Most service agreements measure the wrong two things, or measure the right two things in a way that cannot be improved. The confusion is almost always between first response time and resolution time, and the two respond to completely different interventions.
First response time is a staffing and queue problem. Resolution time is a process and information problem. Treating them as one number produces targets that improve on paper while the customer experience stays the same.
First response time: what you are actually measuring
First response time is the interval between a customer opening a conversation and a human reply appearing. In a synchronous channel this is usually short and mostly determined by whether anybody is available. In an asynchronous channel — email, a messaging app outside business hours, a ticket queue — the first reply may not be the useful one, and measuring it encourages behaviour that makes things worse.
The specific failure is the automated acknowledgement. Sending an immediate "we have received your message" response satisfies the metric and adds a turn to the conversation. For a customer with a real problem, it converts one message into two, and it is read as being fobbed off.
Two ways to avoid it:
- Exclude auto-acknowledgements from the measurement entirely, so nobody gets credit for them.
- Measure the first substantive reply instead — first reply from a human that addresses the question. This is harder to automate and considerably more useful.
Resolution time: the number that reflects the operation
Resolution time is when the problem is actually fixed. It is where a shared mailbox with clear ownership beats one where nobody owns anything, and where a knowledge base beats a fast typist.
Resolution time is dominated by two things: how many times the conversation goes back and forth, and how long each round trip takes. The first is a process problem — how many questions the agent has to ask to get enough information to act. The second is a queue problem.
The improvement that moves resolution time most is almost never a faster reply. It is fewer rounds: a better first reply that asks for everything needed in one message rather than three. That is worth measuring directly — count the customer messages per resolved conversation — because it is the variable that actually moves.
Why one blended number hides both problems
A single SLA figure combining both is unfixable in a useful sense. Improve first response and the number barely moves, because resolution dominates. Improve resolution and the number moves, but you cannot tell whether you did it by changing staffing or by changing the process.
Publish them separately, per channel, and for asynchronous support per language if the team is multilingual. The reason to split by channel is that an overnight message and a message sent during business hours have different achievable targets for reasons that have nothing to do with performance.
What an SLA is actually for
An SLA has two audiences and they want different things. A customer wants to know what happens when something goes wrong — which is a commitment about remedy, not about speed. An internal team wants a threshold that triggers a change in staffing or priority.
Most agreements are written for the first audience and then measured by the second, which produces a number that is visible to customers and inert internally. A useful internal SLA names the trigger and the action: if the queue for this language exceeds N for two consecutive hours, the on-call person is paged. If unassigned conversations exceed a certain age, they escalate to a named person. Both are checkable; "we aim to respond within 24 hours" is not.
The commitment side needs the same specificity. What does the customer actually receive when a target is missed — a partial answer, a status update, a refund, nothing? Leaving it unstated means the remedy is renegotiated at the worst possible moment, which is when somebody is already unhappy.
Practical version
For a small team, three numbers and one rule are enough:
- First substantive reply, split by channel.
- Customer messages per resolved conversation.
- Age of the oldest unassigned conversation.
Plus the rule that any unassigned conversation over a set age escalates to a named person. That set of measures catches queue problems, ownership problems, and process problems separately, which means you know which one you are actually fixing.
For how ownership is assigned in the first place, see shared mailbox rules that survive a growing team and a WhatsApp team inbox for small teams. For targets that have to be set per language, see chat translation quality for cross-border teams.
Source note: This article draws from CIO.com's "What is an SLA? Best practices for service-level agreements" (https://www.cio.com/article/274740/outsourcing-sla-definitions-and-solutions.html), which discusses service-level agreement definitions and practice. The separation of first substantive reply from first response time, the rounds-per-conversation measure, and the trigger-plus-action design for internal SLAs are our own operational analysis. No vendor claims from the source are used as verified data here.