Enterprise Knowledge Search · Sales Automation
Tipalti needed a secure way to give internal teams timely, consistent AI-backed business insights without moving data into external inference services or depending on specialist prompt engineers.
Reported impact
Tipalti democratized internal insight generation, shipped more than ten new AI features to over 50 users, enabled more than 600 ad hoc analyses, and avoided building separate external AI infrastructure while supporting a $75 billion payments business.
Tipalti built an internal AI prompt store on Snowflake so sales and marketing teams can explore enterprise data through reusable, governed LLM workflows. Because its warehouse already runs on Snowflake on AWS, the team kept data in place and layered Cortex AI directly into the platform instead of introducing separate inference infrastructure. Predefined business questions are executed through AI_COMPLETE and TRY_COMPLETE, which return structured results inside SQL workflows and isolate single-row failures. Tipalti also used Snowpark to build a custom parallelization function so multiple prompt executions can run simultaneously and deliver faster answers. Outputs surface in tools such as Salesforce, giving nontechnical users consistent insights without manual prompt engineering or data wrangling. The design reduces overhead today and leaves Tipalti ready for future semantic-model, search, and RAG capabilities on the same governed data foundation.
Internal business and finance data remains in Snowflake on AWS, where predefined prompts call Cortex AI through AI_COMPLETE and TRY_COMPLETE. Snowpark parallelizes execution, and structured answers are surfaced in Salesforce and other daily tools for sales and marketing teams.
AI_COMPLETE integrates LLM capabilities from Cortex directly into SQL workflows
TRY_COMPLETE ensures that any single failure won't cause an entire batch process to fail
The solution enables Tipalti to integrate AI insights into the platforms its teams use every day, such as Salesforce
built an AI-prompt store on Cortex AI
multiple queries can run in parallel
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
LinkedIn's Organization JSON-LD lists addressLocality "Foster City", addressCountry "US", and numberOfEmployees value 1084.
Related
MongoDB Sample May 18, 2026 LekoTech LekoTech needed to digitize millions of unique car parts to promote reuse but faced latency of up to 35 seconds with its existing search system. LekoTech needed to digitize millions of unique car parts to promote reuse but faced latency of up to 35 seconds… MongoDB Atlas · MongoDB Atlas Search · MongoDB Atlas Vector Search MongoDB Sample Apr 29, 2026 Tenali AI Tenali AI needed to help enterprise sellers answer deeply technical questions during live calls in under one second by searching Slack, PDFs, CRM data, and prior meeting content without the latency and synchronization… After switching embedding models, sales reps actively using Tenali's answers on live calls increased by over 40%. MongoDB Atlas · MongoDB Vector Search · Voyage AI 
IBM Sample Jan 15, 2026 IBM IBM needed to reduce the time employees spent on repetitive work across HR, procurement, IT, supply chain, tax, and sales while creating a repeatable enterprise blueprint for scaling AI, automation, and hybrid-cloud… 15K AI agents proposed by employees during the 2025 IBM watsonx Challenge IBM Apptio · IBM Turbonomic · IBM Watson Discovery 
