Shopify
Stripe
Slack
Notion
Google Workspace
Confluence
Ask one question of
five thousand tickets.
A Study points an AI extractor at a set of records, fills in the columns you defined, and hands you a grid. One row per record, one column per thing you wanted to know.
| Ticket | Root cause short text | Refund requested boolean | Product area single select | Sentiment single select |
|---|---|---|---|---|
| #20831 | Carrier lost the parcel, no scan for 5 days | Yes | Shipping | Frustrated |
| #20847 | Wrong size shipped, warehouse pick error | Yes | Fulfilment | Neutral |
| #20852 | Discount code expired before checkout | No | Billing | Neutral |
| #20861 | Duplicate charge on a retried payment | Yes | Billing | Negative |
| #20874 | analysing… | |||
A spreadsheet you never had to fill in, about work you already did.
Instead of opening one ticket at a time and asking, you point a Study at a set of records, describe the columns you want, and every record is analysed in parallel. Doing it by hand is a week. Doing it through chat is one conversation per ticket.
One · Input
The records
A Zendesk search query and an optional date range decide what is in scope. Any expression you would type into Zendesk search works here.
Two · Schema
The columns
Each column has a type, a label and optional guidance telling the model how to fill it. This is the part that turns a vague question into a repeatable one.
Three · Output
The grid
One row per record, filled in and sortable. Browse it, filter it, export it, or push it back into Macha as a knowledge source.
Questions worth pointing a Study at.
Every one of these is a question a support lead already has and usually cannot answer, because answering it means reading a few thousand tickets.
Revenue
What are we actually refunding for?
Run it over every ticket that mentions a refund and get the real driver behind each one, not the macro that closed it.
Columns it fills
Deflection
Which tickets should never have reached a human?
Classify a quarter of volume by whether an existing help centre article already answered it. That list is your automation backlog.
Columns it fills
Product
What is support telling us about the product?
Tag every ticket with the feature area and whether it is a bug, a gap or a misunderstanding, then hand product a ranked list instead of anecdotes.
Columns it fills
Compliance
Did we say anything we should not have?
Audit outbound replies for promises about delivery dates, refunds or medical or financial claims, and get the exact ticket back with the quote.
Columns it fills
Quality
Where does the queue actually go wrong?
Score resolution quality and first-contact outcome across a month, then compare weeks, brands or teams on the same rubric.
Columns it fills
Voice of customer
What are people asking for that we do not sell?
Extract every explicit request for a product, size, region or integration you do not offer yet, with the wording the customer used.
Columns it fills
The schema is the whole trick.
A column is not just a label. It has a type, so the output is machine-usable, and it can carry guidance in your own words so the model fills it the way you would.
Column guidance, in plain English
“Mark this true only if the customer explicitly asked for money back. Someone complaining about a delay is not a refund request.”
True or false. The one to reach for when you are counting things.
One value from a list you define. Keeps categories tidy enough to group by.
Several values from your list, for tickets that are genuinely about two things.
A figure pulled out of the text — an amount, a count, a number of days.
A phrase. Good for a root cause or a name you want to read at a glance.
A paragraph, when the answer needs the reasoning and not just the label.
Then do something with it.
A finished Study is not a report you read once. It is a dataset, and it can go straight back into the product.
Read it in the grid
Sort and filter in place. The columns you defined are the columns you filter on, so the answer is two clicks away.
Export to CSV
Take it to a spreadsheet, a BI tool or a board deck. It is ordinary tabular data with no lock-in.
Feed it to your agents
Push the result back into Macha as a knowledge source, so the analysis you just ran informs the answers your agents give tomorrow.
Four features that sound alike. They are not.
A Study extracts structured facts from records. No agent runs, no tools are called, nothing is graded — you get a grid.
Questions about Studies.
Short answers. If something is missing, ask us — we will tell you straight.
Pronto para turbinar seu time com IA?
Comece em minutos. Conecte suas ferramentas, configure seus agentes e deixe a IA cuidar do resto.
500 créditos grátis · sem limite de tempo, sem cartão de crédito