The escalation escalator at a financial services firm

Rough monthly cost of the escalation path:

• L1: 9,000 tickets × £12 = £108,000
• L2: 55% of 9,000 = 4,950 tickets × £45 = £222,750
• L3: 40% of 4,950 = 1,980 tickets × £120 = £237,600
• Total ≈ £568,350 per month

Scaling to ~£6.8m a year, of which the majority is spent at the two most expensive tiers.

What the number tells you:

1. The pyramid is inverted. The highest-paid engineers are doing the highest-volume, lowest-complexity work. That is a capability misallocation, not a volume problem.
2. Escalation is the default, not the exception. A 55% L1 escalation rate means first line is functioning as a switchboard. A 40% L2 escalation rate means second line is not resolving either - so the design assumption that each tier filters is not holding.
3. Handle time is driving the behaviour. Agents are being rational: the measured metric rewards handing off quickly. The 3.1 CSAT is the predictable result of a customer being transferred repeatedly.
4. Illustrative upside: if L1 escalation fell to 30% and L2 escalation stayed at 40%, monthly cost drops to roughly £359,000 - around £2.5m a year - with the same ticket volume.

Why AHT is harmful here:

• It measures activity, not outcome. A 40-second call that escalates the customer three times scores well; a six-minute call that resolves the issue permanently scores badly.
• It creates the escalation behaviour directly. The instruction not to 'waste time' on diagnosis is the AHT target expressed as a policy. First line is incentivised to transfer.
• It hides the true cost. AHT is capped at first line. The cost of the escalation appears in a different cost centre, so the metric that drives the behaviour is invisible to the people held to it.

Replace with a small balanced set:

• First-contact resolution - the primary outcome measure for first line.
• End-to-end resolution time for the customer, not per-tier time.
• Repeat contact rate within 7 days for the same issue - catches premature closure.
• Cost per resolved ticket end-to-end, spanning all tiers, so escalation shows up where the decision is made.
• CSAT on the resolved ticket.

Guard against: moving from AHT to FCR without changing incentives. If FCR becomes the only measure, agents will avoid difficult tickets, which pushes complexity back up the tiers in a different way. Pair the outcome measure with a quality sample and with explicit authority to resolve.

Design:

1. Segment the demand first. Take the top 10 issue types by volume - likely access changes, password resets, VPN and common application errors. These are the shift-left candidates; the long tail is not.
2. Remove the top of the funnel. Automate self-service for password reset and standard access requests so they never become tickets at all. This is cheaper than resolving them at L1.
3. Move the knowledge and the authority together. For each candidate type: a validated article, a documented decision rule, and formal authority for L1 to act without asking L2. Authority without knowledge creates incidents; knowledge without authority creates escalation.
4. Create a feedback loop from L3 down. L3's 30% of time on resets is the best source of candidates. Require every L2/L3 resolution of a repeatable type to produce a draft article and a training note - this is the classic 'earned escalation' discipline.
5. Train and certify by issue type, not by tier. Use a skills matrix so 'can resolve access changes independently' is a recorded competence.
6. Track the migration. Target L1 escalation at or below 30% within two quarters, with the savings re-invested rather than banked, or L1 will resist.

Risks and controls:

• Wrong resolution at first line → certify before authorising, sample QA, and keep a fast escalation route.
• Shadow escalation via chat → make escalation visible and measured, never informal.
• L2/L3 feel devalued or threatened → give them ownership of the shift-left pipeline as a career step, not as job loss.
• Volume disguises improvement → report the escalation ratio and end-to-end cost, not raw counts.
• Security exposure on access changes → standardise and automate rather than loosening control; automated identity workflows are usually more auditable than manual email approvals.