AI-Scored Business Model Analysis of Recently Funded AI Companies

Investors, corporate innovation teams and strategy functions that need to separate genuinely new AI business models from familiar ones with AI in the marketing copy. This case is most useful if:

• you track a fast-growing set of AI startups and AI-enabled offers, many recently funded
• analysts read pitch pages but score them inconsistently
• you want AI to do the first pass without inflating every score
• the assessment criteria need to evolve without losing comparability

The market for AI-enabled companies and offers was growing faster than anyone could assess it by hand. Venture funding, launch announcements and polished websites made almost every company look differentiated.

The client needed a repeatable way to find the small share of offers that were structurally different: by the brief's own estimate, the 2 to 5% that were not conventional services with AI in the copy.

We built Sensing Studio, a market sensing instrument in which an AI model scores each offer against a structured business-model rubric and a human analyst makes the call.

• A multi-factor rubric: seven dimensions scored from 1 to 5 (operating-model novelty, system depth, where AI sits in delivery, proprietary leverage, edge specificity, pricing posture and method legibility), plus a categorical read on team shape. Maximum 35.
• AI scoring with calibration: one command fetches an offer page and has Claude score it with a calibration-first prompt (a median of about 2.5, with 5s rare), and an explicit "insufficient evidence" exit so thin pages are not scored at all.
• Analyst workflow: each offer gets a verdict (Pursue, Watch or Pass), a one-line "why interesting" note, per-dimension notes, a shape class and red flags, reviewed in about 15 minutes a day.
• Sources and signals: 21 seed sources in three tiers, including venture portfolios and launch platforms, with funding and other market signals recorded alongside each firm across 20 industries.
• A versioned rubric: every score records the rubric version it was made against, so criteria can change without breaking comparisons.

52 firms and 49 offers assessed on one consistent rubric, with the results carried forward into the client's wider intelligence platform.

AI that refuses to flatter. Calibration and the insufficient-evidence rule stop the model from rating every well-funded company as exceptional.

Human judgement where it counts. The model does the reading and first scoring; the analyst decides what to pursue.

When every company claims to be AI-native, the scarce skill is telling the difference. A calibrated rubric makes that judgement repeatable.