FullSession | Chatbot vs live chat: when to use AI, a person or both | Roman Mohren | October 2026
Article: https://www.fullsession.io/blog/chatbot-vs-live-chat

OPERATING-MODEL CHECKLIST: AI CHATBOT ONLY, LIVE CHAT ONLY, OR BOTH IN ONE THREAD
A reusable working tool. Every field is blank on purpose: fill it in from your own queue and your own setup. Nothing here reports a completed test or a measured result.

HOW TO USE THIS FILE
Part A sorts a sample of your real conversations into jobs.
Part B asks the five decision questions.
Part C maps each job to an owner and a handover rule.
Part D records your hours and the waits you can honestly promise.
Part E compares billing units for the models you are considering.
Part F is a two-question walkthrough test: one repeat question and one exception.
Use "visitors" for people on your public website and "customers" for people inside your product. Keep the two separate in your notes.


PART A: QUEUE SAMPLE
Take a recent sample of conversations (for example, the last 100 across website and in-app chat).

Sample period (dates):
Number of conversations sampled:
Channels included (website chat / in-app chat / other):
Who sorted the sample:

Count each conversation once, under the job that best describes it:

Job 1. Repeat question with an approved answer
Count:
Three typical examples, in the customer's or visitor's own words:
1.
2.
3.
Approved source that answers them (article, page or policy, with version or date):

Job 2. Question that arrived outside staffed hours
Count:
How many of these needed a person, rather than an answer?

Job 3. Account-specific question (plan, seats, order or usage)
Count:
Account data needed to answer:
Is that data available to an AI through a permitted connection? (yes / no / partly):

Job 4. Account exception (refund outside policy, plan exception, disputed charge)
Count:
Who holds the authority to decide:

Job 5. Judgment or sensitive topic (security, data deletion, contract terms, an upset customer)
Count:
Topics your policy says must always go to a person:

Job 6. Sales question from a website visitor (pricing, plan fit, product)
Count:
Approved sources that answer it (pricing page, plan comparison, product pages):

Job 7. Negotiation, custom terms or procurement
Count:
Who owns these conversations:

Job 8. Multi-step task inside the product (setup, configuration)
Count:
Steps that could be guided or completed with permission:
Steps that must stay with a person:

Conversations that fit no job (describe):


PART B: THE FIVE DECISION QUESTIONS
Answer in writing. Short answers are fine; blank answers are a finding.

1. How many conversations in the sample have an approved answer you would let software give?
Answer:

2. Which exceptions need authority, who holds it, and during which hours is that person available?
Answer:

3. What hours do you staff, and what wait can you honestly promise outside them?
Answer:

4. Which account data and actions should the AI be permitted to use, and through which connection?
Answer:

5. Which billing unit moves the way your volume and staffing move (seat, human conversation, AI resolution or outcome, AI-assisted conversation)?
Answer:

Model your answers point to (AI chatbot only / live chat only / both in one thread):
Reason, in one sentence:
If AI chatbot only: where exceptions go, and how customers and visitors are told before they need it:
If live chat only: what happens to repeat questions during peaks and outside staffed hours:
If both in one thread: the three handover conditions below must be checked in Part F.
- The person replies in the same conversation.
- The context travels: transcript, what was tried, permitted account context, page.
- One named owner, and an honest time for the next update.


PART C: JOB-TO-OWNER MAP
For each job, decide who answers first, when a person takes over, and who that person is.

Job | Answers first (AI / person) | Handover rule (when a person takes over) | Person or team who owns it | Reply time you will promise
1. Repeat question       |          |          |          |
2. Outside staffed hours |          |          |          |
3. Account question      |          |          |          |
4. Account exception     |          |          |          |
5. Judgment / sensitive  |          |          |          |
6. Sales question        |          |          |          |
7. Negotiation / terms   |          |          |          |
8. In-product task       |          |          |          |

Request for a person at any point: what happens, and how fast?
Answer:


PART D: HOURS AND HONEST WAITS
Office hours (days, times, time zone):
Away message, word for word:
Does the away message say when the team is back? (yes / no):
Does it promise anything nobody is staffed to deliver? (yes / no):
Where the reply will appear (same thread / other channel, tested):
How questions left after hours are assigned at the start of the next business day:
Who covers when the exception owner is out:
How the AI is set to answer outside office hours, per surface (website / in-app), as confirmed in your own setup or walkthrough:


PART E: BILLING-UNIT WORKSHEET
Fill one column per option you are evaluating. Use each vendor's own current pricing page and write the date you checked it. Do not compare prices until the units are written down.

Option name:
Billing unit for people (seat / human conversation / none):
Price per unit and commitment (monthly or annual), from the vendor's page:
Date checked:
Billing unit for AI (resolution / outcome / conversation / other):
How the vendor defines that unit, in its own words:
Monthly minimum, if any:
What is included (inbox, human chat, integrations):
What is excluded (recording, add-ons, taxes, third-party subscriptions):
Seats your coverage needs (people x hours):
Estimated monthly AI units from Part A (your estimate, labelled as an estimate):
Your team's time on exceptions (hours per month, your estimate):

FullSession Max AI reference terms (recheck at https://www.fullsession.io/pricing?solution=desk before use):
- $0.39 per AI-assisted conversation; $99 monthly minimum credited toward conversations (you pay the higher of $99 or your conversation charges).
- One conversation = one user in one thread for a fixed 24 hours from Max's first substantive answer, guidance or supported action.
- Greetings, clarification-only exchanges, human-only chat, internal tests, failed starts and handoffs before substantive Max help are not billed. Human replies add no charges.
- Desk app, shared inbox, human chat and supported integrations included. No seat, MAU, app or per-lead fees.
- Optional Core CX / UX recording, taxes and third-party subscriptions are separate.
Your estimated AI-assisted conversations per month:
Max charge = the higher of $99 or (conversations x $0.39):


PART F: WALKTHROUGH TEST (ONE REPEAT QUESTION, ONE EXCEPTION)
Run both in a walkthrough or a test workspace. Record what you actually observe, with the date and device. Expected results are what to look for, not results.

Test 1: repeat question
Question used (from Part A, Job 1):
Surface (website / in-app):
Date, time and device:
Expected: an answer from your approved source, with the source identifiable.
Observed answer:
Source the answer came from:
Was a supported step guided or co-driven? (yes / no / not applicable):
Did the customer or visitor need a person? (yes / no):
Notes:

Test 2: exception
Exception used (from Part A, Job 4 or 5):
Surface (website / in-app):
Date, time and device:
Expected: the AI explains what it can, the request for a person is honored, a teammate takes over in the same thread, and an owner is assigned.
What the AI said before the handover:
How the person was requested (customer or visitor asked / policy rule):
Did the teammate reply in the same thread? (yes / no):
Could the teammate see the earlier conversation without asking again? (yes / no):
Owner assigned (name or team):
First human reply, word for word:
Did it state when the next update would come? (yes / no):
Repeat outside office hours: away message shown, word for word:
Question left after hours reached the inbox? (yes / no):
Reply appeared in the same thread when the customer or visitor returned? (yes / no):
Notes:

Decision after testing (model chosen, date, owner):
Next review date:

Related reading:
Chatbot-to-human handoff: https://www.fullsession.io/blog/chatbot-human-handoff
AI chatbot pricing: https://www.fullsession.io/blog/ai-chatbot-pricing
Support & Sales Desk: https://www.fullsession.io/solutions/support-sales-desk