Why teams outsource chat, and what actually goes wrong
Outsourcing chat is usually a coverage decision rather than a cost one. A small team cannot staff
evenings, weekends and holidays without either burning people out or hiring far ahead of demand, and
a contract team fills those hours at a price that makes sense. That logic is sound. What goes wrong
is almost never the agents' ability — it is that they were given a login and a product name
and left to infer everything else.
An in-house agent absorbs context continuously: they hear the release discussion, they know which
customer is angry this week, they know that the pricing page is out of date. An outsourced agent
gets none of that by default. Every failure mode of outsourced chat traces back to that gap, which
means the fix is not stricter supervision but better transfer of context.
Write the brief before the contract
The brief is the whole job. If you cannot write it, you are not ready to hand chat over, and
writing it usually improves your in-house operation too. It does not need to be long — a few
pages beats a manual nobody reads — but it needs to be specific.
- Scope. Which questions they answer, and which they pass on. Be concrete: name
the topics, do not describe a philosophy.
- The top twenty questions and your actual answers. Not links to where the answer
might be. The answers.
- Voice. Two or three rules with examples of good and bad. Formal or relaxed,
whether contractions are fine, whether you use the customer's first name.
- Authority. What they can decide alone — a refund up to a value, a
shipping goodwill gesture, a discount — and what always escalates.
- Escalation. Who, by what route, at what hours, and what they say to the visitor
while it happens.
- The never list. Commitments they must never make: delivery dates, roadmap
promises, security or compliance claims, anything about a competitor.
That last one deserves emphasis. An agent who does not know your compliance position will improvise
one under pressure, and a confident wrong claim about how customer data is handled is far more
expensive than an unanswered question.
Decide the access question deliberately
Outsourced agents need enough access to do the work and no more, and the default of “give
them what everyone else has” is rarely right. Every person needs their own named login rather
than a shared team account, because a shared login makes it impossible to attribute a conversation
or to remove one person when they roll off. Expect turnover on a contract team to be higher than
yours, and treat offboarding as a scheduled routine rather than an event — ask your provider
for joiners and leavers in writing on a fixed cadence, and reconcile it against your own account
list.
Be realistic about what the tooling can enforce. The role model in most chat products is coarse
— administrators and agents — so the meaningful controls are which accounts exist, who
holds admin, and how quickly an account disappears. Assume an outsourced agent can read the
transcripts they handle, decide whether that is acceptable for your content, and if it is not, keep
the sensitive categories in-house rather than trying to partition them with settings.
Run the first fortnight properly
The pattern that works is a narrow start that widens on evidence. Give the contract team one
clearly bounded slice to begin with — a single product area, or the overnight window only
— and read everything they send for the first week. It is tedious and it is the highest-value
time you will spend on the engagement, because the corrections you make in week one become the
standard for everyone who joins the account later.
Give feedback to the team lead rather than to individual agents, in writing, with the transcript
attached and the better version written out. “Be warmer” is not actionable across a
language and a company boundary; “here is what we would have said, and why” is. And ask
them what is missing — an outsourced team hits your documentation gaps faster than anyone
in-house, because they cannot fill them from memory.
Keep the loop running once it is quiet
The failure mode after a successful launch is silence. Chat runs, nobody complains, and nobody
reads a transcript for four months, by which time the answers have drifted away from your current
pricing and policies. Put a small recurring review in the calendar — ten conversations a week
read by someone in-house is enough to catch drift long before a customer does — and treat
every product or policy change as an event that triggers a brief update, with a named owner on your
side rather than the provider's.
What to measure
Compare outsourced and in-house chats on the same metrics rather than inventing separate ones,
and segment rather than averaging the two together. Escalation rate is the most
informative single number: rising escalations usually mean scope has quietly widened past the brief,
while a suspiciously low rate can mean agents are guessing instead of passing things on. Watch
repeat contacts within a week, which catches answers that satisfied the visitor in
the moment and were wrong. Track first response time in the covered hours, since
coverage was the reason for the arrangement. And read a sample yourself every month — no
dashboard will tell you that the tone has slowly stopped sounding like your company, and that is
the thing customers notice first.
If the partner team does not work in English, check the specific screens they will live in before you promise anything — which parts of your dashboard change language covers how far the translation actually reaches.