Chat onboarding has an unfair advantage
Every conversation your team has ever had is written down. New phone agents inherit folklore; new chat agents inherit a searchable corpus of real questions with real answers attached. A one-week ramp is realistic if you actually use it.
Day 1-2: read before you write
Assign fifty transcripts, curated: twenty routine (the shape of normal), ten escalations (where the edges are), ten great saves (what good looks like), ten failures (honestly labeled as such). The reading assignment does what no manual can — it teaches tone, pacing, and the top questions simultaneously. Pair it with the canned response library, presented as scaffolding to personalize rather than scripts to fire.
Day 3: shadow live
The trainee watches an experienced agent work real conversations — not a demo, the actual Tuesday queue. The narration matters more than the watching: WHY this canned response, WHY that silence got narrated, WHY this one goes to a call. One hour of narrated shadowing beats a day of silent observation.
Day 4-5: answer with a safety net
- One conversation at a time. Concurrency is a skill for month two, not day four.
- A named reviewer nearby who can be pinged mid-chat and who transfers in cleanly (with the transcript) when a conversation outgrows the trainee.
- Scoped topics: the routine tier first — password resets, billing lookups, how-tos. Angry customers and refund exceptions stay with seniors until the fundamentals are automatic.
Make the mistakes cheap by configuration
Set the new agent up so early errors cannot compound: their own login (never a shared one — the transcript history IS the coaching record), routing that sends only their scoped departments, and no admin or billing permissions on week one. In MyLiveChat, per-agent accounts and department routing are standard settings — the safety net is configuration, not vigilance.
The habit that makes it stick
Week two onward: a weekly fifteen-minute review of the trainee's own transcripts — two they are proud of, two that felt hard, read together with the reviewer. This is the same transcript-review ritual the whole team should run; starting it in onboarding means it never has to be introduced later as criticism. Graduation is measurable: concurrency raised when narrated-silence and first-response habits hold; scope widened when the escalation rate to the reviewer falls.
What to hand over on day one
New agents are usually given access and a product tour, then left to infer everything that
actually governs the job. The inference takes weeks and produces inconsistency, when most of it
could have been written down once.
Give them the boundaries in writing. What they can decide alone, what needs approval, and what
must never be done in chat — the refund limit, the discount rule, the fact that passwords
and card numbers are never collected. A new agent without these makes them up under pressure, and
the resulting decisions are the ones that get escalated later.
Give them the honest map of what is broken. Every product has known issues, common workarounds
and areas where the documentation is wrong. Teams treat this as tribal knowledge and let newcomers
discover it by getting caught out in front of a customer; a single page listing the current known
problems saves a fortnight of that.
Give them the vocabulary. Which internal names map to which customer-facing names is a constant
low-level source of confusion, and getting it wrong in a transcript is visible to the customer.
And tell them explicitly that saying “let me check that properly” is encouraged. The
most common new-agent failure is guessing rather than pausing, and it happens because nobody said
that pausing was allowed.
The first weeks, sequenced
Reading before writing is the right order, and it is worth being deliberate about which
conversations they read. A random sample teaches the average; a curated set of ten or fifteen
— including a difficult complaint, an escalation, a technical question and an outstanding
example — teaches the range, which is what they actually need.
Then let them draft without sending. Having a new agent write the reply while an experienced
agent handles the conversation gives real practice with no risk, and the comparison between the
two answers is the fastest feedback available.
Move to live conversations with a safety net rather than a deadline. The variable is confidence
and it differs per person; someone ready on day three and someone ready in week two both become
equally good agents, and forcing the schedule mostly produces cautious, over-hedged replies.
Start them on a narrow band of conversation types — order status, straightforward
how-to — and widen it deliberately. Being thrown into billing disputes in week one teaches
avoidance rather than competence.
Keep the feedback specific and frequent early, then taper. Daily review of a few transcripts in
week one is proportionate; the same intensity in month three is surveillance and is read as
such.
What to measure
Do not measure new agents on speed for the first weeks. Handling time falls naturally with
familiarity, and making it the early target produces fast, shallow answers and a habit that is
hard to undo later.
Watch escalation rate over time instead. It should start high — that is correct behaviour
— and decline steadily. A rate that stays high suggests missing authority or missing
knowledge; one that drops very quickly can mean the agent has started guessing, and the
transcripts will show which.
Ask new agents what they were unsure about at the end of each week and keep the answers. It is
the cheapest source of onboarding improvements you have, and it is only available for a short
window — once someone knows the job, they can no longer remember what was confusing.
A new agent’s first question is often simply which application to open. Which agent console your team should use covers the four builds on offer, and why a mixed team is a normal configuration rather than a problem.