Shopify
Stripe
Slack
Notion
Google Workspace
Confluence
What Is Macha Studies? Run AI Over Up to 20,000 Tickets at Once
Macha Studies runs an AI model across a whole set of Zendesk tickets at once and returns one structured row per ticket, with columns you define. Point it at a year of tickets and you get a table of contact reasons, resolutions, sentiment or categories you can export or give to your agents.
Key takeaways
- Macha Studies runs a chosen AI model over Zendesk tickets, up to 20,000 per run, and returns one structured row per ticket with columns you define.
- Studies columns can be Yes/No flags, single or multiple choice, numbers, or short or long text, so one feature can build a knowledge base, score sentiment or tag a backlog.
- Before a full Studies run, Macha suggests testing on a sample of 10 to 50 records, then shows the record count and cost estimate on a Review screen.
- Studies results can be sorted, charted in a Report view, exported to CSV, or pushed into a knowledge source that linked Macha agents search.
- Studies is included in Macha's one plan, which starts at $299 a month for 750 tickets, and also runs on the free trial.
Macha Studies runs an AI model over a set of Zendesk tickets, up to 20,000 per run, and fills in columns you define, so you get one row of structured results per ticket.
| Question | Answer |
|---|---|
| Input source today | Zendesk tickets, matched by a Zendesk search query and an optional date range |
| Column types | Yes/No, single choice, multiple choice, number, short text, long text |
| Run limit | 20,000 records per run |
| Before a full run | A test on a small sample, then a Review screen with record count and cost estimate |
| What you do with results | Sort, group, open the report, export to CSV, or push into a knowledge source |
| Plan | Included in Macha's one plan, from $299 a month for 750 tickets, and on the free trial |
What does Macha Studies do, in one line?
Studies runs your chosen AI model over a set of records and fills a structure you define, producing one row of results per record. Think of it as turning thousands of messy tickets into a tidy spreadsheet where every column is something an AI read and decided. The sidebar calls it "AI Analysis".
How do you set up a Study?
Every Study is a four-step wizard:
- Source: pick what to run over. Today that's Zendesk Tickets.
- Scope: choose which records (any Zendesk search expression, such as
tags:refund, plus an optional date range) and which fields the AI reads from each one. - Extract: define the columns you want filled, pick a model, and add guidance per column. This is the heart of it: each column is one piece of information pulled from every record.
- Review: see the record count and cost estimate, then run.
The columns can be any shape: Yes/No flags, single or multiple choice from a fixed list, numbers, or short or long text. That flexibility is why the same feature can build a knowledge base, run sentiment analysis or tag a backlog: you're only changing the columns.
On the Scope step, Subject and Description are enough for most jobs. Custom fields and the full comment thread are available too, but the docs flag both as expensive, because they add API calls and make the prompt much larger.
What can you use Studies for?
Because you define the output, Studies is a general-purpose "run AI over my data" engine. The common jobs:
- Turn past tickets into a knowledge base: extract the question and the resolution from each ticket, then push the results into a knowledge source your agents search. (Full how-to.)
- Analyze a year of support: sentiment, true contact reasons, refund, bug and churn signals, trends. (Full how-to.)
- Tag or categorize a backlog: apply consistent categories across thousands of old tickets that were never tagged.
- Find knowledge gaps: flag which questions your help center couldn't answer, so you know what to write.
- QA and compliance: score tickets against criteria you define.
If you can describe what you want pulled out of each ticket, Studies can extract it.
How do you keep a big run from going wrong?
Three design choices keep a large run under control:
- Test run first. Run the same extraction on a small sample (the docs suggest 10 to 50 records), check the answers, tighten the column guidance, and only then commit to the full set.
- Estimate and confirm. The Review screen shows how many records the query matches and the estimated cost, and clicking Run opens a second confirmation.
- Pay per processed record, cancel any time. The run page shows live progress. Cancelling finishes the records in flight and stops; records that errored or never ran aren't charged.
Every run is also a frozen snapshot of the config, schema and model it used, so editing the Study later doesn't rewrite history. The hard cap is 20,000 records per run. For a year of 60,000 tickets, the docs recommend four quarterly Studies instead of one giant one.
What do example schemas look like?
The whole feature comes down to "what columns do you want?" Three examples:
Knowledge base, to turn tickets into reusable answers: question (short text) · answer (long text) · category (single choice) · product_area (single choice)
Sentiment and insight, to understand a year of support: sentiment (single choice: positive / neutral / negative) · contact_reason (single choice) · frustration_driver (short text) · was_resolved (yes/no)
Backlog cleanup, to tag old tickets consistently: topic (single choice) · priority (single choice) · is_billing_related (yes/no) · summary (short text)
Keep one question per column: "category and severity" should be two columns.
What happens after a run finishes?
The results page is a sortable table with one row per ticket and a link back to each Zendesk ticket, so you can check any answer against the original. From there you can group by a column, open the Report view (bar charts for choice and Yes/No columns, histograms for numbers), export to CSV, or push the rows into a knowledge source. Each row becomes a document that any linked agent can search.
Who is Studies for?
Studies is an admin feature. It's for the person who wants to turn a year of support data into agent knowledge, find what customers struggle with, or clean up a tagging mess, without exporting CSVs or writing scripts.
Which Macha plan includes Studies, and what does it cost?
Macha sells one plan with every feature, priced by ticket volume: $299 a month for 750 tickets, up to $3,999 for 10,000, with setup and monitoring by the Macha team included. Studies also run on the free trial, which gives $50 of free usage with no credit card. A saved Study costs nothing until you run it, and you always see the estimate before a full run.
Frequently asked questions
What is Macha Studies in one sentence? A feature that runs an AI model over a set of records (today, Zendesk tickets) and extracts a structured result for each, one row per record.
What can I extract? Anything you can describe as a column: yes/no flags, categories, numbers, or free text such as a summary, a resolution or a question.
Is it just for tickets? Today the input source is Zendesk Tickets, with more sources on the roadmap.
How is this different from chatting with an agent? Chatting handles one request at a time. Studies runs a model across your whole dataset at once and returns structured results.
Can the results feed my agents? Yes. Push a completed run into a knowledge source and any linked agent can search it.
How many tickets can one Study process? Up to 20,000 records per run. Split larger sets by date range, for example one run per quarter.
How fast can you get a first Study running?
Define the columns you want, scope the records, test on a small sample, review the estimate and run.
See what's in your tickets: start with $50 of free usage, no credit card required, and talk to the Macha team about running your first Study. Start free.
