The Ticket Is Disappearing
The thing you count when you invoice a client is being retired, and the companies retiring it are the ones who sell you your tools. The ticket.
Start with Atera, which makes the software a lot of providers run their service desk on — thirteen thousand customers, about half of them MSPs. Its autonomous agent is called Robin, and the company does not sell it on a demo. It sells it on a guarantee. Within ninety days of onboarding, half your Tier One and complex Tier Two tickets get handled by Robin, with no person touching them. If that doesn’t happen, the fees are waived. That is not a projection. That is a contract term.
It’s worth understanding how a company gets confident enough to write that down. CEO Gil Pekelman walked through it on this show in June. Before you sign anything, you sign an NDA. Then you hand Atera six months of your own ticket history. Atera runs those tickets through Robin, one by one, and Robin marks each one as something it could have closed on its own, or not. Then they come back and show you the list. Here are your tickets. Here are the ones you are never going to see again.
The company is explicit about where that leads. In reporting from ChannelE2E, Atera calls it the death of the ticket, and draws the pricing conclusion out loud — providers should move to outcome-based billing, and start running on new numbers. Avoided tickets. Autonomous resolution rate.
Acronis shipped a piece of the same thing in July. Its AI Service Desk turns alerts into tickets and drafts the resolution. In the company’s own announcement, it’s free — but only to partners on a commitment contract. Pay-as-you-go partners are excluded.
NinjaOne, which says it serves forty thousand customers across a hundred and forty countries, wired its endpoint data straight into ServiceNow. An alert opens an incident. Asset and patch state update on their own. Nobody types anything.
And Channel Insider quotes Ingram Micro’s Victor Baez on what partners are saying out loud now. A lot of them, in his words, are talking about doubling their business without hiring anybody.
Every one of those is a ticket that never gets created, or never gets worked. Which leaves an open question about what you are going to put in front of a client at renewal.
To see why that question is harder than it sounds, look at what a ticket actually is.
You Can’t Invoice an Absence
A ticket is a record that work happened. An avoided ticket is a record of nothing at all — and nothing is very difficult to invoice.
So everything replacing the ticket rests on a number that can only be produced one way: by counting what used to happen, before you changed it. And that count has to be taken in advance, because once the automation is running, the thing you needed to measure is gone.
Which is why Atera asks for six months of tickets, and why it asks before you sign. Only one party in that transaction has a commercial reason to build the baseline, and it is the party selling the automation. The measurement does get made. It just gets made by the counterparty, on the counterparty’s model, and handed back to you as a finding. And consider what accumulates on the other side of that. Atera has thirteen thousand customers. Every evaluation it runs hands over another shop’s complete operational history — so the party best positioned to know what normal looks like across this entire market is not an analyst firm, and it is not you.
And you can’t reconstruct it afterward, because the ground moves. Pekelman said something in June that got almost no attention: ticket volume goes up when Robin is switched on. Not down. Up. People who never bothered to ask for help start asking, because the thing answering is fast and doesn’t make them feel stupid. The denominator grows while the labor underneath it shrinks.
One company can state its number, and the reason is instructive. AT&T says its incident system prevented about three point one million field dispatches and twelve million hours of customer downtime in a single year. Twelve million hours is roughly fourteen hundred years of somebody’s service being down. It can produce that because it started building in 2018 and has been counting dispatches ever since. Eight years of before.
Without it, what you have is assertion. Pekelman says Atera has customers who took headcount down ninety percent. Visa is cutting twenty-six hundred people and naming AI. Microsoft took its service organization from roughly fifty thousand to forty, claiming seven hundred and fifty million a year. Ten thousand people, seven hundred and fifty million dollars. That works out to about seventy-five thousand a head — which is to say the savings and the payroll are the same number.
Now the measured picture. Stanford’s economic policy institute finds that since 2022, unemployment among the workers most exposed to AI rose seventy-seven hundredths of a point. The least exposed rose eighty-five. The exposed group did slightly better. And the productivity gains that show up reliably in controlled experiments have not shown up in the aggregate numbers yet. Read that the right way around. The workers everyone said were about to be automated out of a job are, so far, doing slightly better than the ones nobody worried about.
