This week at ServiceNow's Knowledge 2026, Atera became the first AI IT vendor to put a specific outcome on the line: Robin, their autonomous agent, will resolve 50% of your Tier 1 and Tier 2 tickets in 90 days, or you don't pay.
It's a real shift from two years of vendor pitch decks. It's also a contract, and contracts deserve the reading nobody wants to do at 4pm on a Friday.
We run an AI-first managed IT practice, so this is the kind of announcement we have to think clearly about. Some of it is genuine progress. Most of the marketing around it is the standard hype loop we've watched for two years, and the people most likely to get hurt are the ones who treat the press release as a substitute for the schedule.
The Three Numbers Atera Controls
Stripped down: Robin runs in your environment for 90 days. If it doesn't autonomously resolve 50% of the Tier 1 and Tier 2 tickets it touches, Atera waives its fees. There's a 72-hour proof of concept up front so you can see Robin handle live tickets before signing.
Read that twice. The 50% is the headline. The three definitions underneath it are the whole game.
"Resolved" is whatever the platform says it is. In most ticketing systems, resolved means the ticket was closed, not that the user's actual problem went away. Auto-close timers, low-confidence acknowledgments, and the same issue reopening as a new ticket all push the resolution number up without changing what the user experienced. The number you want to see in the SOW is reopen rate inside 30 days, and how user-confirmed fixes are counted differently from system-confirmed closes.
Atera decides what counts as Tier 1 or Tier 2. Tickets Robin can't handle cleanly can be reclassified out of scope. The denominator is theirs. A 50% guarantee on a controlled denominator is a different product than a 50% guarantee on every ticket your users file.
The unresolved half is undefined. The contract speaks to what Robin does. It does not speak to what Robin's failed attempts cost you. A tool that misroutes 10% of tickets before escalating creates triage debt that lands on whoever is left to clean it up. That is rarely Atera.
This isn't a hit piece. Outcome-based pricing in IT was always going to start here, and Atera deserves credit for going first. The point is that "50% or your money back" is a useful first move, not a finished product. You read it the same way you read any vendor contract: the marketing is the wrapper, the schedule and the definitions are the deal.
What the Guarantee Doesn't Cover
The tickets Robin handles best are the ones a well-built self-service portal already absorbs: password resets, software access requests, basic VPN reconnections. There's real time savings there. Has been since Microsoft shipped Authenticator's password reset flow. The marginal lift from autonomous IT on top of decent self-service is smaller than the marketing implies.
The tickets that hurt your business are the ones that look easy and aren't.
A routine account lockout. Robin sees it, runs the unlock, closes the ticket. A technician who has watched your network for two years sees the second lockout this week from the same IP range and pulls the audit logs. One of those is a feature. The other is how a credential-stuffing attempt gets caught at hour two instead of week three.
An intermittent printer issue. Robin pushes a driver update and closes it out. Your network engineer notices the printer is on the same segment as your voice traffic and that VoIP quality has been off all week. Robin solved a printer ticket. The engineer caught a bandwidth saturation issue before sales started complaining about dropped calls.
A SharePoint access request. Robin checks the role-based controls and provisions the access. In a regulated environment (healthcare, legal, financial services, anything CMMC-touching), that's the moment your audit trail starts. Auditors don't talk to Robin. They want a name on the approval, and they want to know that the approver verified the request actually came from the manager, and that the data classification fits the user's clearance.
Three tickets, three resolution checkmarks, three things Robin missed. The pattern isn't that AI is bad at IT. The pattern is that AI is good at executing inside the boundary of a known problem and structurally blind to when a known problem is hiding something else.
That blind spot doesn't show up in a 90-day resolution rate. It shows up in a postmortem six months later, where a cyber incident traces back to an event the tool quietly closed without flagging.
What an MSP Does That No Guarantee Covers
We use AI tooling at SMS the way a surgeon uses an electrocautery. It's a tool. The surgeon stays accountable for the patient. The hospital does not waive its malpractice premium because the cautery worked.
In practice, that means the work that doesn't fit on a guarantee:
We hold vendors to their own SLAs when an issue spans Microsoft, your line-of-business app, and your internet provider. Robin doesn't sit on hold with Microsoft Premier at 11pm. We do.
We catch the cybersecurity signal hiding inside the routine ticket, because we've watched your environment long enough to know what normal looks like. That instinct doesn't transfer to a tool that started observing your stack last quarter.
We own the documentation trail when an auditor shows up. Auditors want a name next to every access change, every config push, every policy exception. The audit doesn't care that the agent ran the command. It cares who was responsible for the agent running the command.
We make judgment calls when the standard playbook would do the wrong thing. This is the highest-leverage thing an MSP does and the hardest thing to automate, because by definition the data is ambiguous. A tool optimized to act on ambiguous data confidently is a tool optimized to be wrong with confidence.
We know your business, not just your tickets. The technician who configured your backups two years ago notices when last night's job ran 40% smaller than it should have, and pulls on that thread before anyone files a complaint.
Atera is right that AI in IT is moving from features to outcomes. The version of that move where it works is leverage on the volume, plus a competent person on the judgment. The version that fails is the one where someone treats a 50% autonomous resolution rate as a substitute for the half nobody understood.
If you're evaluating any AI IT product (Atera, Kaseya, ServiceNow, or anyone else shipping in this category), the contract question matters, but it's secondary. The first question is what shape your IT environment is in when the 90 days are up, and who you're calling for the half Robin didn't resolve.
FAQ
What is Atera actually guaranteeing?
That Robin, its autonomous IT agent, resolves 50% of Tier 1 and Tier 2 tickets in the first 90 days of onboarding, or Atera waives its fees. Atera defines what "resolved" means and which tickets fall into the guaranteed tiers, so the contract language matters more than the headline number.
Does this mean autonomous IT can replace a managed IT partner?
No. Autonomous IT handles a specific class of tickets well. It doesn't handle multi-vendor escalation, compliance documentation, cybersecurity pattern recognition, on-premises hardware work, or judgment calls on ambiguous data. The capable AI IT tools assume there's a competent human accountable for the environment behind them. Atera doesn't actually disagree with that. The marketing just buries it.
What's the difference between automation and autonomous IT?
Automation runs predefined scripts when predefined conditions are met. Autonomous IT uses an AI agent to interpret a ticket, choose an action, execute it, and verify the result. Autonomous tools handle variation better than rule-based automation. They're also harder to audit, because the reasoning is opaque. In regulated environments that audit gap is a meaningful constraint.
How should I evaluate an AI IT guarantee before signing?
Three contract questions. How is "resolved" defined, and what's the reopen-rate window. Who controls ticket classification into the guaranteed tiers. What happens, financially and operationally, to the tickets that fall outside the guarantee. Get those in writing. The harder, fourth question, who is accountable when an autonomously closed ticket turns out to have been a security event the tool didn't flag, is the one nobody wants to sign for, including you.
Where does AI fit in IT support if guarantees are this narrow?
It fits as leverage on volume, not as a substitute for judgment on consequence. Routine password resets, software access requests, common connectivity issues are reasonable to automate. Anything where a wrong answer creates a security exposure, a compliance issue, or a damaged client relationship is a place where a person with skin in the game still has to sign off.