← All posts
Operations ·

How to Enable L1, L2 and L3 Support

How to enable L1, L2 and L3 support in a DevOps team — a 90-day transformation built on runbooks, clear escalation and swarming, from someone who has run it in production.

Turning a flat, overloaded team into a layered support model that scales, without the org-chart theatre. This is the work I do with engineering teams, usually in places where downtime has a price tag attached.

Every flat team works fine right up until the night it doesn’t. They all find out the same way: an alert fires, three people open the same dashboard, and twenty minutes pass before anyone admits they assumed someone else had it. Nobody owned the incident because, on a flat team, everyone owns everything. Which is the same as no one owning anything.

I’ve seen what an unowned gap actually costs. A client in the financial sector had hit a connectivity fault once before. The vendor shipped a fix, and it got implemented, but only halfway. The missing half was the backup line, which was meant to be configured on our side and never was. A year later the primary connection dropped, the failover everyone assumed existed wasn’t there, and the market-data feed went dark. Traders stopped receiving rates and sat watching trades they couldn’t make. We escalated hard, on our side and the vendor’s, and it still took the vendor walking the connection end to end to notice the backup line had never been set up. Two hours. An eternity in a trading room. Afterwards we ran a proper post-mortem, owned the part that was ours, stood up reactive monitoring on the connection, and started actually testing failover instead of trusting it. The fix from a year earlier had been real. It just was never finished, and nobody owned finishing it.

Seniority becomes the only triage mechanism. Your best engineer becomes the default first responder for everything, their week gets shredded by interruptions a junior with a decent runbook could have closed in ten minutes, and the platform work that would prevent the next page never gets done. Knowledge stays locked in two or three heads, so onboarding crawls and holidays are tense.

Layering the team fixes this, though not for the reason people assume. It’s not about rank. It’s about knowing, before the next 3 AM page, who picks it up and when it leaves their hands.

What the three layers actually do

The labels are old, but the jobs underneath them have moved on. L1 isn’t a phone queue, and L3 isn’t a fancier firefighter.

Tier Core function What they own Escalates when First target
L1 — First response Service desk Triage, containment, runbooks, ticket ownership, comms Outside the runbook, or response SLA breached First response ≤ 15 min (P1)
L2 — Platform engineers Incident management Diagnosis and fix, runbook upkeep, pipelines, reliability work Cause is systemic or needs a design change Fix or workaround ≤ 4 h (P1)
L3 — Architects Problem management Root cause, architecture, standards, mentoring Recurring cause, regulatory impact, or a major incident Post-mortem / design ≤ 48 h; off the rotation by default

The two mistakes I see most often hide inside that table. The first is treating L1 as a relay that only re-routes tickets. If your first line has no authority to fix anything, you haven’t built support, you’ve added a hop. Give them a real runbook library and let them close the common cases. The second is leaving L3 on the front line. An architect’s job is to make incidents rarer and smaller, not to keep getting paged for the same one. If they’re firefighting every week, that’s a runbook gap wearing a staffing costume.

One more thing worth saying out loud, because it’s where most tiered models quietly fail: not every incident should climb the ladder one rung at a time. For a genuine P1, sequential escalation is just downtime with extra steps. Keep the explicit path for the routine ninety percent. For the majors, pull the right people in together regardless of tier. That’s also, for what it’s worth, where current ITIL thinking has landed.

Incident flow across L1, L2 and L3, with a swarming path for major incidents and a continual-improvement loop back into runbooks Incident flow across L1, L2 and L3, with a swarming path for major incidents and a continual-improvement loop back into runbooks

The first 90 days

You can’t declare an L1 team into existence on a Monday. You build the conditions that make one possible, in roughly this order.

Weeks 1–2. Read the last quarter of incidents. Not the dashboards, the actual tickets. Sorted by category and business impact, the picture is always lopsided: a handful of failure modes eat most of the senior time. That list is the whole transformation in miniature; everything after it is just working the list down.

Weeks 3–6. Build the runbook spine. This is where the real work lives, and where most teams have nothing usable. “Check the logs and restart if needed” is not a runbook. It’s a shrug in prose. A runbook names the dashboard, the log line, the exact command, and how you know it worked. Until your top ten failure modes are written down that precisely, you don’t have an L1 layer. You have juniors guessing.

Weeks 5–8. Make routing and SLAs mechanical. With runbooks in place, the escalation matrix lives in the runbook system as a table, not in someone’s memory. On-call policies and ticket timers escalate automatically on breach. This kills the worst habit in incident response: the “give me another ten minutes, I’ve nearly got it” that turns a fifteen-minute handoff into a two-hour outage. The rule I enforce is blunt. Escalate on the clock, not on a hunch.

Weeks 8–12. Protect L3 and close the loop. Recurring causes get pulled out of firefighting and into deliberate root-cause work. A short weekly review lets L1 surface the patterns that are invisible one ticket at a time. That review becomes your improvement backlog, generated for free by the people closest to the failures. By the end of the quarter the change shows up in numbers you can put on a slide: share of incidents closed at L1, time-to-escalate, and the senior hours handed back to actual engineering.

Where to start

Write the runbooks. The tier structure mostly assembles itself around them. And if you can’t yet write down how to handle your top ten alerts, that’s the finding. It’s exactly where we’d begin.

This is the transformation I run: flat, interrupt-driven teams turned into something that gets steadier with each incident instead of more brittle. If that sounds like your team right now, get in touch.

Work together?

I help teams turn ad-hoc operations into resilient, automated systems.

Request a call →