Chat Session
Definition
Chat Session
A chat session is one live text conversation between a customer and an agent, human or bot, carried on a website widget, mobile app, or messaging channel. It runs from the first message to a defined close event, such as an inactivity timeout.
Every live chat, WhatsApp thread, and Facebook Messenger dialogue a brand handles counts as one session. Sessions carry their own timestamp, transcript, queue history, and outcome tag, and that record feeds reporting on volume, handle time, and satisfaction.
Digital-first buyers pushed chat past email in most Western markets. Salesforce’s State of the Connected Customer research reports that 57% of consumers prefer messaging channels for routine service, and 71% expect a response within five minutes.
Because a session can be human, bot-assisted, or fully automated, it sits where the contact center, the CRM, and the workflow automation stack meet. Live chat widgets still generate most sessions, though messaging apps keep taking share.
Key takeaways
- A chat session is one complete text conversation, opened by the first message and closed by inactivity, resolution, or a manual end.
- Sessions are the base unit for chat KPIs: concurrency, average handle time, first-contact resolution, and post-chat CSAT.
- A trained agent typically runs two to four sessions at once, against one voice call at a time.
- Chatbots deflect most tier-0 sessions, and humans take over on escalation through a warm handoff.
- Offshore providers in the Philippines, India, and Colombia staff chat around the clock at 40–60% of onshore cost.
How it works
A chat session opens when a customer clicks the widget, messages a business number, or triggers a proactive prompt. The platform issues a session ID, routes the request through queue logic, and drops it into the next free agent’s console.
The session closes on one of three events: the customer confirms resolution, the agent clicks “end chat,” or an inactivity timer expires after 5–10 minutes with no reply.
Everything between open and close belongs to that session — bot handoffs and transfers included.
Concurrency is what separates chat from voice. One agent runs several sessions in parallel — the norm is 2 to 4, though skilled agents on simple queues push past six. That multiplier is why workforce management maths for chat never copies voice.
Five formulas govern almost every chat capacity plan you will be asked to sign off.
| KPI | Formula | Healthy benchmark |
|---|---|---|
| Chat handle time | Session open to close, agent time only | 6:00–10:00 min |
| Concurrency | Active sessions ÷ logged-in agents | 2.0–4.0 |
| Chat abandonment | Sessions ended before agent reply ÷ total offered | Under 5% |
| First-contact resolution | Sessions closed without transfer or re-open ÷ total | 70–80% |
| CSAT (post-chat) | Positive responses ÷ surveys returned | 85%+ |
Those bands come from ContactBabel’s *US Contact Center Decision-Makers’ Guide*, 2024 edition, and from ICMI operational surveys. A result below the healthy band usually points at routing, staffing, or knowledge-base gaps rather than agent skill.
Average handle time behaves differently here too. It counts agent-active minutes, not wall clock — which is why the same hour can log as three separate sessions of six minutes each.
A chatbot usually owns the opening exchange, much as an interactive voice response menu fronts a phone queue. Anything the bot cannot close escalates to a human, and the transcript travels with it inside the same session ID.
Quality teams then score the closed session on two outcomes: first-call resolution, adapted to mean no transfer and no re-open, and customer satisfaction from the post-chat survey.
Zendesk, Intercom, Salesforce Service Cloud, and LivePerson all log intent and sentiment beside those two scores, and that metadata is what feeds coaching and forecasting.
Examples
Offshore providers publish chat numbers, and they cluster tightly. Read the five cases below and the same pattern shows up: about three sessions per agent at once, a bot absorbing password reset traffic, and resolution rates inside the ContactBabel healthy band.
Concentrix runs 24/7 chat pods for a US telco out of Manila and Bogotá. Agents handle three concurrent sessions on average, and the bot deflects roughly 45% of tier-0 password reset traffic before a human sees the queue.
Against the 2024 bands above, that pod sits mid-range on concurrency.
Teleperformance staffs Shopify’s post-purchase chat channel from Lisbon and Cairo. Session volumes triple across the November and December peak, so forecasting treats chat, email, and voice as three separate backlogs with three different service level agreements.
TTEC’s Digital unit pairs chatbots with human escalation for a US health plan client. In its 2024 case study, average session length fell from 12 minutes to 7 after bots authenticated members before the human agent joined the conversation.
Foundever, the provider formed from Sitel and Sykes, reports that its Colombia and Philippines chat teams average 3.2 concurrent sessions per agent, with 78% first-contact resolution for retail clients — inside the 2024 ContactBabel band on both counts.
Alorica and Genpact both run hybrid pods, where one agent takes two chat sessions plus a back-office ticket during quiet minutes. Through 2025 that blend has kept spreading, because it bills one seat while clearing two queues.
Related terms
These terms sit closest to a chat session, and buyers tend to meet them in the same breath. Each one either feeds the session, measures how it went, or sits one channel across from it.
- Contact Center: the multi-channel operation that houses chat, voice, email, and social service in one queue system.
- Live Chat: the synchronous text channel that still produces most chat sessions.
- Chatbot: the automated agent that handles tier-0 sessions before human handoff.
- Average Handle Time: the per-session duration metric, and the core chat efficiency KPI.
- First Call Resolution: the chat-adapted measure of sessions closed without transfer or re-open.
- Customer Satisfaction Rating: the post-chat survey score that gates coaching cycles.
- Interactive Voice Response: the voice-channel equivalent of the pre-chat bot greeting.
FAQ
What counts as one chat session?
One session runs from the customer’s first message to a defined close event: resolution, an agent-ended chat, or an inactivity timeout. Bot minutes and warm handoffs to a human stay inside the same session record.
How many chat sessions can one agent handle at once?
The contact centre norm is 2 to 4 concurrent sessions per trained agent. Simple scripted queues push past six, while complex technical or regulated queues drop back to one or two.
How is chat handle time different from call handle time?
Chat handle time counts only the minutes a specific agent is active on the session, not the wall-clock duration. Because agents run sessions in parallel, the same 15 wall-clock minutes can log as 5 minutes of handle time on each of three sessions.
What is a good chat abandonment rate?
Under 5% is the healthy benchmark in ContactBabel’s 2024 guide. Above 8% usually signals under-staffing at peak, weak proactive routing, or a pre-chat form that asks too much.
Do chatbots count as chat sessions?
Yes. Bot-only sessions log the full transcript, intent tags, and outcome, whether that outcome is deflected, escalated, or abandoned. Most dashboards report them apart from human-handled sessions so the CSAT figures stay comparable.
Where do offshore BPOs sit in the chat stack?
Philippines, India, and Colombia providers deliver 24/7 chat coverage at 40–60% of onshore cost, and Concentrix, Teleperformance, Foundever, TTEC, Genpact, and Alorica all publish chat-specific pods.
See which providers deliver 24/7 chat coverage at offshore rates in the Outsource Accelerator directory.
Independent
