"Evaluate and Enhance an AI Startup"

Why a Specific Approach for AI Startups?

Valuing an early-stage startup is already an art in itself. When it comes to AI startups, the complexity increases even further. Between technological hype, execution difficulties, and real competitive advantages, it is easy to be seduced by promises that struggle to materialize.

A widely used approach by business angels and VCs is the Scorecard Valuation Method. I have adapted it to specifically address the challenges of artificial intelligence.

The Classic Scorecard Method

The traditional Scorecard method is based on a principle of relative comparison. It involves:

1. Establishing a baseline by identifying the average pre-money valuation of comparable startups that have recently raised funds in the same region and sector.
2. Evaluating the target startup according to several weighted criteria:
   • Quality of the founding team (up to 30%)
   • Market opportunity size (up to 25%)
   • Product/technology (up to 25%)
   • Competitive environment (up to 10%)
   • Go-to-market strategy (up to 10%)
3. Applying an adjustment coefficient for each criterion to obtain a global score that will modulate the reference valuation.

Simple and effective, isn't it? However, this generalist approach does not sufficiently capture the specificities of AI startups.

Adapting the Scorecard to AI Startups

AI is not just a technological layer; it's a profound transformation of how products are designed, developed, and deployed (see the dangers of AI investment). The enriched method integrates this reality by retaining the overall structure while refining the analysis.

The first step remains the same: evaluating the startup according to the classic criteria (team, market, etc.). However, for the "product/technology" criterion, which can now weigh up to 50% in the overall evaluation, we apply a sub-scorecard specific to AI.

Here are the key axes to use for evaluating the product and technology:

Evaluation Axis	Questions to Ask	Why It’s Crucial
Product Development	Is the product functional? Early sales? Real customer feedback?	Validates the beginning of product/market fit
Value Proposition	Real customer pain or just an AI gimmick?	Avoid "AI washing" projects. Important note: sovereignty alone cannot constitute a sufficient value proposition.
Proprietary Data	Proprietary data? Mastered fine-tuning?	In a world where models are becoming generalized, data often becomes the real competitive edge.
Scalability	Large-scale deployment possible? Solid infrastructure? Access to significant capital.	The scale effect and thus access to significant capital can make a difference.
Differentiation	Entry barriers? Does the startup have a relevant vertical specialization?	Generic AI models will be able to meet 80% of needs without these startups. It's generally better to start in a specific niche. General use cases will be addressed by the big tech companies.

Typology of AI Startups and Specific Weightings

Each type of AI startup has its own priorities. Here is a summary table of weightings to apply according to the typology (see my segmentation of AI startups).

Type of Startup	Prod. Development	Value Proposition	Proprietary Data	Scalability	Differentiation
Fundamental Models	20%	0%	0%	80%	0%
Infrastructures	20%	0%	0%	80%	0%
Fine-Tuners	20%	20%	60%	0%	0%
AI Tools	20%	50%	0%	0%	30%
AI Governance	30%	50%	0%	0%	20%
Solutions	20%	20%	20%	0%	40%

Let's illustrate this methodology with a concrete example: Mistral AI, which is developing a new artificial intelligence model. As a startup in the "Fundamental Models" category, the critical factor for its success will be its ability to mobilize significant financial resources (funding) and attract top talent.

Conversely, for a startup building a solution based on existing AI models, the determining factor will be its ability to effectively target a specific niche and provide a unique added value, perfectly adapted to the particular needs of that vertical segment.