AI Now Closes 7 in 10 Support Tickets. The 3 Left Behind Will Break Your Team.

Salesforce just published a number that should change how every support leader plans next year. On its own help portal, AI agents handled more than 5 million conversations, against 2.4 million handled by people. Across the industry, roughly 7 in 10 customer service sessions now run autonomously, according to ZDNET’s reporting on agent adoption. Ticket deflection is no longer a promise on a vendor slide. It works, it scales, and adoption tripled in a single year.

But the celebration hides a structural shift. When AI absorbs the easy 70 percent, your human team stops working a mixed queue. They start working a queue that is 100 percent hard cases. Every hour of every shift, nothing arrives except the tickets AI could not close. Most support organizations have never operated that queue. Almost none have planned for it.

The Deflection Math Nobody Runs

Deflection dashboards celebrate the numerator. Five million conversations closed by AI sounds like pure savings, and in one sense it is. Fewer tickets reach people, so cost per contact drops and executives applaud.

Now run the other side of the equation. The 2.4 million conversations that still reached humans were not a random sample. They were the residue. The ambiguous cases, the angry customers, the problems that live in the physical world.

Your average handle time will climb, because the average just lost its easy half. Your CSAT will wobble, because every remaining conversation starts from frustration. Your agents will burn out faster, because the breather tickets, the quick password resets between hard calls, are gone.

None of that means deflection failed. It means the job changed. As a result, the team, the tools, and the metrics have to change with it.

Ticket Deflection Leaves the Physical Problems Behind

Look at what actually survives ticket deflection. AI agents excel at problems that are text-shaped. Billing questions, account access, policy lookups, order status. If the answer lives in a database or a knowledge base, the AI closes the case before a human ever sees it.

What remains is disproportionately physical. The router with a blinking light the customer cannot name. The machine on a factory floor making a noise no knowledge article describes. The installation where step four looks nothing like the diagram. We wrote before about the customer who cannot describe what they are looking at, and why that case was always the most expensive one in the queue. After ticket deflection, that case is no longer an outlier. It is the queue.

So the residual queue has a specific character. It is visual, ambiguous, and emotionally loaded. Text got filtered out. Reality got left in.

Your Metrics Were Tuned for a Queue That No Longer Exists

Every benchmark in your operations review assumes a mixed queue. Handle time targets, tickets per agent per day, escalation rates, staffing models. All of them date from an era when easy tickets subsidized hard ones.

For example, an agent who used to close 40 tickets a day might now close 12, and that agent is probably working harder than before. If you keep grading against the old baseline, your best people will look like your worst performers. Meanwhile, finance sees the deflection savings and asks why cost per human ticket went up. It went up because the tickets did.

The honest move is to reset the baseline. Measure the residual queue as its own operation, because that is what it has become. New handle time targets, new quality rubrics, new staffing ratios. Otherwise you will punish the exact people carrying your hardest work.

The Residual Queue Needs Eyes, Not More Text

Here is where most leaders reach for the wrong fix. The instinct is to buy more AI, tune the prompts, expand the knowledge base. But the surviving tickets did not survive because your AI was weak. They survived because text cannot resolve them at all.

A customer staring at a leaking valve does not need a smarter chatbot. They need someone who can see the valve. Because the bottleneck is visual, the highest leverage tool for the residual queue is live video. One session lets an agent see the blinking light, the error screen, the miswired cable, and resolve in minutes what a text thread would bat around for days. It also settles, with evidence, whether a truck roll is actually necessary.

This is exactly where Viewabo sits. An agent sends a link, the customer taps it on a phone, and the agent is looking at the problem. No app install, no account creation, no asking a frustrated customer to hold the manual up to a computer camera. The hard queue stops being a guessing game and becomes a diagnosis.

What to Do Before the Queue Hardens

If your deflection rate is climbing, act now, because the residual queue gets harder as that number rises. Four moves matter most.

First, re-baseline every metric. Split reporting into “AI-resolved” and “human-resolved” and never compare today’s human numbers to last year’s blended ones.

Second, re-profile your hiring. The residual queue rewards judgment, patience, and diagnostic skill over speed and script adherence. Pay accordingly, because these people now handle only your most expensive moments.

Third, give humans tools AI does not have. Seeing the customer’s physical world is the clearest example. Ticket deflection removed the work that needed typing. What is left needs eyes.

Fourth, route physical-world problems to video early. Every text exchange about an unnameable blinking light adds cost and frustration before the inevitable escalation. Skip the theater and look at the thing.

Deflection Was the Easy Part

The industry spent two years asking whether AI could handle customer conversations. That question is settled. Seven in ten sessions already run without a person, and the share will grow.

The next question is harder and much more interesting. What happens to the people, tools, and metrics serving the 30 percent that remains? Ticket deflection cleaned the easy work out of your queue. What it left behind is the truest test of your support organization, because every ticket that reaches a human now genuinely needs one. Build for that queue on purpose, or watch it grind down the team you have.