Ask any support manager what percentage of their tickets could be automated, and they'll guess 20 to 30%. Then show them the data, and it's almost always higher.

The gap between perception and reality exists because support teams are too close to the work. When you handle tickets all day, every ticket feels unique. You remember the weird edge cases, the angry customers, the complex multi-touch issues. You forget that you just answered "what are your business hours?" for the 200th time this month.

An audit makes the invisible visible.

The 2-Hour Audit

You need two things: your ticket data from the last 30 days and a spreadsheet.

Step 1: Export your tickets. Every help desk can do this. You want ticket subject, first customer message, status, and resolution time. If you can get tags or categories, even better. Export as CSV.

Step 2: Categorize by intent. Go through the tickets and tag each one with a simple label. "Password reset." "Billing question." "Feature request." "Bug report." "How-to." "Shipping status." "Refund request." "Cancel request."

If you have hundreds of tickets, don't tag them all. A random sample of 200 is enough to see the pattern. Or use AI classification to categorize them automatically, which turns a 2-hour task into a 5-minute task.

Step 3: Count the categories. Sort by frequency. You'll see something like this:

• Password reset / login issues: 18%
• Shipping status: 15%
• How-to questions: 14%
• Billing questions: 12%
• Refund requests: 8%
• Feature requests: 7%
• Bug reports: 6%
• Cancellation: 5%
• Everything else: 15%

The exact percentages vary by business, but the shape is always the same: a small number of categories represent the majority of volume.

Scoring Each Category

For each category, ask two questions:

"Is the answer the same (or nearly the same) every time?" If yes, it's a strong automation candidate. Password resets always follow the same steps. Shipping status is always a tracking number lookup. Business hours are always the same answer.

"Does it require human judgment?" If no, automate it. If yes, keep it human.

Build a simple scoring grid:

In this example, about 47% of tickets (password, shipping, how-to) can be fully automated. Another 20% (billing, refunds) can be partially automated (AI handles the simple cases, routes complex ones to humans). And 13% (features, bugs) can be auto-routed to the right team even if they can't be resolved by AI.

Total automation potential: 50 to 60% of tickets touched by AI in some way. That's typical.

Calculating the ROI

Once you know which categories to automate and their volumes, the math is straightforward.

Take your monthly ticket volume. Multiply by the percentage that's automatable. That's how many tickets AI handles.

Cost per AI resolution: $0.20 to $0.30 with Supp. Cost per human resolution: $5 to $15 depending on agent salary and tickets-per-day.

Example: 500 tickets/month. 50% automatable (250 tickets). Human cost for those 250: $2,500 (at $10/ticket). AI cost: $75 (at $0.30/ticket). Monthly savings: $2,425. Annual savings: $29,100.

For the human-handled 250 tickets, AI classification still helps by pre-categorizing and routing them, saving 1 to 2 minutes per ticket. That's 250 to 500 minutes saved per month. About 6 to 8 hours of agent time.

The Quick Wins

After the audit, start with the highest-volume, lowest-complexity categories. These give you the biggest return with the least effort.

Password resets and login issues are almost always the top win. These follow the exact same troubleshooting steps every time: check email, try reset link, clear cache, try different browser, escalate to IT if none of those work. AI can walk customers through this automatically.

Shipping status is another quick win if you have a tracking system. AI receives "where is my order?", looks up the tracking number by email or order ID, and responds with the current status.

Business hours and location questions. These are the simplest possible automation. AI responds with your hours and address. Done.

After the quick wins, move to the partially-automatable categories. Billing questions where AI handles the simple lookups and routes complex disputes to humans. Refund requests where AI processes requests under $50 automatically and escalates larger ones.

What Not to Automate

Some things should stay human, even if technically they could be automated.

Complaints from angry customers. AI is bad at reading emotional cues and even worse at genuine empathy. An angry customer who gets a bot response feels dismissed. Route these to your best agents.

VIP or high-value accounts. If a customer represents significant revenue, they deserve a human. AI can still classify and prioritize the ticket so it gets to the right person faster, but the response should come from a person.

Anything involving sensitive data. Account security issues, data deletion requests, and legal matters should always involve a human who can verify identity and make judgment calls.

The audit tells you what can be automated. Good judgment tells you what should be.

Repeat Quarterly

Your ticket distribution changes over time. New features generate new questions. Bug fixes eliminate old ticket categories. Pricing changes shift billing inquiries. A product launch spikes a previously small category.

Re-run the audit every quarter. It takes an hour once you have the process down. Each audit usually reveals 2 to 3 new automation opportunities and occasionally shows that an existing automation is no longer relevant.

Support optimization isn't a one-time project. It's an ongoing discipline. The teams that do it consistently spend less, respond faster, and have happier customers and happier agents.