Howzer v2.5.2-DE: everything the analysis knows, in every answer
The analysis was always deeper than what it handed back. Now the full picture comes with every response, purely as an addition, while answers arrive 17 percent faster and long messages are triaged far more accurately than before.
By Howzer Team, Product · product ·
Twelve analyses run on every incoming message. Until this release, the answer you got back contained a summary of them. Everything else stayed inside. v2.5.2 opens it up, adds nothing you have to adapt to, and makes the whole thing faster while it is at it.
What is in the answer now
- How the sentiment breaks down, not just which way it leans, plus the detail behind the emotion and root cause.
- How confident the urgency assessment is, and the four separate dimensions it is built from.
- Which risk indicators fired, and the escalation entry with its deadline, its reasons and a recommended next step.
- Whether someone is signalling a switch to a competitor, where the message came from geographically, and a language and text section.
- Reference numbers appear as type and count only. The actual numbers never leave your system.
All of it is added alongside what was already there, so an existing integration keeps working without a single line changed.
Faster, with identical results
Analyses that do not depend on each other now run at the same time instead of queueing up. A typical message comes back 17 percent faster, and the results match the old ones field for field. Writing to storage was batched as well, which took a heavy load off the database.
Long messages, properly triaged
The urgency model can now read four times as much of a message at once, and it was retrained with long examples. Short messages were never the problem and stayed excellent. Long ones were, and are now much closer.
Churn warnings, re-tuned on real volume
The point at which Howzer warns you that a customer may be about to leave was re-tuned against ten thousand real messages. It now catches 80 percent of them instead of 72, and it does so without putting noticeably more noise in the queue.