Learn how to work out what each support conversation really costs, from AI credits to human handoffs, and how to use the number to guide decisions.
Most teams can tell you what they pay for their support tools each month. Far fewer can say what a single customer conversation costs them. Without that number, it is hard to judge whether automation is paying off, which channels deserve investment, or when a plan upgrade makes sense. This guide shows you how to calculate cost per conversation step by step, including the parts that are easy to miss.
Why cost per conversation matters
A monthly bill tells you what you spent. Cost per conversation tells you what you got for it.
Once you have the figure, you can:
- Compare channels such as website chat, WhatsApp and Messenger on equal terms.
- See whether a busier month made support cheaper or more expensive per customer.
- Judge whether a change, like improving your knowledge base, actually reduced cost.
- Build a realistic budget before you launch a new channel.
The metric is simple in concept and slippery in practice, because the answer depends on two decisions: what counts as a "conversation", and which costs you include.
Step 1: Define what a conversation is
Before any arithmetic, agree on the unit. Different tools and teams count differently, and mixing definitions produces numbers nobody trusts.
Common definitions:
- A thread: every message exchanged with one customer on one channel until it goes quiet.
- A session: a thread broken into separate conversations after a period of inactivity, for example 24 hours.
- A resolved issue: one customer problem, however many messages it took.
For cost work, the resolved issue is the most meaningful, but a thread or session is easier to count automatically. Pick one, write it down, and keep it constant. Changing the definition halfway through a quarter makes trend lines meaningless.
Step 2: List every cost that belongs in the number
There are usually three layers of cost. Teams that only count the first one understate the true figure.
Direct platform costs
This is your subscription plus any usage charges: the plan fee, extra credits, add-ons such as storage or additional agents, and any per-message fees charged by messaging providers.
Human costs
Every conversation that reaches a person consumes staff time. Take the loaded hourly cost of the people involved (salary, benefits, overhead), multiply by the minutes spent, and attribute it to the conversations they handled.
Maintenance costs
Someone keeps your knowledge base current, reviews flagged answers and tunes the assistant. This time is real, though it is often left out. A fair approach is to estimate the monthly hours and treat the cost as a fixed overhead spread across all conversations.
Step 3: Use the basic formula, then split it
The simplest version is:
Cost per conversation = total support cost in the period ÷ number of conversations in the period
That blended figure is a starting point, but it hides a lot. The more useful approach is to calculate two numbers and combine them.
- 01Gather platform, human and upkeep costs
- 02Count conversations with one fixed definition
- 03Split into AI-only and human-handled
- 04Divide and compare by channel
Split by who handled it
An AI-only conversation and a conversation that ends with a human cost very different amounts. Calculate:
- AI-handled cost per conversation: platform cost attributable to those chats, divided by their count.
- Human-handled cost per conversation: platform cost plus staff time, divided by their count.
- Blended cost: weight the two by the share of conversations each represents.
A worked example
These figures are hypothetical, chosen to show the method.
Suppose a shop receives 1,000 conversations in a month:
- 700 are resolved by the AI agent alone.
- 300 are handed to a person, who spends an average of 6 minutes each.
- The loaded cost of staff time is $30 per hour, so 6 minutes costs $3.
- The platform and upkeep costs total $600 for the month.
Human time: 300 × $3 = $900. Total cost: $600 + $900 = $1,500. Blended cost per conversation: $1,500 ÷ 1,000 = $1.50.
If you allocate the platform and upkeep cost evenly ($0.60 per conversation), the AI-only conversations cost $0.60 each and the human-handled ones cost $3.60 each.
The takeaway is that the blended number moves mostly with the handoff rate. Reducing handoffs from 30% to 20% in this example would cut the total noticeably, without changing a single price.
Step 4: Break it down by channel
Channels differ in volume, customer behaviour and cost structure, so a single average can mislead.
| Channel | What to watch |
|---|---|
| Website chat | Often high volume with many short, simple questions; usually cheapest per conversation |
| Longer, stretched-out threads; provider fees and the 24-hour messaging window can affect cost | |
| Messenger | Similar to WhatsApp in thread length; check for repeat visits inside one thread |
Run the same calculation per channel. If one channel has a far higher handoff rate, its knowledge gaps are probably the cause, and fixing them is the cheapest way to lower the figure.
