How Ramp builds customer-first AI

Summary

Ramp describes principles for building customer-focused AI, contrasting embedded outcomes with superficial chatbot experiences it calls “AI washing.” The approach emphasizes integrating models into workflows, separating general models trained on aggregated and masked data from sensitive models that use private customer data temporarily through in-context learning, and requiring explicit opt-in for sharing contract data. It prioritizes user control and feedback over lengthy explanations, while using guardrails that constrain outputs through predefined interactive data blocks rather than relying only on censorship filters. Ramp cites contract term extraction, spend intelligence, pricing benchmarks, Copilot, and Jsonformer as examples of these practices. The stated goal is AI that saves customers time and money while protecting privacy and producing safer, more useful results.

Context

Ramp argues that many AI products provide superficial chat experiences without meaningful customer value, while AI systems also raise privacy, trust, and safety concerns. The principles are intended to keep models focused on saving customers time and money.

Approach / What changed

Ramp embeds AI in customer workflows, separates general models from sensitive models using private data temporarily through in-context learning, requires explicit opt-in for sharing contract data, incorporates user feedback, and constrains outputs with structured interfaces and techniques such as Jsonformer.

Takeaways

• General models use aggregated and masked customer data for common tasks, while sensitive models can temporarily use private customer data through in-context learning without storing it.
• Ramp favors customer feedback and model improvement over lengthy explanations, such as letting users correct spend-intelligence categorization errors for future learning.
• Copilot generates search queries from customer data but presents results in predefined interactive blocks, reducing the risk of unwanted plain-text output.

Related reading