Product analytics
This page is for the site owner. It opens after the owner account confirms with a passkey.
Product analytics
What people do across the Kalemio products, and where they get stuck.
Where people get stuck
People who started in the selected period and how far each got, counted once per person. Test projects and the owner's accounts are left out.
Where they stopped
Last step reached by people who did not finish the journey.
Failures and abandonment
Jobs and projects created in the period that failed, expired or were removed, by stage.
Active people
Active means the person started a job, project or meeting. For the app, “opened” uses the last time the app contacted the server.
Weekly retention
People grouped by the week they signed up. Each cell is the share active in that week after signing up. Weeks start on Monday (UTC).
Minutes by product and plan
Completed work in the period, rounded up to whole minutes per job. App plans are each person’s current plan.
Minutes by runner
Every transcription through the gateway in the last 120 days. Web products reach it through the owner console. Older jobs have no runner recorded.
Where visits come from
Anonymous website events by campaign tag. Event counts, not people. Country and device are not recorded.
Arabic Subtitles free minutes
Five free minutes a day per network, shared by transcription (on the M1 only) and translation. Networks are counted by a salted hash that changes every UTC day, so the same network on two days counts twice.
Free use by day
Minutes claimed, minus minutes returned by files that never produced subtitles. “At cap” counts networks that used all five minutes that day.
Waiting and processing
Completed transcriptions. Queue: uploaded and paid or claimed until the site sent the file to the gateway. Gateway wait: until a runner took it. Processing: until subtitles were ready.
Free job outcomes
Site: free files. Gateway: free chunks on the M1 pool, where a long file is several chunks.
Free claims refused
Recorded per upload session from the claim itself. “Longer than quoted” means the audio, or a translation’s text, needed more minutes than its quote and no longer fitted.
Free to paid
Cost avoided
Free minutes that ran on the M1, priced at the gateway’s recorded cost per minute.
Arabic Subtitles translation cost
OpenRouter spend per translation job, as OpenRouter billed it. Jobs from before costs were recorded are estimates from their length, priced at today’s OpenRouter rates. Test projects are included, because they are billed too, and shown separately.
Spend by period
Days cover the selected period; weeks (from Monday, UTC) and months cover the last 120 days. Customer spend leaves out test projects.
Free, paid and test
Paid minutes are charged at the refill price less the card fee. Arabic speech is recognised on our own GPUs first; that share is the modelled cost per minute, not a provider bill.
Margin per paid minute
What a paid minute would keep at the cost per minute measured so far, per kind of job.
By model and engine
Translation is the default model; fallback is the previous model, used only when the default could not translate a batch. Speech to text transcribes other languages before translating to Arabic.
Recent translation jobs
Newest first. Minutes are the job’s billable length, rounded up.
Current OpenRouter prices
Read from the OpenRouter models API when this page loads, cached for an hour.
People
Pseudonymous service IDs, most recently active first. Arabic Subtitles has no accounts, so one upload session stands for one person.
Website events
Anonymous daily counts for the selected period compared with the one before it. They count events, not people.
Top pages
Navigation transitions
Media uploads
Arabic Subtitles by media and length; Qarar meetings uploaded or recorded, by length.
Actions and dimensions
Privacy and data controls
Website events are anonymous daily counts without IP addresses. Product records come from the services themselves and are shown here by pseudonymous ID, never by email. A successful owner passkey sign-in excludes its current network from future website events using a one-way digest. Older website rows are daily aggregates without IPs, so they cannot be matched to your network later; use the site and date deletion control to remove them.
Checking current exclusion…