Step 5: Watch the hidden multipliers
Several things quietly raise cost per conversation:
- Long conversations. If your platform charges per reply, a chat that takes ten replies costs more than one that takes two. Concise, accurate answers save money.
- Repeat contacts. A customer who returns because the first answer failed generates two conversations for one problem.
- Special features. Some features cost more than a standard reply. Check your platform's rules for anything billed at a multiple.
- Unused capacity. Paying for a larger plan than you use raises the effective cost of every conversation.
Step 6: Turn the number into decisions
A cost figure is only useful if it changes what you do. Review it monthly and ask:
- Is the AI-only share rising or falling?
- Which topics cause the most handoffs? Add or improve knowledge for those.
- Does any channel cost far more than the others? Investigate before expanding it.
- Would the next plan up lower the per-conversation cost at your volume, or just add unused capacity?
Plan comparisons are straightforward when you work in credits. For example, if you use around 3,500 replies a month, a plan with 4,000 included credits has headroom, while one with 500 would push you into paid recharges. Do the arithmetic with your own volumes rather than guessing.
Doing this in AI SNS CHAT
AI SNS CHAT bills in credits, and one reply uses one credit. That makes the platform side of the calculation direct: count the replies, and you know the credits used. A conversation's platform cost is roughly the number of replies in it, multiplied by the effective price of a credit on your plan.
Here is how to gather the inputs:
- Credits used. Open the Usage page to see credits per agent. Compare this against the credits included in your plan each month. Plans include 500 (Hobby), 4,000 (Standard), 15,000 (Pro), 40,000 (Corporate) and 16,000 (Private AI) credits, at the default prices shown on the pricing page. Credits reset each billing cycle, while credits bought through auto-recharge never expire.
- Conversation counts. Go to Activity → Chat logs, where conversations from every channel except email are listed. Email conversations are in the Inbox. Use filters and export to count conversations by channel and period.
- Human-handled share. In Chat logs you can see which conversations were escalated or taken over. Count those separately to get your handoff rate.
- Quality signals. Thumbs up and down ratings in Chat logs help you spot answers that cause repeat contacts. Analytics shows Chats, Topics, Question history and Sentiment, and advanced analytics require the Pro plan.
- Staff time. Helpdesk → Scheduling shows who is on shift to receive handoffs, which helps you estimate how many staff hours the human-handled conversations consume. The time itself you will need to estimate or take from your own records, since AI SNS CHAT does not provide a cost view for it.
Note that there is no customer-facing view of underlying AI provider costs, so your calculation is based on credits and plan price, not on the vendors' raw charges.
To lower the figure, the levers inside the product are practical ones. Add missing answers under Data sources (Files, Text snippets, Website, Q&A and more) so fewer chats need a person. Edit a reply in the Inbox or Chat logs into the ideal answer to train the agent. Use Playground to test changes before they reach customers. Details on all of these are on the features page.
Voice is a separate case: one voice minute uses 4 credits on the Private AI quote, and the credit rate per minute is set by the administrator, so check your own terms before including phone calls in the same calculation.
Conclusion
Cost per conversation is not one number but a small model: a clear definition, a complete list of costs, a split between AI-only and human-handled chats, and a breakdown by channel. Build it once with real figures, review it monthly, and let it tell you where to improve your knowledge base, which channels to grow and when a plan change is worth it.
Frequently asked questions
Should I include staff salaries in cost per conversation?
Yes, for any conversation a person handles. Use a loaded hourly cost and the minutes spent. Leaving it out makes human-handled chats look as cheap as AI-only ones and hides the real benefit of reducing handoffs.
How often should I recalculate it?
Monthly is a good rhythm for most teams. Recalculate sooner after a major change, such as adding a channel, rewriting your knowledge base or switching plans, so you can see its effect.
Is a lower cost per conversation always better?
Not necessarily. If you cut cost by making answers shorter or discouraging handoffs, customers may return with the same problem, which raises the cost per resolved issue. Track satisfaction signals such as ratings alongside cost.
How do credits translate into a cost per conversation in AI SNS CHAT?
One reply is one credit, so a conversation with eight AI replies uses eight credits. Divide your plan price by the credits you actually use in a month to estimate an effective price per credit, then multiply by the replies per conversation. Add staff time for any conversations that were handed to a person.


