What's Left of the Service Desk

The service desk as most companies still staff it — a tiered L1 taking calls and chats, an L2 handling escalations, an L3 for the hard problems, a device pipeline running through the same queue on the side — was designed for a ticket mix that doesn't exist anymore.

It was designed for a world where most tickets were questions: how do I do this, I can't get into that, my password isn't working, the application is behaving strangely. Those tickets rewarded a large L1 layer with call-handling skills and a knowledge base. Around them, physical tickets — new hires, breakfixes, offboardings, moves — ran through the same queue as a kind of by-product, absorbed into the same headcount, measured with the same metrics.

Two things are happening to that ticket mix at the same time, and neither of them is a Surya product. They're what's happening across the industry, whether any given company has named them yet or not.

Force one: AI is eating the software half

The tickets that lived at the top of the L1 funnel — how-do-I, access, password, application questions, policy lookups, request-for-approval — are being absorbed by AI assistants embedded in the tools employees already use. Chat interfaces on top of the knowledge base. Ambient agents inside the identity provider. Self-service inside the collaboration platform.

This isn't a prediction; it's happening now, inside companies that never bought an "AI service desk" product from anyone. Some of it is native to the tools those companies were already licensed for. Some of it is built by internal teams on top of language models. Some of it is offered by their existing vendors as an included capability. The delivery vehicle varies; the direction is uniform. A large share of what a first-line human used to answer, an assistant now answers — often better, always faster, always cheaper on the margin.

This is an industry reality, not a service line. Companies are not buying "AI containment" from Surya. They are doing it themselves, or their existing platforms are doing it for them, and the effect on the ticket mix is happening regardless.

Force two: the physical half moves to logistics

The other half of the queue — the half software cannot touch — is the pile we've named in the companion piece: joiners, leavers, breakfixes, IMACs, refreshes. Those tickets are resolved by a device moving to the right place in the right state, not by a conversation. They belong in a fulfillment operation with receiving, imaging, kitting, dispatch, retrieval, and a bench. When they're run through a support queue instead, they cost more, take longer, and consume the wrong people.

Once an organization has admitted the physical half is a logistics workload, it moves. Not necessarily to us — the point of the piece isn't that. It moves out of the service desk one way or another, because leaving it there is expensive in a way that becomes visible the moment anyone looks.

What's left in the middle

Subtract the software half and subtract the physical half. What remains is the actual, irreducible service-desk workload: the tickets that are questions of judgment, that require human context, and that don't have a device to be shipped.

• Complex application troubleshooting that assistants can't yet reason through end-to-end.
• Identity and access edge cases that need a human to interpret intent and policy.
• Escalations from the AI layer — cases the assistant correctly declined to close.
• Incidents that cross systems and need someone to hold the picture together.
• The judgment calls: exceptions, priority reads, situations where the right answer isn't in any playbook.

That work is real, and skilled, and often the highest-value thing the desk does. But it's a narrower band than what the traditional L1/L2 model was staffed for. It doesn't require a large front-line layer optimized for call handling; it requires a smaller number of experienced engineers whose entire job is judgment work — closer to what used to be L2 or L3.

Why this is a CIO problem, not an operational tweak

The instinct, watching this play out, is to treat it as a workflow question: automate some tickets, outsource some others, keep the org chart mostly intact, absorb the change through attrition. That approach can survive a year or two. It doesn't survive the trend line.

The reason it doesn't survive is that the traditional staffing model has two structural assumptions built into it, and both are being removed at once.

• Assumption one: there is a large volume of first-line questions that needs a large first-line team. The AI half of the shift removes this.
• Assumption two: device work belongs in the same queue as support work, because it's all "IT tickets." The logistics half of the shift removes this.

Take away either assumption and the tiered model still mostly works. Take away both and the tiered model is the wrong shape for what's left. It has too much front-line capacity for the ticket volume, and it's still trying to run a logistics operation as a side effect of a support function.

The restructure isn't about cutting headcount. It's about matching the org design to the actual work. Fewer, more senior engineers doing judgment work. A logistics operation — internal or vendored — running the physical pipeline as a real fulfillment function with real metrics. And the AI layer, wherever it lives, doing the volume work that used to justify the front line.

The old staffing model is going away not because it was bad but because the workload underneath it changed. The companies that see this early get to redesign deliberately. The ones that don't will discover, one budget cycle at a time, that they're paying for a shape that no longer matches the work.

Our role in this — the whole role — is the physical layer. The AI half is happening with or without us. What we do is make sure that when the physical tickets leave the service desk, they land somewhere that runs them as the logistics operation they always were.

Talk to us about the physical half