Strategy Becomes Numbers
Strategy is only valuable if it shows up in the metrics. The vertical AI thesis — domain expertise, workflow embedding, proprietary data, switching cost — all ultimately produces specific financial outcomes that distinguish genuinely defensible businesses from hype. Net revenue retention is the metric where that distinction shows up most clearly.
The NRR Math
For a horizontal AI tool with real churn risk — a general-purpose productivity assistant, an undifferentiated writing tool, a simple API wrapper — NRR might be 85-95%. Customers cancel when they find alternatives, when their budgets are cut, or when the product stops feeling essential.
For a well-embedded vertical AI platform with expansion modules and genuine switching cost — a healthcare revenue cycle system embedded in Epic, an insurance underwriting platform integrated with claims processing — NRR of 115-130% is achievable. Customers don't cancel because the switching cost is too high. They expand because the product has natural extension points into adjacent workflows.
That 30-point spread, starting from a $20M ARR baseline, compounded over five years: at 120% NRR, existing customer revenue grows to nearly $50M. At 90% NRR, it shrinks to under $12M. The gap is $38M — from the same starting point, with the same customer count, from nothing more than the difference in retention and expansion dynamics. No sales advantage can close a $38M structural gap.
The NRR Compounding Effect
Burn Multiple: The Efficiency Signal
Beyond NRR, the burn multiple is the key efficiency metric in the repriced market. Net cash burned divided by net new ARR. Below 1x means elite — you're generating a dollar of ARR for less than a dollar of cash spent. Between 1-2x is efficient and fundable. Above 3x is concerning in the current environment where investors expect capital discipline.
Burn multiple matters more now than it did in 2021. When capital was effectively free, burning $3 to generate $1 of ARR was defensible as long as growth was fast enough. In a repriced market with 5% risk-free rates, the math is different. Investors underwriting to real exit distributions need efficiency to be real, not projected.
When the Numbers Don't Match the Story
The most revealing moment in any AI investment evaluation is when you ask for the cohort data. Not just annual revenue, not just total ARR — the actual retention and expansion data broken out by customer cohort and vintage.
Companies with genuine vertical AI defensibility are proud to show this data because it tells the story of compounding retention better than any pitch can. Companies that resist sharing it usually have NRR that doesn't match the switching cost narrative.
When retention and switching cost are real — not just claimed in a deck — downside risk compresses. That matters more than people think when you're modeling exits under uncertainty. The companies with 120%+ NRR don't just generate better returns. They generate more predictable returns, which is exactly what investors need to model confidently.