Why these mistakes matter
This page covers companies building their own models, ML infrastructure or research-led products, including spin-outs. It focuses on the technology itself; the artificial intelligence page covers products built on existing models.
Most rounds do not fail on one big problem. They stall because an investor finds something the founder should have fixed or explained earlier, and confidence drops. Each mistake below says why it happens, how investors tend to react and what to do instead.
Mistake 1: University IP left unresolved
Why it happens: Spin-out negotiations are slow and founders start building anyway.
How investors react: Investors will not fund a company that may not own its core technology.
What to do instead: Agree the licence or assignment before raising, and understand its royalty and equity terms.
Mistake 2: Benchmarks nobody can reproduce
Why it happens: Results were produced once under ideal conditions.
How investors react: Technical diligence exposes the gap and trust falls.
What to do instead: Publish or share reproducible experiments and compare against fair baselines.
Mistake 3: Compute budgets with no room for failure
Why it happens: Plans assume each training run succeeds.
How investors react: Investors expect overruns and doubt the runway.
What to do instead: Budget for failed experiments and show how you control compute spend.
Mistake 4: Research with no buyer
Why it happens: The technology is exciting in itself.
How investors react: Investors ask who will pay and when.
What to do instead: Link each technical milestone to design partners and a commercial use.
Mistake 5: Hiring before the milestone is defined
Why it happens: Talent is scarce, so teams hire early.
How investors react: Burn rises without clear progress.
What to do instead: Define the milestone first and hire to reach it.
Mistake 6: Overstating the moat
Why it happens: Founders assume a better model is defensible.
How investors react: Investors know open research moves fast.
What to do instead: Explain what compounds over time: data, deployment, integrations or speed.
Mistakes every sector shares
Alongside the sector-specific points, these general errors come up in almost every round:
- Raising without a clear milestone the money is meant to reach.
- A messy cap table or missing IP assignments found late in diligence.
- Pitching investors who do not back your sector or stage.
- Starting to raise with too little runway left to negotiate calmly.
Machine learning: common mistake versus better approach
| Mistake | Better approach |
|---|---|
| University IP left unresolved | Agree the licence or assignment before raising, and understand its royalty and equity terms. |
| Benchmarks nobody can reproduce | Publish or share reproducible experiments and compare against fair baselines. |
| Compute budgets with no room for failure | Budget for failed experiments and show how you control compute spend. |
| Research with no buyer | Link each technical milestone to design partners and a commercial use. |
| Hiring before the milestone is defined | Define the milestone first and hire to reach it. |
| Overstating the moat | Explain what compounds over time: data, deployment, integrations or speed. |
Questions to prepare before you pitch
- Do we own or licence all our core IP?
- Could an outside engineer reproduce our results?
- What is our compute budget if half our runs fail?
- Which customer benefits from our next milestone?
How KJ Enterprises evaluates machine learning businesses
KJ Enterprises backs ML-first companies with clean IP, honest benchmarks and technical milestones tied to real customer value.
If that describes your company, you can apply for investment. Related reading: the complete funding guide and the step-by-step how-to and costs and terms and the readiness checklist and applied AI mistakes to avoid.