The AI layer takes the repeat questions
Order status, delivery windows, policy questions — resolved instantly, with live system access rather than a canned reply.
Live Chat Support
RunThe visitor with a question does not come back tomorrow.
Staffed chat with an AI layer on the repetitive share — covered to a response window you set, including the hours nobody wants.
Typical stack
- Intercom
- Zendesk
- Front
- Shopify
- Claude
- Slack
Written for: A team whose chat widget is answered when someone happens to be free.
Who it is for
Is this built for a business like mine?Built for businesses whose visitors arrive outside office hours.
Chat pays where the visitor is close to a decision and the question is small. These are the shapes where that is true.
What we see
Pre-purchase questions about sizing and delivery arrive at eleven at night, which is when the traffic is.
What we build
Coverage across your real traffic curve, with order lookups answered live rather than promised by email.
The detail that matters
Agents read live order state. "Let me check and get back to you" is not an answer.
What we see
Tier-one questions interrupt engineers who are in the rotation and resent it.
What we build
A trained tier-one desk grounded in your current docs, escalating with a written diagnosis rather than a forwarded transcript.
The detail that matters
Escalations arrive reproduced, not relayed.
What we see
Enrolment season multiplies enquiry volume tenfold against the same three admissions staff.
What we build
Seasonal capacity that scales for the window and scales back down after it.
The detail that matters
Capacity is contracted by window, not by permanent headcount.
What we see
Portal and site enquiries sit unactioned while agents are out showing property.
What we build
Immediate qualification and routing, with the enquiry in the CRM before the agent is back in the car.
The detail that matters
Speed is the product here. Minutes decide the conversion.
What we see
Users ask about a pending transfer, a card decline or a stuck verification at two in the morning, and money questions do not wait until business hours.
What we build
Tier-1 resolution on account, payment and verification questions with live system access, escalating to Tier-2 on written conditions with the case already investigated.
The detail that matters
The Tier-1 / Tier-2 line is a written condition, not a judgement call — and anything touching a compliance decision leaves the desk entirely.
And who it is not for
If your chat volume is under a handful of conversations a day, an AI layer alone will serve you better than a staffed desk. We will size that honestly.
The problem
Do they understand what is actually going wrong?A chat widget nobody watches is worse than no widget.
It advertises availability and then fails to deliver it. The visitor who typed a question and waited four minutes did not go and find your phone number — they went back to the results page they came from, and the widget converted them for somebody else.
Chat is answered by whoever is free, which at 8pm is nobody.
First response time is unmeasured, so nobody knows it is four minutes.
Agents cannot see order or account state, so every answer is a promise to check.
Volume triples in a peak week and the same two people are covering it.
What we build
What exactly would I be buying?A queue with a number on it.
What you are buying is a response window that holds under load, not a number of people. That means the repetitive share is absorbed automatically, the rest reaches a trained person, and the split is measured rather than assumed.
People take the rest
Trained on your product and your tone, with the transcript and context already attached when they pick up.
Escalation by rule
Handoff happens on conditions you set — topic, order value, an explicit request — not on a model scoring its own confidence.
A service level that is reported
First response time, resolution rate and the automated share, reported weekly against the target you agreed.
How it works
How does this actually function?What happens in the first thirty seconds.
The routing decision is made before anyone is waiting, which is why the response window holds when volume spikes.
Message
Classify
Resolve or route
To a personRespond
Log
Classification happens on arrival, so a routine question is answered while a complex one is already queued to a person.
The agent receives the full conversation and the account state. The customer never repeats themselves.
Every conversation is written back to your systems, so the record does not live in our tool.
What changes
What is different afterwards?What is different afterwards.
Coverage matches your traffic
Staffed against your real traffic curve, including the evening and weekend share that currently goes unanswered.
Chats answered inside the response window, by hour.
Peaks stop breaking the desk
The automated layer absorbs the repetitive spike, so a campaign or a season does not become a hiring problem.
Response time held during peak weeks.
