You deployed AI, it handles thousands of contacts, and your total ticket volume is flat or even up. This is normal, and it has two causes. The AI removes the easy contacts, which leaves harder work for your people. And making it cheaper to ask brings more people to ask.
The business case for AI in customer service almost always promises the same thing: fewer tickets, so fewer agents, so lower cost. A few months in, the AI is clearly working, the dashboards say so, and yet the total volume has not dropped the way the case predicted. The cost per ticket has barely moved.
This is the AI volume paradox, and it surprises teams only because the business case ignored two ordinary effects. This article explains both, and what to measure so the next forecast is honest.
The promise: fewer tickets, lower cost
The standard plan is simple arithmetic. If the AI handles 40 percent of contacts, you need 40 percent fewer agents, and your cost falls by roughly 40 percent. The plan treats every ticket as the same size, so removing 40 percent of them removes 40 percent of the work.
Tickets are not the same size. That single wrong assumption is where the paradox begins.
The AI takes the easy tickets and leaves the hard ones
An AI agent is good at the simple, common contacts: order status, password resets, opening hours, returns. Those are also the contacts your human agents handled fastest. So the AI does not remove an average slice of the work. It removes the easiest slice.
What is left for your people is the harder half: the messy, multi-part, emotional, or unusual cases. The average human ticket now takes longer, because the quick ones are gone. Your agents handle fewer contacts, but each one costs more time. Cut headcount by the deflection rate and you will watch handle times climb and a queue form, because the people who remain are doing harder work than the forecast assumed.
Making it cheaper to ask brings more asking
The second effect is about demand. Before AI, contacting support had a cost to the customer: a hold queue, a wait for an email reply, a callback. That friction quietly suppressed some contacts. People with a small question often decided it was not worth the wait.
Mark Levy, who writes Decoding Customer Experience, describes what happens when you remove that friction. His piece on why volume goes up when your AI is working makes the point plainly: an instant, always-available channel is cheap to use, so customers use it more. Questions that were never worth a phone call are worth a quick chat. The AI did not fail to reduce volume. It changed the price of asking, and lower prices bring more demand.
Jens Stark, in the Scaling Customer Value newsletter, frames the same tension as a question every team should answer before deploying: is your AI customer service a conversion booster or a churn catalyst? More contact is not automatically good or bad. It depends on whether those extra conversations help the customer or just absorb your capacity.
What to measure, and what to tell finance
The fix is to forecast the two effects instead of ignoring them. Before deployment, look at your contacts and estimate how the complexity mix changes once the easy ones are automated. Model the human handle time going up, not staying flat. And expect total contact volume to rise somewhat as the channel gets cheaper to use.
The paradox is a planning failure, not an AI failure
None of this means the AI is underperforming. It is doing exactly what it was built to do. The paradox lives entirely in a business case that assumed tickets were interchangeable and that demand was fixed.
A team that plans for both effects gets the real result: the AI absorbs the simple work, the people are freed for the hard work, and the cost case rests on handle time and demand rather than on a single deflection figure.