Older conversations can be tagged and scored the same way new ones are, so your reports and Year in Review show real topics and sentiment for past months instead of "Unclassified" or "Insufficient data".

How classification works

When a chat is resolved, Velaro reads the transcript and fills in line of business, intent, outcome and product. With automatic tagging on, line of business and intent also become conversation tags. A second step scores each visitor message for sentiment, using an English word list rather than AI, and records where the bot handed off to a person. Both steps run on the same chats the same way every time, so a past chat gets exactly the result it would have got on the day it closed.

Chats whose text has already been cleared after archive cannot be read again. They keep showing "Insufficient data" for the sentiment parts, never an estimate. Everything else is classified from the text that is still on file.

How we recommend tags

You do not need to design a tag list first. Run the sample: Velaro classifies 20 recent conversations and tells you the mix it found, for example "home 8, auto 6, renters 3" for line of business and "claim 9, billing 5" for intent. That mix is the set of tags we would apply to your history. It reads your own chats, so the proposal reflects your business.

If the proposal looks right, approve it by running the full history classification. If a tag is wrong, correct it on any conversation. Corrections feed back into the tagger, so the same mistake is not repeated.

Our process

1. Sample. Run history-classification-sample for a date range. Nothing is written beyond the 20 sample tags.

2. Propose. Read the mix in the result.

3. Approve. Run history-classification for the range you want. It first shows what it would do (a preview). Add apply to run it.

4. Backfill. Applying queues the run and returns right away with a run id; it finishes in the background, and you read the result with the status check. Only one run per site goes at a time, so asking again while one is going just returns that run. Each run handles up to 100 conversations and skips any chat already looked at (including chats with nothing to tag), so you can run it again and it continues where it stopped, and it tells you when nothing remains. A range can be at most one year at a time.

5. Results appear. Enterprise Reports and Year in Review pick up the new tags and sentiment on their next refresh.

Run it

You must be a manager of the site, and AI Field Extraction must be turned on in Chat Intelligence settings. If it is off, the run tells you so and changes nothing. Your monthly AI tagging allowance still applies; if it is used up, the run pauses and says so.

With the Velaro command line:

velaro report-backfill history-classification-sample --from 2025-10-01 --to 2026-09-30
velaro report-backfill history-classification --from 2025-10-01 --to 2026-09-30
velaro report-backfill history-classification --from 2025-10-01 --to 2026-09-30 --apply
velaro report-backfill history-classification --status

The same actions are available to your AI assistant through the Velaro MCP tool report_backfill (pass status to read the latest run).

What you will see

History is classified rather than estimated. Where a chat could not be read, the report still says "Insufficient data" for that part, and it tells you how many chats that covers.