Two AI support products can advertise very different prices while producing almost the same bill. One charges for users and adds AI usage. Another charges for conversations. A third links AI charges to completed outcomes. The headline numbers are not comparable until you define the workload and the billable unit.

Start with the work you expect the system to do. Then ask every provider to price that same scenario. This avoids choosing a tool on an entry plan that excludes your channels or on an optimistic resolution estimate that has never been tested.
1. Understand what starts the meter


	

		

			
Billing unit
			
What you need to understand
			
A question to ask the provider
		
		

			
Seat
			
Who needs a paid user account and what role they receive
			
Do supervisors, reviewers and occasional agents require full seats?
		
		

			
Message
			
Which incoming, outgoing or generated messages count
			
Does an unsuccessful answer or repeated clarification still consume allowance?
		
		

			
Conversation or ticket
			
How messages are grouped into a chargeable interaction
			
When does a returning buyer start another billable conversation?
		
		

			
Active contact
			
How a person becomes active in the billing period
			
Do outbound messages or multiple channels change the count?
		
		

			
Resolution or outcome
			
The conditions under which completed work is charged
			
What happens after human intervention, a wrong answer or a reopened issue?
		
	



These units describe commercial models, not quality guarantees. A per-resolution model still needs an understandable resolution definition. A per-message model can be economical when messages are short and accurate, but less attractive when an assistant creates several unnecessary turns.

Check whether more than one unit applies. A subscription may cover the inbox while AI, messaging channels or extra stores are billed separately. “Included AI” can refer to a limited allowance rather than unlimited use.

2. Build a complete cost sheet
List the recurring subscription, required users, store connections, AI charges, external channel fees, overages and necessary add-ons. Keep one-time onboarding or migration costs separate, then decide whether to spread them across a planning period.

Do not assume an annual price is available month to month. Record the billing commitment and currency. If a Singapore seller receives a US-dollar quotation, currency conversion and applicable tax can affect the cash paid. Use the actual invoice terms rather than an exchange-rate screenshot copied from another article.

Official pricing pages are useful starting points. For example, eDesk, Tidio and SleekFlow publish different plan structures and allowances. Check their current pages and obtain a quotation for your configuration; the presence of a published entry price does not establish the total cost of your intended setup.

3. Send every vendor the same workload
Use a recent representative month, not only your busiest sale day. Count users, stores and eligible conversations. Separately estimate routine enquiries, cases needing human help and the messages or active contacts relevant to a provider's billing rules.

Here is an illustrative request, not a claim about typical sellers:


	

		

			
Input
			
Hypothetical workload
		
		

			
Connected stores
			
Four stores across supported marketplaces
		
		

			
Staff access
			
Three agents and one supervisor
		
		

			
Eligible service conversations
			
5,000 in a month
		
		

			
Independently completed by AI
			
Test scenarios at 1,000, 2,000 and 3,000
		
		

			
Human-assisted conversations
			
The remaining eligible conversations
		
		

			
External messaging
			
Quoted separately for any chargeable channel use
		
	



Ask for all three resolution scenarios. A quote that looks affordable only at the highest assumed automation level is risky before a pilot has established what your knowledge and permissions can support.

Include a reopening example: the buyer returns about the same unresolved problem, and a person finishes the work. Ask the vendor to explain the charge in writing. The explanation is more useful than a sales statement that the plan is “fair”.

4. Calculate effective cost rather than comparing unit prices
For an internal planning measure, divide total monthly service-system cost by independently completed conversations that pass your quality definition. Keep human service cost in a separate view if you want to evaluate the full operation.

Suppose a hypothetical system costs $240 for the month and produces 1,600 independently completed conversations. Its system cost per independent completion is $0.15. If only 800 pass review, the same bill becomes $0.30 per completion. Neither figure is a vendor quotation or a forecast for Duoke.

The denominator matters. Do not count greetings, unanswered requests for information or incorrectly closed cases as successful work. Otherwise, your spreadsheet rewards a system for closing conversations that customers still need help with.

Also inspect total cost per eligible incoming conversation. That second measure prevents a high completion rate on a tiny, cherry-picked scope from hiding the cost of the larger service operation.

5. Understand Duoke's outcome-based distinction
Duoke's current product positioning links AI conversation credits to qualifying independent resolution. The important distinction is between AI completing the buyer's valid service need and merely sending a reply. Human substantive intervention and non-qualifying outcomes need to be treated according to the applicable plan and resolution rules.

Do not interpret “no resolution, no charge” as meaning every other service entitlement is necessarily free. Check subscription, credits and channel-related charges separately. Use the current English quotation for the applicable market rather than converting a Chinese price informally.

Ask for examples involving an incorrect answer, a partial answer, a necessary human approval and the same issue reopening. A transparent billing model should make those cases understandable.

6. Decide what the pilot must prove
Before paying for a longer commitment, review answer quality, the proportion of eligible work completed independently, human rework and the invoice generated by the pilot. Keep the workload and exclusion rules visible.

If the team must spend substantial time repairing wrong answers, include that work in the decision. If a reliable assistant covers a smaller scope than expected, calculate its value at that smaller scope. The useful price is the one attached to dependable work your business actually needs.