From 51 FlexLM CSVs to conversational license analytics

Jul 7, 2026 · case study

A chip simulation waits when FlexLM denies a license checkout. The software licenses behind one simulation can cost six figures, and a project runs thousands of simulations. Too few seats stop engineers; too many leave expensive capacity idle.

Bazinga Labs builds design-automation software for teams managing that tradeoff. FlexLM already records every checkout and denial (feature, user, host, timestamp, granted or refused), but turning those events into a capacity decision still required manual analysis.

One question meant a week of exports

The working set began as 51 CSV exports on NFS and S3, across 9 test runs and teams of up to 650 engineers. Answering a license question meant exporting, joining, and reading the result in a spreadsheet, typically about a week of work.

The questions were concrete:

Which features denied the most checkouts, and in which week?
What was peak concurrent usage for each feature?
Who held a license without using it?
Which licenses were under-subscribed, over-subscribed, or in range?

Fifty-one exports became one queryable dataset

Bazinga consolidated tens of millions of rows into four queryable datasets: license events, a utilization series, and feature and denial dimensions.

Once the events were a table, denial counts by feature and week were one aggregation:

denial_storms.sql

SELECT   feature, date_trunc('week', ts) AS week,
         count(*) AS denials
FROM     license_events
WHERE    event = 'DENIED'
GROUP BY feature, week
ORDER BY denials DESC LIMIT 15;

A question produces inspectable SQL

Bazinga put a conversational layer in front of the tables. A Bazinga customer asks a license question in English; the layer writes SQL, runs it against Infino, and returns a chart with the query one click away. The saved SQL makes every answer inspectable.

Conversational analytics empty state, with suggested questions about denials, idle holds, and peak concurrent usage — Suggested questions for this dataset: denials, idle holds, peak concurrent usage. Screenshots use sample feature names and counts.

For “Which features have the most denials?”, the result is the aggregation above, a horizontal bar, and a short summary:

A question about which features have the most denials, answered with a horizontal bar chart, the SQL, and a short summary — A question, the SQL that ran, the chart.

Dashboards

A customer can pin the resulting chart beside under-subscribed licenses, over-subscribed licenses, idle holds, and quarterly peak usage. On the next load, the dashboard executes the saved query and chart specification against the current data.

License utilization dashboard with denials by feature, peak concurrent licenses, a denials metric, and a table of idle holds — Charts from the conversation, pinned.

The model authors the dashboard; the application serves it

The conversational model translates a question into SQL and produces the chart object. After pinning, the application reruns the saved SQL and chart specification directly against Infino on each dashboard load.