How to Analyze a Year of Zendesk Tickets with AI
Your support tickets are the most honest dataset your company has. Every one is a customer telling you, in their own words, what's confusing, broken, or missing. The problem is volume: nobody is going to read 20,000 tickets, so all that signal sits unused while you make decisions off gut feel and a few loud complaints.
Macha's Studies feature reads them all for you. Point it at a year of Zendesk tickets, tell it what to look for, and it returns a structured row per ticket — sentiment, the real reason they contacted you, whether they were resolved — that you can actually analyze. This is the analysis playbook for Studies. (If you want to turn tickets into agent knowledge instead, see past tickets as a knowledge base.)
What a year of tickets can tell you
Run AI across the whole set and you can finally answer questions that are normally guesswork:
- **Why do people actually contact us?** The true contact-reason breakdown — which is almost always different from your existing tags, because tags are applied inconsistently (or not at all).
- What's making customers unhappy? Sentiment plus the specific friction driving it.
- What's trending? Spikes in a topic after a release, a pricing change, or an outage.
- Where are the revenue signals? Refund requests, cancellation risk, upsell openings, hidden across thousands of tickets.
- What's missing from our docs? The questions your help center couldn't answer — your content roadmap, handed to you.
Step by step: analyzing your tickets
The flow is the same four-step Study wizard — the difference is what you extract.
1. Source. Start a New Study and choose Zendesk Tickets.
2. Scope the records. Set a date range — for a real analysis, widen it to 6–12 months (a single run is capped at 20,000 tickets). Add a filter only if you want to focus (e.g., one product line); for a broad analysis, leave it open to get the full picture. Then pick what the AI reads per ticket — the Full comment thread gives the richest signal for sentiment and reasons.
3. Extract — define your analysis columns. This is where an analysis Study differs from a knowledge-base one. Instead of question/answer, you define insight columns:
| Column | Type | What it surfaces |
|---|---|---|
sentiment | Single choice | positive / neutral / negative |
contact_reason | Single choice | the real reason they wrote in |
frustration_driver | Short text | what specifically annoyed them |
was_resolved | Yes/No | did the ticket actually get solved |
refund_requested | Yes/No | revenue signal |
doc_gap | Yes/No | could this have been self-served? |
Add a line of guidance (e.g., "Judge sentiment from the customer's messages, not the agent's. Pick the single best contact reason.") and a low-cost model — extraction doesn't need your most expensive model.
4. Review and run. Run a test on a small sample first to make sure the AI is judging sentiment and reasons the way you would — tweak the guidance if not. Then check the cost estimate and run the full study. It processes in the background with live progress, and credits are deducted per ticket as it goes.
What you get — and what to do with it
The result is a frozen, immutable table: one row per ticket, your columns filled. From there:
- Spot the patterns. Sort and count: which contact reason dominates? What share is negative sentiment, and on which topics?
- Quantify the gut feelings. "We get a lot of billing complaints" becomes "billing is 22% of negative-sentiment tickets, mostly about the renewal date."
- Build your roadmap. The
doc_gapandfrustration_drivercolumns tell product and content teams exactly what to fix and write. - Feed it forward. Push the results into a Knowledge Source so agents benefit, or re-run quarterly to track whether your fixes moved the numbers.
Because each run is a snapshot, you can compare this quarter's analysis to last quarter's and actually see the trend.
Example analyses to try
- Contact-reason audit — just
contact_reason+was_resolved, across the year. The single most clarifying report most teams never run. - Release post-mortem — scope to the two weeks after a launch; extract
topic+sentimentto see what broke. - Churn signals —
cancellation_risk(yes/no) +reason, to find at-risk accounts hiding in support. - Self-service opportunity —
doc_gap+topic, to rank what to add to your help center.
Cost and best practices
- Studies is on the Professional plan and up ($699/mo); runs are billed in credits per ticket, and you approve an estimate first.
- Test before the full run — get the schema and guidance judging the way you would on a sample.
- Use a cheap model — analysis extraction rarely needs a premium one; across 20,000 tickets, model choice is your main cost lever.
- Keep schemas focused — 4–6 well-chosen columns beat 15 fuzzy ones.
- Re-run on a cadence so you can watch trends, not just snapshots.
Frequently asked questions
Can AI really analyze my support tickets accurately? Yes, for well-defined extractions like sentiment, category, and yes/no flags — especially when you test the schema on a sample first and refine the guidance.
How many tickets can one analysis cover? Up to 20,000 per run; widen the date range to cover your year.
What's the difference between this and the knowledge-base use case? Same feature, different columns. Knowledge base extracts question/answer; analysis extracts insights like sentiment and contact reason.
How much does it cost? Credits per ticket, depending on the model; you see an estimate before the full run. Professional plan and up.
Can I export or reuse the results? The run is a stored, immutable snapshot you can review, and you can push it into a Knowledge Source for your agents.
The bottom line
You don't have to guess what your customers struggle with — the answer is already in your tickets. Studies reads a year of them in one run and hands you a structured table of sentiment, reasons, and signals. Scope the year, define a handful of insight columns, test, and run — and turn a pile of tickets into a report your product, content, and support teams can act on.
Find out what's really in your tickets: run your first analysis Study. 7-day free trial, no credit card required. Start free.
