services productivity, not financial infrastructure; watch for expansion into deeper financial workflow automation. |
| Norm Ai⭐ | Compliance | Growth (post-Series B) | yes | 23 | AI-core compliance agents deployed at $30T+ AUM institutions; best advisory board in RegTech; zero disclosed revenue metrics and investor-customer overlap are the key questions. |
| Hebbia | Back-office | Series B | yes | 20 | Best-in-class AI document analysis platform with elite financial services logos; thesis-adjacent — this is analytical productivity, not financial infrastructure; watch for deepening into underwriting or compliance workflows. |
About this thesis
What “AI-native” actually means here.
We’re looking for companies rebuilding back-office and middle-office financial workflows with LLMs and agents at the core — not GPT wrappers bolted onto incumbent suites. The test is binary: remove the AI and the product still works? It’s cosmetic. Remove it and the product disappears? It’s core.
Memos commit to two judgments every time: AI core vs. cosmetic, and founder-workflow fit. The rubric on the right scores six dimensions out of five each. 22+ earns a star and a place on the “build conviction” shortlist.
| Dimension | 1 | 3 | 5 |
|---|---|---|---|
| AI centrality | AI is a feature label on a traditional SaaS product | AI handles a meaningful but optional workflow | AI is the product; remove it and nothing is left |
| Workflow depth | Surface-level tool (notifications, summaries) | Owns one full workflow end-to-end | Owns multiple connected workflows; replaces a role |
| Data loop | No proprietary data accumulating | Some proprietary data, weak loop | Strong loop: more usage → better model → more usage |
| Founder-workflow fit | Founders have no domain background | Founders have adjacent domain experience | Founders have done the exact job being automated |
| Traction signal | Pre-revenue or pilots only | Real ARR, mixed retention | Real ARR with strong NDR and logo quality |
| SFV thesis fit | Tangential to the thesis | Clearly within the thesis | Central to the thesis; SFV should know this company |
Research stack
How a memo gets made.
Each memo is one agent run. The agent picks its own tool calls; the prompts enforce citation and numbers discipline.
Step 1
User input
Company name
→
Step 2
Claude Agent SDK
claude-opus-4-6 · streaming query()
→
Step 3
Exa MCP
Four research tools
web_search_exacompany_research_exalinkedin_search_exacrawling_exa
→
Step 4
System prompts
system · memo template · rubric · operator context
→
Step 5
File output
output/<slug>-memo.md · output/market-map.md
Step 1
User input
Company name
Step 2
Claude Agent SDK
claude-opus-4-6 · streaming query()
Step 3
Exa MCP
Four research tools
web_search_exacompany_research_exalinkedin_search_exacrawling_exa
Step 4
System prompts
system · memo template · rubric · operator context
Step 5
File output
output/<slug>-memo.md · output/market-map.md