As AI pitches flood the scene, rising investor Jennifer Neundorfer is breaking it to founders that incremental “10x better” features won’t be enough to cut through the noise. The companies that survive will build entirely new behaviors and workflows — not just smarter buttons. It’s also a deceptively simple idea with punishing implications for product design, data strategy, unit economics, and storytelling.
Why New Behaviors Trump 10x Tweaks in AI Products
Neundorfer’s bar for backable AI is simple — don’t polish up an existing flow; imagine anew what it could be. Think tools that transform the way work itself is conducted: creative software where video is created not from a timeline, but from prompts; research assistants that talk back and cite sources instead of dumping links at your feet; clinical documentation that transcribes its own notes based on audio around it so doctors get time back with their patients.
You’re successfully in that zone of category-creating when you’ve got behavior change that can be measured: tasks done that weren’t previously done, time-to-decision that has collapsed from days to minutes, or workflows that leap from low hundred-person teams into high hundreds because collaboration is native, not bolted on.
And if a demo resembles the old process with an AI layer, investor fatigue is inevitable. If it feels like a novel way of working, you may get mindshare — and a sliver of budget.
Proprietary Data and Distribution as Moats
In a market flooded with accessible models, differentiation is moving up to data and down to distribution. Founders should be able to articulate a sustainable data advantage: permissioned datasets others cannot easily access, feedback loops that continuously label domain-specific edge cases, or deployment in the wild resulting in unique telemetry.
That benefit depends on its oversight. Business buyers will want to see proof you can be trusted with sensitive information. What we call SOC 2 Type II, solid audit trails, and role-based access, etc., aren’t “nice-to-haves”; they are table stakes in healthcare, finance, and the public sector. IDC projects that worldwide spending on AI will exceed $500B by 2027, yet that checkbook opens to vendors who successfully pass security review on the first shot.
On distribution, an intelligent wedge matters. Native integrations into systems of record, channel partnerships with cloud marketplaces, and bottom-up usage that turns champions into buyers will always outstrip cold outbound motion. The question: Why can’t a larger platform simply copy you and promote the clone?
Unit Economics in the Age of Inference at Scale
If your AI UX is great, you can cover up some shaky economics. Neundorfer’s advice returns to the fundamentals: provide a credible path to healthy margins as usage scales. The former should always go down as the latter goes up. That includes caching, retrieval augmentation, model distillation, and task-specific fine-tunes for reducing context windows and tokens per task, respectively.
Investors want to dig into your cost-per-task, not merely the cost per token. Will the product be able to deliver gross margin expansion as workloads shift from pilot to production? SaaS-like businesses should be in a position to reach >70% gross margins, which is usually the expectation, although early cohorts may start lower. Multi-model routing will also reduce vendor lock-in and allow you to match model selection with your price, latency, and accuracy requirements.
Most importantly, pay for delivered results. When a customer can see that they save time, with fewer errors or improved conversion rates, your conversations about discounting get shorter and your renewals stronger.
Show That Your Team Wins with Evidence, Not Slogans
With dozens of companies that look alike in each new category, founders are tasked with telling the world why their startup should be the one to get behind. Neundorfer seeks “lived experience” — operators who felt the pain firsthand, engineers with shipping experience at scale, or researchers with peer-reviewed work that maps to the product’s core thesis. “Team–market fit” isn’t a slogan; it’s a story supported by the evidence of artifacts.
Put that into evidence, not platitudes. Bring live demonstrations on actual customer data. Release an evaluation harness with clear benchmarks. Share week-4 and week-12 retention, task-level accuracy vs. human baselines, and partner letters with ROI spelled out. For years, CB Insights has cited “no market need” as the No. 1 reason startups fail; hard proof that users depend on your product beats any slide about TAM.
Read the Market Carefully and Plan Ahead for a Reset
Call it a bubble or call it a boom, but there is almost always a reset. Neundorfer recommends founders build for resilience: raise with 24–30 months of runway, map milestones to the next round’s proof points, and don’t have brittle dependencies on a single model provider or customer. “In enterprise, sales cycles are getting longer; pilots require clear success criteria and a 90-day path to production.”
Deal-gating on compliance and IP questions. Be clear about data provenance, opt-in processes, and how you deal with copyrighted material. An open policy and a flexible implementation model — cloud, VPC, or on-prem — can convert a skeptical prospect to your fastest reference.
What to Put in the Pitch to Stand Out with Investors
The founders who stand out in today’s AI market reveal six things up front:
- The old workflow you’re changing, and how the new one you bring about leads to behavior change
- The set of proprietary data assets (in addition to labeled training sets) that will compound from your product or operation over time
- The path on unit economics as usage scales
- What your go-to-market wedge is and why it gets bigger with scale
- A differentiated team uniquely qualified to use machine learning or deep learning in an unusual way that others can’t easily follow (as well as the ability to attract other people who would gain such advantage by joining)
- What risks and opportunity costs have been mitigated
Keep it short, quantified, and tied to customer pull, not model capabilities.
The throughline in Neundorfer’s playbook is discipline. Develop at the edge of what is possible today and ship in order to meet the demands buyers will require next quarter. But in a crowded field, it is clarity and courage that divide the durable from the disposable — not clever wrappers.