Studies (AI Analysis)

How it works

You point a Study at an input source (today, Zendesk tickets matched by a search query and optional date range), pick the fields the AI should read, and define a schema — the columns you want filled in, each with a type (boolean, single/multi select, number, short or long text) and optional guidance. Macha runs the same extraction over every record in parallel and writes the answers into the grid. One record in, one row out.

Before any run, a Review screen shows the record count, credit cost per record, and total estimate, so you never accidentally run a 10,000-ticket Study without seeing the cost. Credits are deducted per successful record; errored and cancelled records cost nothing. Runs are capped at 20,000 records.

What it's for

Studies answer questions that span many records — audit the last 5,000 tickets for refund mentions, classify support volume by root cause, find every ticket where a customer raised billing. Doing that by hand is hours of work; a Study does it in one pass and gives you a Report view with per-column charts and drilldowns.

Crucially, results can feed your agents: push a run to a knowledge source and each row becomes a searchable document, so a Study of past resolutions becomes context your support agents can pull during live conversations.