The channel becomes measurable
Volume, topic, resolution and escalation reported weekly, which turns chat from a widget into an operation.
Weekly service level reporting against target.
How we deliver
How does this start, and what do I get at each step?Six stages, and every one has an exit.
You can stop after any stage with something useful in hand. That is the point of naming the artefacts rather than the activities.
- 01
Volume study
Real conversation history — when it arrives, what it asks, what share is repetitive.
- Volume by hour and day
- Topic taxonomy from real chats
- Automatable share, quantified
- 02
Coverage design
The response window, the hours, and what happens when volume exceeds forecast.
- Coverage plan and service level
- Overflow procedure
- Escalation rules, written by you
- 03
Knowledge build
The answers, the tone, and the boundaries — what the desk must never answer alone.
- Knowledge base
- Tone guide
- Never-answer list
- 04
Integration
Access to the systems holding the answers, so the desk resolves rather than relays.
- Platform access configured
- Order and account lookups live
- CRM write-back
- 05
Pilot
One shift pattern, monitored, with your team reviewing transcripts daily.
- Pilot transcripts reviewed
- Service level measured
- Corrections applied
- 06
Run
Full coverage, weekly reporting, and a standing review where the knowledge base actually changes.
- Weekly service level report
- Knowledge base updates
- Quarterly review
Who runs the desk
How does this start, and what do I get at each step?Every desk ships with a lead and a QA function.
A pool of agents with no accountable owner is how outsourced quality degrades — slowly, invisibly, and then all at once. Four roles exist on every account, and the smallest engagement gets all four.
- 01
Agents
Trained on your product, your tone and your escalation boundaries, and assessed before they touch live work. Named and consistent on dedicated plans.
- 02
Team Lead
Accountable for the queue rather than working in it — coverage, SLA, escalations and the shift handover. One person you can name when something goes wrong.
- 03
QA
Samples completed work against a written rubric, independently of the lead. Disagreements between agents are treated as a defect in the knowledge base, not in the agent.
- 04
Account lead
Runs the weekly report and the standing review where the knowledge base actually changes. The person who tells you the number went the wrong way.
QA sampling rate and the review cadence are agreed during onboarding and reported weekly against target — including the weeks it was missed.
Questions
But what about the thing that worries me?The questions people actually ask.
Are the agents dedicated to us?
On dedicated plans, yes — the same named people, trained on your product. Shared plans cost less and use agents covering several accounts in the same category. We recommend based on your volume rather than on margin.
How much gets handled without a person?
Typically half to three quarters for e-commerce and lower for technical products, and we measure it during the volume study rather than quoting an industry figure. The number is reported weekly, so you can see it rather than trust it.
What if an agent does not know the answer?
They escalate to your team on a rule you set, with the conversation and context attached. Escalation is treated as a knowledge gap and closed, not as an agent failure.
Which platform do you work in?
Yours. Intercom, Zendesk, Front, Gorgias, HubSpot, or whatever you already run — the record stays in your system so ending the arrangement does not cost you your history.
How quickly can this start?
Two to four weeks from the volume study, most of which is knowledge build and integration rather than recruitment.
The other half
The system that fills this queue.
Half the volume arriving here is created by campaigns and by the site itself. When the same team builds both, the desk feeds the knowledge base and the marketing team gets the real objections — which is a loop most businesses never close because the two sides are different vendors.
AI AgentsStart
What is the next step?Tell us the volume, we will tell you the shape.
Send us a week of real numbers — calls, chats, tickets, documents — and we will come back with a coverage model and what it would cost.
Related desks
- ConversationEmail SupportShared inbox and ticket handling with drafting automation and trained agents, worked to a first-response target rather than to inbox zero.
- ConversationVoice CallsA staffed phone line with an AI layer on routine calls — inbound, outbound and overflow, to a service level you set.
- TransactionOrder TakingOrder capture across phone, chat and email — validated against live stock and written into your system, not into a spreadsheet.