Process mining applied to AI agent traces shows the paths agents take, where they loop or repeat work and what drives their token consumption — inside the customer's own Snowflake account.
QPR Software Plc has introduced QPR AI Agent Mining for Snowflake, a solution powered by QPR ProcessAnalyzer, QPR’s Snowflake Native App available on Snowflake Marketplace. The solution turns AI agent execution traces into process maps. It shows how agents carry out their tasks, where they loop or repeat work, and which behaviours drive token consumption, latency and errors. It is available now.
As AI agents move from experimentation into production, organisations need to understand how they behave at scale. Monitoring tools show what happened in an individual run and report aggregate metrics such as token usage and latency. What they do not show is the sequence: which paths agents take across thousands of runs, where those paths branch, loop and stall, and which of them consume the budget.
Process mining for a new kind of log
Snowflake records detailed traces of AI agents, including planning steps, tool calls and model responses, in its AI observability event table. QPR AI Agent Mining for Snowflake uses the same process mining techniques QPR has applied to business processes in ERP and other enterprise systems for more than 15 years.
Ready-made dashboards show:
- Which execution path agents take, and how often each occurs
- Where agents loop, retry or repeat work
- How runs diverge from the most common path
- Which paths and runs drive token consumption, latency and errors
"Companies are putting AI agents into production faster than they can see what those agents do. Agent developers need to know why a deployed agent takes the long way round, and AI leaders need to know where their token spend goes. Process Intelligence answers both questions by showing how agents work across thousands of runs, not one run at a time," says Matti Erkheikki, CEO of QPR Software.
"An agent run is not a single request. It is a process. When we turn the traces agents already write into a process map, we see things a trace view cannot show. And because the analysis runs where the data already is, in the customer's own Snowflake account, there is no separate pipeline to maintain and the agents need no changes," says Olli Vihervuori, CTO of QPR Software.
Built for Snowflake
Agent traces are sensitive. They contain user prompts, the business data agents retrieve and the answers models generate. As a Snowflake Native App QPR ProcessAnalyzer analyses the traces inside the customer's Snowflake account, under the access controls and governance already in place. Nothing is exported to a separate analytics platform. Customers also decide the scope: all agents in the account or selected agents only, with or without conversation content. Without content, the analysis still covers tool calls, token usage, latency, models and errors.
From agent behaviour to business outcomes
QPR has already brought Process Intelligence to enterprise AI. Through its MCP server, QPR ProcessAnalyzer gives AI agents operational context about how enterprise processes run. QPR AI Agent Mining for Snowflake takes the next step and turns the lens on the agents themselves: how they carry out their work and what that costs.
With QPR ProcessAnalyzer analysing both the agents and the end-to-end processes they work in, such as order-to-cash, organisations can also measure the business outcomes of agent work: whether the cases agents handle are completed faster, at lower cost and with less rework. This grows in value as agents take a bigger role in end-to-end processes.
Together, these capabilities strengthen QPR’s vision of Agentic Process Intelligence as part of enterprise AI.
Learn more: www.qpr.com/ai-agent-mining
Explore QPR ProcessAnalyzer: https://www.qpr.com/products/qpr-processanalyzer
For further information:
QPR Software Plc
Matti Erkheikki
Chief Executive Officer
Tel. +358 40 717 2570
