How AI Applications Earn Their Premium
Summary
- Lower inference costs improve margins but enable customers to replace application parts with general-purpose models, making pricing critical.
- Applications face vulnerability when customers can reproduce core outputs using general models with minimal effort.
- Founders must rigorously develop and price domain expertise, deep integrations, reliable execution, & hard-to-access context.
- Repsol's legal team chose Harvey over Microsoft Copilot & an internal prototype after an eight-week blind pilot showed higher quality, accuracy, & time savings.
- Harvey enabled lawyers to save four to six hours weekly with 96% adoption & improved contract templates by identifying recurring objections.
- Three rules for sustained success are: operate workflows better than customers can, make context cheaper to rent than to rebuild, & prove value before pricing.
- Vendors must outperform internal alternatives across full workflows, including applying standards, ensuring edit consistency, & preserving negotiation context.
- Specialized applications like Harvey use specialized agents to reconcile edits, evaluate performance against rubrics, & preserve review context for future learning.
- Context justifies premiums through exclusive access or by making preparation of customer-owned data cheaper than rebuilding it internally.
- Vertical software companies can spread the cost of collecting & validating proprietary data across many customers to earn margins.
- Applications earn fees by handling permissions, normalization, retrieval, & validation of customer data that buyers cannot easily manage alone.
- 94% of enterprise LLM decision-makers expect total spending to increase in the next year per an August 2026 survey.
- Pricing should be based on recognizable work units like contract reviews or resolved support tickets, including full delivery costs & failure rates.
- Decagon's solution for Notion delivered a 34% improvement in ticket resolution time, validating the choice to buy over building internally.
- Applications should price at the highest measurable layer of value rather than per token to ensure predictability & profitability.
- Pricing models should separate fixed orchestration costs from volatile model inputs to handle varying workload economics effectively.
- Margins can improve through routing, caching, & better models while keeping the offer attractive relative to customer alternatives.
- The test for earning a premium is whether buying offers a better combination of total cost, quality, speed, & risk than the customer's best alternative.
- Application advantages must widen faster than model progress to sustain premiums as customer alternatives improve.
Lower inference costs can improve margins for application companies, but more capable models and easier development tools can also make parts of an application easier to replace. The commercial question is what customers will continue to pay for as their alternatives improve. This exposure varies by product. An application is vulnerable when a customer can...
- Keywords
- Growth