So the number you carry into a renewal is one somebody else produced, about work nobody can see, in a category the public record already says does not add up.
Which would be survivable, if the person you were telling it to hadn’t already made up their mind.
The Client Already Stopped Believing
Bring that into the room where you have to say it out loud, because the person on the other side of the table changed while you were building the offer.
Gallup, with Bentley University, has run the same survey on American attitudes toward AI for four years now. This year, thirty-nine percent said AI does more harm than good. Nine percent said more good than harm. That is not a split. That is more than four to one. Among adults under thirty — the people staffing your clients’ front lines — the share calling it more harmful went from twenty-seven percent to forty-seven. That took two years. And roughly three out of four Americans now say they have little or no confidence in companies to use AI responsibly.
Read the object of that last sentence carefully. Not the technology. Companies. The ones deploying it.
Then the people who actually have to use the stuff. In a survey from Adaptavist — a technology consultancy that sells services into this exact problem, so weigh it accordingly — forty-two percent of workers said they spend more time checking AI output than the AI saves them. Fifty-five percent said it is making their team less efficient, not more. Sixty-five percent said they feel nostalgic for how work went before.
Those are your client’s employees. Those are the people whose experience gets reported upward when somebody asks whether the new arrangement is working.
So here is the meeting you are actually scheduling. You walk in with a number you did not produce, describing work that left no trace, and hand it to a buyer whose own staff have been telling them all year that AI costs more than it returns. That is not a hard conversation. That is a conversation you lose on arrival, no matter how well the automation actually performed.
There is a version of this that goes well, and it doesn’t require you to be AT&T. It’s the provider who walks into that room with two dated numbers — what the year looked like before, and what it looks like now — and lets the client draw the conclusion themselves. That provider isn’t arguing about whether the AI worked. They’re the only one in the room who can show it.
Which makes this a calendar decision before it is a strategic one. You can build your own baseline while you still have tickets to count — your volume, your categories, your mix, on your books, dated and kept — so the number you carry into that room is one you produced and can produce again next year. Or you can hand six months of your history to the company quoting you the savings, and spend every renewal from here forward negotiating against a measurement you do not own, cannot rebuild, and never took.
Which turns a large strategic problem into a much smaller and more urgent one.
Why Do We Care?
Because this turns out to be a records question, and records questions get settled by whoever configured retention four years ago and never looked at it again. Go find out how far back your PSA actually keeps closed-ticket detail, and whether that history survives a platform migration — because the vendors pitching you autonomy are frequently the same ones pitching you a migration. Export the last twenty-four months into something you control, this month, while the export still contains something worth having — because that baseline is going to get built either way, and the only question actually in play is whether you end up holding a copy.
What to Consider
- Check what your retention policy keeps, not just how far back it goes. Most PSAs hold the ticket record considerably longer than they hold the parts that make it scoreable — time entries, technician notes, resolution detail, reopen history. A count of four thousand closed tickets proves nothing at a renewal; the distribution underneath it is the entire asset, and that’s usually the layer on a shorter purge schedule.
- Freeze your category taxonomy before you automate anything, and write down the version you froze. The fastest way to destroy a baseline is not deleting it — it’s re-mapping your categories eighteen months from now so that the before and the after aren’t comparable, which happens by default during any platform change. If your ticket types shift underneath you, you will still have all the data and no longer have a measurement.
- Count the demand that never became a ticket, because that’s the number about to move. Atera says half of employees bypass ticketing entirely, and Pekelman says volume climbs once the autonomous agent is switched on — which means your ticket count was already understating real demand and is about to understate it differently. Pick one month, deliberately count what arrives by hallway, direct email, and chat, and keep that figure next to your ticket total so you know what you were absorbing off the books.
If this trend continues: within twelve to eighteen months, “show me your ticket volume from before you automated” becomes a standard line item in renewal negotiations and acquisition diligence, and the providers who can’t produce it will discover that the case for their last two years of pricing lives in a file their vendor owns.