Case file — BE44AB44
The idea
“COBOL modernization assistant — AI that reads legacy COBOL bank code, documents what it does in plain English, and generates equivalent modern code. Banks can't find COBOL developers.”
The panel
Market Analysis You're entering a crowded field with at least two direct competitors already live: COBOL Copilot (offering documentation, semantic search, and web-based collaboration) and Cobol Intel (free tier with code explanation, conversion to Java/Python/SQL, and architecture mapping). Both are actively positioning around the same pain—undocumented legacy systems and developer scarcity. The market is real: banks genuinely need COBOL modernization help, and the shortage of COBOL developers is well-documented. However, the red flag you're ignoring is adoption friction. Banks move glacially on tooling decisions and fear vendor lock-in with AI systems touching mission-critical code. Free trials and slick demos don't overcome procurement timelines measured in quarters. Your genuine advantage: timing. The wave of COBOL developer retirements is accelerating, and regulatory pressure (especially post-2020) is pushing banks toward modernization budgets. If you can solve trust and compliance (audit trails, explainability, change control) better than competitors, you have an opening.
Your core blindspot: COBOL semantics are deceptively complex. Business logic embedded in variable naming conventions, implicit state across batch runs, and undocumented regulatory constraints won't parse cleanly into "plain English." You'll hit a wall translating financial calculations where rounding rules, audit trails, and compliance requirements are baked into the code's structure, not its syntax. Build-vs-buy trap: Don't build your own COBOL parser. Use GrammTech or Synopsys's existing tools—they've spent decades handling dialects and edge cases. Your differentiation must be the AI layer, not reinventing parsing. No moat here yet. Competitors like Astera and Blu Age already do this; you need a wedge—maybe targeting a specific vertical (insurance claims processing?) or a particular pain point (test case generation?). Generic COBOL-to-modern translation is crowded. One win: Using LLMs to generate documentation from COBOL is actually smart and underserved. That's your real product if you focus there first.
Your CAC will be brutal. Enterprise sales to risk-averse banks take 9-18 months minimum; you're looking at $50K-$200K per deal but need 6-12 months of runway just to get meetings. You haven't solved how to get past procurement gatekeepers who don't believe AI can handle mission-critical code translation. Your pricing assumption: you're probably thinking per-line-of-code or per-project. Banks will demand fixed-price contracts with liability guarantees you can't give. Actual pricing power is near-zero until you have proven case studies. You'll burn through 18-24 months of runway with zero revenue. That's your real problem—not the technology, but the sales cycle mismatch. One thing working for you: switching costs are astronomical once a bank picks a tool. If you solve the first customer, retention and expansion are nearly automatic. That's worth the long winter.
Timing verdict: Late, but with a compressed window. The COBOL crisis peaked around 2020-2022 when banks panicked during COVID. Most large institutions have already hired contractors, built internal teams, or committed to specific migration vendors (IBM, Accenture, Deloitte). You're entering when urgency has cooled but budgets haven't reset. Critical macro trend: Regulatory pressure on legacy system risk. Banking regulators are tightening requirements around system modernization timelines. This creates sustained (not spike) demand, but it favors established vendors with compliance track records, not new entrants. Opportunity window: Closing. The 18-month gap between now and late 2027 is when most banks finalize their modernization strategies. After that, they're locked into chosen paths. One genuine advantage right now: AI code generation has matured enough to actually work on COBOL syntax—this wasn't true in 2023. But you need pilots with real banks immediately, not a polished product in six months.
Competitors found during analysis
Live dataCOBOL Copilot
Documentation, semantic search, web editor
Cobol Intel
Free tier, code explanation, architecture mapping
Cause of death
The sales cycle will outlive your runway
Banks take 9–18 months from first meeting to signed contract. You have no product, no case studies, and no procurement relationships. Your CFO panel is right: you're looking at 18–24 months of near-zero revenue while burning through development and sales costs. That's not a cash flow problem — it's a structural mismatch between your capitalization (zero) and your go-to-market reality. Even well-funded startups die in this gap. You haven't even entered it yet.
The window is closing, not opening
The COBOL panic peaked in 2020–2022. Most large banks have already committed to modernization vendors or built internal teams. Your timing panel says the strategy-lock window closes around late 2027. That gives you roughly 18 months — but you need 12+ of those just to build a credible product and get a pilot running. You're not early. You're not even on time. You're showing up to the wedding during the cake cutting and asking if they've picked a caterer.
COBOL semantics will eat your accuracy claims alive
This is the one that kills the demo. COBOL business logic lives in implicit state across batch runs, undocumented rounding rules baked into financial calculations, and regulatory constraints encoded in structure rather than syntax. Your AI will produce plausible-looking modern code that is subtly, catastrophically wrong in ways that only a domain expert would catch. Banks know this. That's why they don't trust AI tools with mission-critical translation — and why your competitors are still struggling to close deals despite being ahead of you.
⚠ Blind spot
You're thinking about this as a technology problem. It's a liability problem. When your AI mistranslates a COBOL routine that calculates mortgage interest and a bank deploys it, who's on the hook? Banks will demand contractual liability guarantees, errors-and-omissions insurance, and compliance audit trails before they let your tool anywhere near production code. You can't provide any of that. Your competitors who've been live for years are still navigating this. You haven't even thought about it yet, and it's the actual reason banks won't buy — not because the AI isn't good enough, but because no one will sign the indemnification clause.
What would need to be true
At least two mid-tier insurance companies would need to agree to paid pilots within 6 months — not LOIs, not "we're interested," but contracts with money attached — proving the documentation-first wedge actually bypasses enterprise procurement paralysis.
Your documentation output would need to be accurate enough on real-world COBOL (including dialect variations and implicit state) that a domain expert rates it 90%+ correct — because one hallucinated business rule description destroys trust permanently with these buyers.
You would need to secure 18+ months of runway before writing a single line of code, because even the optimistic version of this story involves a long, expensive winter before revenue appears.
Recommended intervention
Stop trying to translate COBOL. Your tech panel nailed it: documentation generation is the real product. Specifically, target insurance companies (not banks — slightly faster procurement, less regulatory scrutiny) and sell an AI-powered "legacy knowledge capture" tool that reads COBOL and produces structured documentation: what each module does, what data flows where, what business rules are encoded, and what test cases would validate correctness. This is lower-risk for the buyer (no one's deploying generated code), faster to prove value (weeks, not months), and genuinely underserved. COBOL Copilot touches this but doesn't own it. Price it as a SaaS subscription per codebase scanned, not per-project. Get three insurance companies using it for documentation, then — and only then — upsell the translation layer once you have the trust, the audit trail, and the case studies. This flips your sales cycle from "let us rewrite your mission-critical code" (terrifying) to "let us help you understand what you already have" (obvious yes).
Intervention unlocking
5seconds
No account needed. One email, no follow-ups.
Want your idea examined? Free triage or full panel →