What is an SLA?
An SLA (service level agreement) is a promise about speed or uptime: we answer web leads within a set time, or a vendor answers you within theirs. For a small shop it is often just the reply window you tell customers. A vendor's SLA is their promise to you, on their contract, not a number you invent.
Also called: service level
In plain English
If you say you call back the same day, the workflow should measure that and escalate when it slips. If you have not decided, do not let an assistant promise 'within the hour' because it sounds helpful. Latency is how long a single tool takes to respond. An SLA is the promise a human still has to keep. An SLA is the promise you are willing to keep on a normal week, not the promise a sales page made. 'We answer new leads the same day' is an SLA if you staff for it. A bot that replies in seconds does not meet the SLA if the reply is useless and the human takes three days. Measure from the customer's view. Write the promise down. Do not invent a percentage you have never counted.
How an owner would use it
Write the one window you can keep. Use the agent to draft inside that window, not to announce a faster one. If a vendor quotes an SLA for their AI product, read the contract. Do not repeat it to your customers as if it were yours.
Common mistakes
Letting a model pick a response time creates an SLA you cannot meet. Copying a software vendor's uptime promise onto your own service is a different business. Do not hide a missed SLA by sending a cheerful bot message that resets the clock. Escalate.
A worked example
A mobile mechanic
The text promises the window on the booking page
This is an illustration, not a customer result. The booking page says the shop confirms next-day jobs by 6 p.m. The missed-call text uses that sentence and nothing shorter. If 6 p.m. passes with no confirmation, the job lands on the owner's list instead of the bot promising a new time.
Related terms
Prompts and guides
Questions owners ask
What is an SLA?
An SLA is a stated promise about how fast or how reliably something happens. For many shops it is the reply window customers were told.
How should an AI tool treat our SLA?
It may repeat the window you wrote and flag misses. It should not invent a tighter promise to sound attentive.
What is latency, next to an SLA?
Latency is the wait for a system to answer a single call. Your SLA is the customer promise. A fast model does not keep a callback promise by itself.
Product guides are on AI apps for business.