17 articles in this tag
Five practitioner lessons from building a ~6,900-company venture-outcome graph — identity is the real bottleneck, the dangerous errors are the plausible ones, survivorship hides in track records too, the founder is the unit not the company, and consensus isn't correctness.
NUVC runs on 24 agentic teams and zero employees. Here's what one ordinary production build — a venture-outcome graph, schema to dry-run to rollback to commit — actually looked like end to end, and why the judgment is still the whole job.
AI collapsed the cost of producing work toward zero. It did nothing to the cost of checking that work. So the bottleneck moved — from writing to reviewing — and leverage now belongs to whoever can verify fastest, not generate fastest.
Everyone reports how fast AI made their team. Almost nobody subtracts what it could lose them. A two-axis framework — Risk and Return — for judging your team's AI honestly, from an investor who runs a company on it.
AI is an amplifier — it multiplies whatever judgment is already there. Here is how a non-technical founder measures their own AI judgment, and their agents', without reading a line of code.
Almost every startup-scoring model is trained on companies that survived — because failures leave no structured trace. The fix isn't a cleverer architecture, it's the negative class: label the dead ones and weight them equally. Field notes on wiring a real failure class into a training corpus, and the two things that broke.
We built an outcome graph of ~6,900 accelerator companies — every Y Combinator batch and publicly announced Startmate cohorts — and linked each to what happened next. Three lessons: failure is the training data everyone throws away, selection and success are different instruments, and the outcome is the only signal that doesn't flatter you.
Venture returns follow a power law — roughly half of investments lose money and a handful return most of the profit. AI changes how cheaply you can rank companies, not the shape of that distribution. Why the right move is to let the machine raise the floor and keep the ceiling human.
AI can analyse a pitch deck across 7 investor dimensions in under two minutes. Here is what that means for founders — what AI finds, where most decks fall short, and how the evaluation process is changing.
NuScore evaluates pitch decks across 7 dimensions using the same framework real VCs use — but with AI precision and 980+ validated training examples. Here is the full methodology, what each score tier means, and why the science behind your number matters.
Your NuScore is not a grade — it is a map. It shows you where VCs will push back, what to fix first, and how likely you are to raise. Here is exactly what each number means and how to use it.
Technology adoption curves are compressing with every generation. From the telephone to TikTok to ChatGPT to OpenClaw — and what each acceleration means for how venture capital works.
$5.1B invested. 61% went to AI. Victoria beat NSW for the first time. We cross-referenced the State of Funding report with 197 NuScore evaluations to find what actually drives investment decisions.
Digital operating models eventually cross and surpass traditional ones as scale grows. That crossing point has now arrived in venture capital — with documented evidence from 2020 to 2026.
Black-box AI scoring is a trust killer. Here's how NUVC's explainability layer breaks down every score into traceable signals, confidence levels, and evidence you can verify yourself.
Everyone now has access to ChatGPT, Claude, and Gemini. But the gap in outcomes between investors using these tools is widening — not closing. Here's what separates the ones getting real leverage from the ones still copy-pasting decks.
Our AI analysed 298 decks from companies worth $116B combined. Here's what the data says about detecting outlier potential, conviction archetypes, and the signals that predict which startups become unicorns.
Browse all tags
Want to know how your pitch deck scores?
Upload Your Deck — Free