Compare chatbot prices by calculating the bill for the same workload. Record the subscription, included usage, extra usage, seats and required add-ons. Calculate a normal month and a busy month. Then check whether the service stops, buys more usage or changes plans when you reach the allowance.
The examples below use invented quotes to explain the calculation. They are not current vendor prices or customer results.
Find out what triggers a charge
| Pricing unit | What to establish before calculating |
|---|---|
| Message | Does the vendor count visitor messages, AI replies, greetings, failed replies or tool calls? |
| Credit | How many credits does each model reply or action consume? |
| Conversation | What starts a session, what ends it, and does a returning visitor start another? |
| Outcome | Which results are billable, and what happens if the customer returns for help? |
| Seat | Which staff need a seat, and does AI usage cost extra? |
| Subscription tier | Which features and allowances does the base price include? |
Several units can appear on one invoice. A seat-based helpdesk can also charge for AI outcomes. A credit plan can sell extra packs. Request the usage definition and a sample export showing which events produced the charges.
For example, 1,000 conversations with four AI replies each means 4,000 replies. It does not establish the number of billable messages or credits. That depends on the contract and model mix. If you do not know the conversion rate, leave it marked unknown.
Use one workload for every quote
Copy this record for each vendor. Use measured usage when available and label estimates.
| Input | Normal month | Busy month |
|---|---|---|
| Conversations | ___ | ___ |
| AI replies per conversation | ___ | ___ |
| Credits per reply or action | ___ | ___ |
| Billable outcomes by type | ___ | ___ |
| Required staff seats | ___ | ___ |
| Required bots, channels and integrations | ___ | ___ |
Record the base price, billing term, included units, extra-unit price or pack size, required add-ons and taxes. Check whether unused usage expires. For an annual offer, record both the monthly equivalent and the cash commitment. The capacity-planning method can help define the busy month.
Remove any quote that lacks a required feature before comparing its price. A plan without production API access cannot satisfy a requirement for that API, even if its usage allowance is large enough.
Worked example: the cheaper quote changes
Suppose two invented quotes meet the same feature and seat requirements:
- Quote A costs $99 per month, includes 1,000 conversations and charges $0.08 for each additional conversation.
- Quote B costs $60 per month, includes 5,000 credits and sells additional credits in packs of 1,000 for $12.
Assume each conversation produces four AI replies, each costing two credits under Quote B. This rate is an input for the example, not a general rule.
| Calculation | Normal month | Busy month |
|---|---|---|
| Conversations | 1,000 | 2,500 |
| Quote B credits: conversations × 4 × 2 | 8,000 | 20,000 |
| Quote A bill | $99 | $99 + 1,500 × $0.08 = $219 |
| Quote B extra packs | 3 | 15 |
| Quote B bill | $60 + 3 × $12 = $96 | $60 + 15 × $12 = $240 |
Quote B is cheaper at exactly 1,000 conversations. Quote A is cheaper at 2,500.
Packs round up to whole units. At 1,000 conversations, Quote B needs (8,000 − 5,000) ÷ 1,000 = 3 extra packs. One more conversation needs a fourth pack, raising Quote B to $108 while Quote A costs $99.08. Check volumes near pack boundaries as well as your normal and busy months.
The example assumes packs are bought only to cover usage beyond the included allowance. It excludes taxes, discounts, minimum commitments and add-ons. An automatic recharge threshold or prepaid balance can change when packs are purchased. Ask the vendor to calculate the bill under its actual purchase rules. The minimum packs needed to cover usage may not match the invoice.
If the model mix changes, recalculate the credits per reply. If your volume is seasonal, calculate the expected months separately. Multiplying one peak month by twelve does not describe a seasonal year.
Check outcome charges against the definition
An outcome is a vendor-defined event. A customer reaching a person, a qualified lead and a resolved question may have different billing treatment. Ask for the definition, the charge-level export and any reversal rule when a customer returns.
For an invented outcome quote with an $80 base fee and $0.50 per billable resolution, 300 qualifying resolutions cost $80 + 300 × $0.50 = $230, before other charges. You cannot compare that bill with the conversation example without establishing how many conversations meet the outcome definition.
Use a sample of charged conversations to check the definition against what happened. The chatbot KPI guide explains how to separate containment from a verified customer outcome.
Check what happens at the allowance
Ask the supplier to show the limit behavior. Does the assistant stop answering, buy another pack, bill overage or require an upgrade? Who receives the warning? Can an administrator control the extra spend? A low monthly price is incomplete without those answers.
Agentkit's usage guide defines one chat credit as one saved AI reply. A visitor message alone does not spend a credit. Image and video costs depend on the model and settings and draw from the same workspace balance. Included monthly credits reset; purchased credits carry over until spent. At zero balance, the chatbot cannot produce another billable AI reply and visitors see an unavailable message. Paid plans can buy credit top-ups. Use current pricing to check allowances, top-ups and required features. Another vendor's credit definition may differ.
Finish with a worked quote for the same normal and busy months from each finalist. Add the staff and setup costs covered in the implementation-cost guide. Keep missing terms visible until the supplier answers them.



