A startup automated their support aggressively. Phase 1: auto-responses for FAQ questions (30% of volume). Phase 2: AI-driven troubleshooting flows (another 20%). Phase 3: chatbot handling billing and account changes (another 15%). Phase 4: AI for complaint resolution (another 15%).

At 80% automation, their dashboard looked incredible. Cost per ticket dropped 60%. The VP sent a company-wide email celebrating.

Six months later: churn was up 12%. NPS dropped 15 points. A Reddit thread titled "How to actually talk to a human at [company]" had 400 upvotes. A competitor was running ads that said "Talk to real people. Not bots."

The cost savings from automation were wiped out by the revenue loss from churn. The net was negative.

Where the Curve Bends

Automation has diminishing returns, and past a certain point, negative returns.

The first 30 to 40% of automation is pure upside. These are the tickets that have one right answer, every time. Password resets, order status, business hours, pricing. Customers prefer instant AI answers for these. Nobody wants to wait 2 hours for a human to tell them the store closes at 6pm.

The next 20 to 30% (total: 50 to 70%) is mixed. These are tickets that can be automated but sometimes shouldn't be. Billing questions where the standard answer is right 80% of the time. Troubleshooting flows that work for most users but confuse some. Refund requests where the policy is clear but the customer's situation has nuance.

Above 70%: danger zone. These are the tickets that exist because they're inherently complex, emotional, or ambiguous. Automating them means giving wrong answers to edge cases, giving cold answers to emotional situations, and trapping customers who need a human behind an AI wall.

The startup in the opening example crossed the line at Phase 4 (complaint resolution). You can't automate empathy. A customer who's upset about a broken product doesn't want a chatbot to process their complaint. They want a person to acknowledge the impact and make it right.

The Hidden Costs of Over-Automation

Customers who can't reach a human churn silently. They don't submit a ticket saying "I'm leaving because of your chatbot." They just stop logging in. The churn shows up in your retention metrics 30 to 60 days later, disconnected from the automation that caused it.

Customers who get wrong automated answers lose trust in all your support. One bad chatbot interaction makes customers skeptical of every subsequent AI response, even the correct ones. Trust is fragile and one-directional: hard to build, easy to destroy.

Customers who can't escalate become detractors. A customer trapped in a chatbot loop who eventually gives up doesn't just churn. They tell people. Social media complaints about "I can't reach a human" are among the most shared and viral content in consumer tech.

How to Find Your Limit

The right automation level depends on your product, your customer base, and the nature of your tickets. But there are signals that you've gone too far:

Repeat contacts increasing. If customers are contacting you about the same issue multiple times, the automation isn't resolving it. It's deflecting it temporarily.

CSAT on automated tickets dropping. If automated responses score 3.5 while human responses score 4.5, your automation is lowering the bar.

"Talk to a human" becoming a top request. If your most common message is "can I speak to a person," you've automated things that shouldn't be automated.

Churn rate increasing after automation deployments. Track churn in the 60 days following each automation expansion. If churn ticks up, the automation is pushing customers away.

The Right Approach

Automate the simple stuff (40 to 50% of volume). Auto-respond to queries with one clear answer. Order status, FAQ, pricing, hours, password resets. Customers prefer AI for these.

Assist on the medium stuff (20 to 30%). AI classifies, gathers context, and pre-populates the agent's response. The human reviews, adjusts, and sends. Faster than fully manual, better than fully automated.

Keep humans on the hard stuff (20 to 30%). Complaints, billing disputes, technical debugging, retention conversations, VIP accounts, anything with emotional weight. These are the interactions that build loyalty and prevent churn. Automating them destroys value.

The total: AI touches 70 to 80% of tickets (through full automation and agent assistance). But full automation only handles 40 to 50%. The rest involves a human at some point. That's the balance.

Monitoring the Balance

Track automation rate alongside customer experience metrics. Plot them on the same chart.

If automation rate is rising and CSAT/CES/NPS are stable or improving, you're in the sweet spot. Keep going.

If automation rate is rising and experience metrics are declining, you've pushed too far. Pull back on the most recently automated categories and re-examine.

The goal is optimal automation, not maximum automation. The point where cost is low, speed is high, and customers feel well-served. That point is different for every company, but it's almost never 80%.

Supp's approach is built around this philosophy. Classification at $0.20 per message handles the routing and understanding for 100% of messages. But automated resolution only applies to the intents where it works well. Everything else goes to a human with context already gathered. The AI does the thinking. The human does the caring. That split is where the value is.