DigestAI news desk

Cut through the AI noise.

Enterprise & Industry2 min read

Ramp’s Glass AI cuts spend management costs by 30% overnight

Ramp introduced Glass, an internal AI tool, after spending $1.5 million in AI tokens in a single week in May. The company compared this to giving employees expensive instruments without a conductor—expensive but uncoordinated. Glass now automates tasks like extracting feature requests from customer calls, reducing manual data entry from 300 monthly entries to capturing 20,000 monthly requests…

1 source

Key points

  • Ramp spent **$1.5 million in AI tokens** in May, prompting the creation of Glass to streamline internal workflows
  • Glass automates feature requests from customer calls, capturing **20,000 monthly entries** vs. prior manual **300/month**
  • Ramp Router optimizes model use, cutting internal AI costs by **30% overnight** and valuing Ramp at **$60 billion**

The tool reflects broader enterprise AI challenges: 62% of companies spent over $10 million on new AI tools in the past year, according to a PYMNTS Intelligence report from September. However, deeper adoption faces hurdles like skill gaps and unclear ownership. Ramp’s $60 billion valuation (post-$1.85 billion funding round) underscores its focus on scaling AI tools internally before broader deployment.

Full story from pymnts.com · by PYMNTS · via Reddit AI communitiesOpen source ↗

Ramp’s $1.5 Million Week of Spending Leads to Glass AI Tool

pymnts.com · 14 October 2026

Loading the full article…

This text was published by pymnts.com and written by PYMNTS. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗

Coverage and discussion

1source
Topics · follow one to build your own front page
RampPYMNTS Intelligence

The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.

Comments

via GitHub Discussions

More in Enterprise & Industry

All →

Related stories