A chat launcher on a client's site is part of THEIR identity: their colors, their greeting, their voice. The agency's fingerprints belong in the results deck, not the corner of the page. Per-site widget configuration — launcher color, label text, greeting, pre-chat fields — is the mechanism; a five-minute theming pass against each client's palette is the habit. (Contrast rules still apply: brand color, legible label.)
Decide who answers — before launch, in writing
The single biggest co-branded-chat failure is the unanswered widget nobody agreed to staff. Three workable models, each honest if stated plainly:
- Client answers. The agency installs, themes, and configures; the client's staff take the conversations. Give them their own agent logins and a one-page runbook (hours, offline promise, escalation).
- Agency answers — chat-as-a-service. Bill it explicitly, staff it honestly (your response-time promises are now a deliverable), and write the answers-boundary: what your team may say about the client's business and what routes to them.
- AI first, client second. Train the assistant on the client's site and FAQ; humans (theirs) take the handoffs. Often the only sustainable model for small-retainer clients — and the AI's wrong answers point at their documentation gaps, which is billable work.
Report with transcripts, not just counts
"47 chats this month" is a number; three verbatim conversations where the widget rescued a lead is a renewal argument. Monthly, pull each client's transcript highlights: questions their site fails to answer (your next content recommendation), objections prospects raise (their positioning gap), and the saves. The page-to-question map per client is an upsell engine disguised as a report.
Structure accounts for clean handover
Retainers end; the widget should survive the breakup gracefully. Two rules from day one: the chat account for a client's site should be transferable to them (their billing-ready account, or documented so ownership can move), and every human gets their OWN login — shared credentials make offboarding an ex-employee or an ex-agency equally messy. Offboarding checklist: revoke agency agent logins, hand over admin, leave the runbook.
The economics that make it work
Per-site licensing punishes the agency model; unlimited-sites-per-account pricing means each new client site is a snippet paste, not a procurement event. Flat plans keep the chat line in a retainer predictable — see the agency playbook for the fuller pitch, including qualifying project leads with pre-chat questions on your OWN site.
Scoping the retainer so volume does not eat the margin
Managed chat is priced per month and consumed per conversation, which is a structural risk if the
agreement does not say what happens when volume moves. The client whose traffic triples after a
successful campaign is a success story that quietly becomes unprofitable.
Write the volume assumption into the agreement in plain terms: the number of conversations
included, what happens above it, and how often it is reviewed. This is easier to agree at the start,
when nobody is under pressure, than in the month the number is breached.
Be equally explicit about scope of subject matter. Answering questions about the client's product
is a different commitment from handling billing disputes or account changes, and the second requires
access, training and liability that the first does not. Agencies lose money on the drift between
those two more often than on raw volume.
Onboarding a new client's chat in a week
The difference between a smooth launch and a messy one is almost entirely front-loaded. Before
anything goes live you need three things from the client: the ten questions they are actually asked,
the boundaries of what you may say, and a named person who can answer what you cannot.
The escalation contact matters most and is usually left vague. Agree who takes the questions your
team cannot answer, in what channel, and how quickly. Without that, your agents will invent answers
under time pressure, and the resulting corrections are expensive in a relationship you are trying to
establish.
Launch narrower than the client expects. Cover the pages and hours where you are confident, prove
it works, and expand. A limited launch that goes well builds far more trust than a full launch with
visible gaps, and it gives you real numbers to negotiate the next phase with.
What to measure
Track conversations per client per month against the contracted assumption, and review it on a
schedule rather than when it hurts. This is the number that determines whether the account works.
Measure how often your team had to escalate to the client, and what for. A high or rising rate
means the knowledge handover was incomplete, and it is usually cheaper to fix that than to keep
paying the coordination cost.
Report outcomes the client's business recognises — qualified enquiries, resolved questions,
issues surfaced — alongside volume. Volume alone invites the question of whether the service is
worth it; outcomes answer it.