Across thousands of startup pitches reviewed, on both sides of the table, the same warning signs keep appearing. None of these is automatically fatal. But each one is a thread worth pulling, because in our experience the thing on the slide and the thing underneath it are often very different.
1. "Our proprietary AI engine"
A startup pitching an AI recruitment platform described its technology as a "proprietary AI engine built from the ground up." There was a whole slide on their "core technology moat." No mention of any third-party model or API.
During technical diligence, the entire product turned out to be a front-end on top of OpenAI's API, with a system prompt and some post-processing. There is nothing wrong with building on a foundation model. Most good products do. The red flag is not the architecture. It is a founder who will not tell you what the architecture actually is.
2. "95% accuracy" with no context
A computer vision startup targeting manufacturing quality control put "95% accuracy" in bold on its metrics slide. No footnote.
The 95 percent was accuracy on their own curated test set. When we asked specifically about false negative rates, the defective products the model misses, the number was closer to 18 percent. For quality control, the false negative rate is the whole game. A number without the conditions that produced it is not a metric. It is decoration.
3. "We have 200,000 data points"
A healthtech startup building a diagnostic assistant cited 200,000 data points as proof of a strong training set, next to a chart implying the dataset was compounding.
The 200,000 came from a single public research dataset that anyone can download. It was not proprietary and it was not growing. When we asked about licensing and patient consent for commercial use, the founder was not sure the dataset's licence even permitted it. It did not. Data volume means nothing without provenance, rights, and relevance.
4. "AI-powered" with no AI team
A fintech deck mentioned AI or machine learning on 9 of 12 slides and described itself as an "AI-first platform." The team slide showed three co-founders: two from consulting, one from sales.
Nobody on the team could build, evaluate, or maintain a model. The plan was to outsource development and manage it as a black box. That works for a demo. It falls apart the moment a model drifts in production or a client asks a technical question during procurement. This was not an AI company. It was a business idea that assumed AI would be added later.
5. "Pre-revenue but scaling fast"
A B2B analytics startup showed user growth from 50 to 2,000 accounts in six months, then a slide with zero revenue and a plan to monetise "once we hit critical mass."
The 2,000 accounts were free-tier signups. Nobody was paying. There was no pricing page, and they had not tested pricing with even a small cohort. Free signups tell you almost nothing about demand for a B2B AI tool. Enterprise buyers do not adopt because something is free. They adopt because it solves a problem worth paying for. That growth chart measured curiosity, not willingness to pay.
The pattern underneath all five
Every one of these is the same failure in a different costume: a claim presented as a conclusion, with the evidence that would support it left off the slide. The job in diligence is not to be cynical. It is to ask, for each impressive claim, what would have to be true for this to hold, and then check whether it is.
If you are weighing an AI investment, acquisition, vendor selection, or training programme, our team is happy to start with a conversation about scope and approach.
The views and findings in this article are shared for general information only. They are high-level perspectives, not legal, financial, regulatory, or other professional advice, and should not be relied upon for any specific decision or circumstance. For guidance tailored to your situation, please consult a qualified adviser.