The Deal Was $4M. Diligence Killed It.

A SaaS founder built an AI-powered customer success platform. $4.2M ARR. 340 paying accounts. 18% MoM growth. Two strategic acquirers at the table. The founder expected a 6x revenue multiple, roughly $25M. She got zero. The deal collapsed in due diligence.

Not because the product was bad. Not because the market shifted. Because three IP gaps, each fixable in advance, made the buyer's counsel walk away.

This is a composite. The details are drawn from patterns identified by Mintz LLP, a law firm specializing in IP and M&A for AI companies. The numbers are directionally accurate across dozens of similar exits I have reviewed through Angel Investors Network's deal flow.

The founder made the same three mistakes I have seen tank deals for 20 years. Here is what went wrong.

Gap 1: Missing IP Assignments

The founder's first two engineers joined before formal employment agreements were in place. They wrote the core inference engine. Three years later, both had left the company.

Mintz identifies missing IP assignments as the single most common diligence issue in software acquisitions. By default, individuals own the IP they create. If your engineers and contractors built your core technology, you need written assignments transferring those rights to the company. Present-tense assignment clauses covering all inventions, code, models, and related IP, applied consistently across employees and contractors from day one.

The buyer's counsel asked for assignment agreements for every contributor to the core codebase. The founder had agreements for employees hired after month 14. She had nothing for the two engineers who built the foundation.

The buyer's legal team flagged it as a "chain of title" risk. If either engineer decided to assert ownership, the company's most valuable asset was contested. The buyer was not going to litigate someone else's paperwork failure.

Estimated valuation impact: 20-30% haircut, or deal collapse. In this case, it was deal collapse.

The fix, in advance: Get IP assignment and confidentiality agreements in place from day one. For every employee, every contractor, every advisor who touches code, models, or data. Retroactive assignments are possible but expensive and sometimes impossible. Do it early.

Gap 2: Open-Source License Contamination

The inference engine used a modified version of a library released under a copyleft license. The modification was minor, about 200 lines of code. But the license terms required any derivative work to be released under the same open-source license.

The customer-facing product, the one generating $4.2M in ARR, was a derivative work.

Mintz warns that if you have modified copyleft-licensed repositories, you may be obligated to open-source your derivative code. Internal tooling is lower risk. Customer-facing products are not.

The buyer's technical diligence team ran a software composition analysis. They found the copyleft dependency in the core product. The founder's engineering team did not even know it was there. A junior developer had pulled the library two years earlier to solve a specific problem. No one reviewed the license.

The buyer's concern: if discovered post-acquisition, the company would face a choice between open-sourcing its core IP or rebuilding the affected module. Either path destroyed the acquisition thesis.

The fix, in advance: Run a software composition analysis before diligence starts. Identify every open-source dependency. Review the license terms. Replace copyleft dependencies in customer-facing code with permissive alternatives. This is a weekend of engineering work, not a six-month project.

Gap 3: Training Data Provenance

The AI model was trained on three datasets. Two were commercially licensed with clear terms. The third was scraped from public websites without explicit permission.

Mintz notes that for AI companies specifically, you need a defensible legal basis for every dataset used in training, especially personal, biometric, or high-risk data. Buyers will ask you to trace training data sources, show applicable licenses and restrictions, and prove your customer disclosures are accurate.

The buyer's counsel asked for a training data manifest. The founder produced documentation for the two licensed datasets. For the third, she had a Slack message from an engineer saying "scraped this from [source], seemed fine."

That is not a defensible legal basis. It is a liability.

The concern was not theoretical. The EU AI Act and emerging US state laws impose increasingly specific requirements on training data provenance. A buyer inheriting undocumented training data inherits the liability. No acquirer wants to close a deal and then receive a cease-and-desist letter from a data subject or content owner.

The fix, in advance: Document every dataset. Source, license terms, date acquired, scope of use, any restrictions. Create a training data manifest. Keep it updated. If you cannot defend the provenance, retrain the model on defensible data before you enter a process.

The Pattern

I spent years at Hartford Steam Boiler as an innovation scout inside Munich Re. We evaluated technology companies for partnership and investment. The diligence process was relentless. One recurring pattern: founders who built brilliant products but treated legal infrastructure as an afterthought. They would come to the table with strong revenue, strong product-market fit, and IP structures held together with handshake agreements and Slack messages.

