01Measure your real baseline
Most teams underestimate their first-reply time because they remember the fast replies, not the slow ones. Pull the actual timestamps — form submission to first reply, for the last 20 enquiries — before you fix anything. You're looking for the average, but more importantly the worst case: the one that sat for two days because it landed on a Friday evening. That worst case is usually where you're losing business, not the average.
02Build a triage rule, not a rota
A rota ("Priya covers evenings") breaks the moment Priya is on holiday. A triage rule survives that: sort every incoming enquiry into one of three buckets before anyone decides who answers it.
- Answerable immediately — pricing, availability, "do you do X" — these should never wait for a specific person to be online.
- Needs a quick human judgement call — a custom quote, an edge-case question — route these to whoever's actually free, not whoever's "supposed" to cover that hour.
- Needs the founder or a specialist — genuinely rare; treat it that way, and don't let everything default into this bucket out of caution.
03Templates beat typing from scratch
Write five templates for your five most common enquiry types, then edit rather than compose from a blank box. This alone typically halves the time a fast reply takes to write, and it's the single highest-leverage change a small team can make with zero new tools.
04Decide what a human must see
Before you automate anything, write down what should never go out without a person reading it first — a refund, a complaint, anything legally sensitive. This list is short for most businesses, and it's the thing that makes automation safe rather than risky: everything outside that list can move fast, everything inside it should always wait for a person.
05Where automation actually helps
Once you've got a triage rule, templates, and an escalation list, you've done the hard thinking — the part that actually determines whether "fast replies" means better service or sloppier service. At that point, an AI agent like Leo is mostly executing the system you've already designed, continuously and outside office hours, rather than replacing judgement you haven't defined yet. That's also why it tends to work better bolted onto a team that's already done this thinking than as a shortcut past it.