The IBBA 2026 Market Pulse Report confirms the timing problem. Owners who began formal exit preparation less than two years before listing closed their deals at a median 15% to 25% below their initial asking price. Owners who prepped for three years or more closed within 5% of ask.

IP cleanup is not a six-week project you run after you sign a letter of intent. It is a three-year project you start the day you incorporate.

The 7-Point IP Checklist for AI SaaS Founders

If you are building an AI company and plan to exit within five years, lock down these seven items now.

1. IP Assignments. Written assignment clauses for every employee and contractor. Present-tense, covering all inventions, code, models, and related IP. Signed and filed. No exceptions.

2. Software Composition Analysis. Audit every open-source dependency. Replace copyleft libraries in customer-facing code. Document permissive licenses.

3. Training Data Manifest. Every dataset: source, license, date, scope of use, restrictions. Update it quarterly.

4. AI Governance Documentation. Written policies on how you build, train, deploy, and monitor AI. Buyers expect this now. At minimum: model evaluation criteria, bias testing procedures, incident response protocols, and data handling policies.

5. Third-Party Licensing Agreements. Explicit foreground IP allocation. Clear carve-outs for your background IP. No default joint ownership. Scope licenses narrowly by field of use, product, or statement of work.

6. Patent Strategy. Evaluate whether provisional patents protect key innovations. Even if you do not plan to enforce them, patents signal defensibility to buyers.

7. Data Room Readiness. Organize all IP documentation into a data room structure before you need it. Cap table, articles of incorporation, all agreements, training data manifests, governance policies. A buyer who receives organized documentation moves faster and offers more.

The Real Cost of Waiting

The founder in this composite eventually cleaned up her IP. It took eight months and cost $180,000 in legal fees, engineering time, and model retraining. She found the two original engineers and negotiated retroactive assignments. She replaced the copyleft dependency. She retrained the model on defensible data.

She went back to market a year later and closed at 4.5x revenue. Not the 6x she originally expected. The delay cost her market timing, buyer goodwill, and momentum.

If she had done this work three years earlier, it would have cost $15,000 to $25,000. The ROI on proactive IP cleanup is not abstract. It is a direct valuation multiplier.

> Doctrine Connection: Due diligence is non-negotiable. Buyers do not take your word for it. They verify. Every assignment agreement, every license, every dataset. The founders who treat IP infrastructure as seriously as they treat product development are the ones who close deals at asking price. Everyone else pays the founder dependency tax.

FAQ

Q: When should I start IP cleanup if I plan to exit in 3 to 5 years?

Start now. IP assignments should be in place from incorporation. Software composition analysis should run quarterly. Training data manifests should be maintained continuously. The earlier you start, the cheaper and cleaner the process.

Q: What if my original engineers are gone and I have no IP assignments?

Contact them and negotiate retroactive assignments. This is common. Most former employees will sign for a small consideration ($1,000 to $5,000). If they refuse, consult IP counsel immediately. The longer you wait, the harder this becomes.

Q: Does using MIT or Apache-licensed open source create risk?

Permissive licenses (MIT, Apache 2.0, BSD) are generally safe for commercial use. The risk is copyleft licenses (GPL, AGPL, LGPL) that require derivative works to be released under the same license. Run a software composition analysis to identify which licenses apply to your dependencies.

Q: Do I need patents to exit?

Not always. But patents signal defensibility. Even provisional patents, which cost $2,000 to $5,000 to file, show buyers you have protectable innovation. For AI companies specifically, patents on novel training methods or inference architectures can add meaningful valuation.

Q: How much does a full IP audit cost?

For a seed-to-Series A company: $15,000 to $50,000 depending on complexity. For a later-stage company with multiple products and data sources: $50,000 to $150,000. Compare this to the valuation impact of unresolved IP gaps, which routinely exceeds $1M.

*Jeff Barnes, MBA has no personal position in any company, fund, or platform named in this article. demg.ai has no current commercial relationship with any party mentioned. demg.ai provides marketing education and systems consulting, not investment advice. Past performance does not guarantee future results.*