A chatbot price is useful only when you know what triggers a charge. One vendor counts AI replies, another counts credits spent on those replies. Convert each offer into a bill for the same normal month and the same busy month. Include the base subscription, usage beyond the allowance, required add-ons and any minimum commitment. Then check what happens when the allowance runs out.
Prices and billing rules cited below were checked October 18, 2026. Confirm current terms before purchasing.
Copy this worksheet for each vendor. Use your own usage report if you have one; otherwise mark every volume estimate as an assumption.
| Item to record | Your input |
|---|---|
| Monthly conversations | ___ |
| AI replies per conversation | ___ |
| Credits per reply, if the model or action changes the credit cost | ___ |
| Billable outcomes by type | ___ |
| Human seats | ___ |
| Base price and billing term | ___ |
| Included units | ___ |
| Extra-unit price, pack size or next-tier price | ___ |
| Required add-ons, taxes and minimum commitment | ___ |
| What happens at the limit; downgrade date | ___ |
The capacity-planning method can help estimate a busy month.
What each meter actually counts
| Meter | Billable unit | Check before comparing |
|---|---|---|
| Message | A vendor-defined message, often an AI reply | Are greetings, retries, tool calls or failed replies included? |
| Credit | Credits consumed by a model reply or action | How many credits does each model and action use? |
| Conversation | A session under the vendor's start and end rules | When does a new session begin? |
| Outcome | A vendor-defined result in a conversation | Which outcomes cost money, and can a later return reverse a charge? |
| Seat | A human agent or workspace member | Does AI usage cost extra on top of the seats? |
| Subscription tier | A base price with allowances and feature limits | What changes at the ceiling: service, price or plan? |
These meters often stack. A seat-based helpdesk can also bill AI outcomes. A monthly credit plan can sell recharge packs. A tiered subscription is not an unlimited flat rate when it has a message ceiling. Read the order form and usage definition together.
Messages, credits and conversations
Suppose 1,000 visitors each have a conversation with four AI replies. That is 1,000 conversations and 4,000 replies. It is not necessarily 4,000 billable messages or credits. A vendor might count a greeting, exclude a failed reply, or charge multiple credits for a stronger model or an action. Ask for a sample usage export that ties the meter to individual conversations.
Conversation pricing removes reply count from the headline unit, but session boundaries matter. A visitor who returns tomorrow may start another billable session. Check the session boundary, included allowance and channel rules.
Chatbase provides a concrete credit example. Its pricing page lists monthly Hobby at $40 for 700 credits, Standard at $150 for 4,000, and Pro at $500 for 15,000. Optional 1,000-credit recharge packs cost $40. Hobby, Standard and Pro include two, three and five workspace seats; API access starts at Standard. The page opens on the annual toggle; the figures above are its monthly list prices. For a team already using Chatbase, take a complete month's measured credits from its usage screen instead of assuming one credit per reply.
At 10,000 measured credits with no prepaid balance or add-ons, Hobby needs 10 packs and costs $40 + 10 × $40 = $440. Standard needs six packs and costs $150 + 6 × $40 = $390. Pro costs $500 and leaves 5,000 included credits unused. This illustrative minimum makes Standard the lowest cash option for that workload. Hobby has no API access, so it would be ineligible if the team needs the API.
The Chatbase credit planner compares the three paid plans using measured credits or an estimate based on conversations, replies and credits per reply. It checks seats and API access, then adds minimum recharge packs and selected add-ons. Enter a normal month and a busy month. Its calculation source is public.
The planner is a minimum volume-coverage comparison, not an invoice forecast. It assumes subscription credits are spent before a prepaid recharge balance; that ordering has not been independently verified. The planner ignores when an auto-recharge threshold fires, so a real month may buy a pack earlier or later than this minimum. It excludes tax, discounts, midmonth changes, annual commitments, voice and Enterprise terms. Agentkit sells a competing chatbot product; the planner compares Chatbase plans only and makes no claim about answer quality or provider savings.
Outcomes and resolutions
"Pay for results" sounds clear until you read the result definition. Intercom's Fin outcome rules list a $0.99 resolution and other billable outcomes. A configured procedure handoff, disqualification and self-serve routing can each cost $0.99; a qualification can cost $9.99. Intercom says these prices apply to Fin over chat and email when used with Intercom. Voice, high-volume and specialized terms require a separate quote. Intercom charges at most one outcome per conversation. It also says the assumed-resolution charge is reversed if the customer returns to the same conversation for more help. A configured procedure handoff can be billable even though it routes the customer to a human.
Zendesk uses a different rule. Its automated-resolution tiers distinguish assisted escalation and contained resolution from verified resolution. The first two do not consume the resolution allowance; verified resolution does. Its pricing page lists a seat-based subscription and additional automated resolutions; the allowance and extra-unit rate vary by plan. You cannot apply one vendor's definition to the other's bill.
Ask each vendor for a charge-level export and its repeat-contact rule. Compare a sample of claimed resolutions with your own quality review. A quiet customer is not proof that the answer solved the problem. The chatbot KPI guide shows how to separate containment from verified success.
Here is an illustrative Fin calculation using its October 18 published rates, not an observed customer invoice. Suppose one month has 300 resolutions, 40 configured procedure handoffs and 10 qualifications, all in different conversations. Outcome charges are 300 × $0.99 + 40 × $0.99 + 10 × $9.99 = $436.50. The full bill is $436.50 plus the applicable base subscription, seats, add-ons and taxes. If you price only the 300 resolutions, you miss $139.50 in this example. Your contract, channels and actual outcome mix decide the real amount.
Seats and tiers
Seat fees track the people who work in the inbox or administer the system. They can be sensible when human agents handle many cases, but they do not tell you what AI automation costs. For a hybrid offer, write two lines: seat count × seat price and billable AI units × rate, after subtracting any included allowance. Add both lines before comparing it with a website-bot subscription.
A tiered subscription buys a stated allowance and feature set. Agentkit's current pricing is one example: at monthly billing, Hobby is $29.99 for 2,000 messages, one chatbot and one seat; Standard is $119.99 for 12,000 messages, two chatbots and three seats; Pro is $399.99 for 40,000 messages, three chatbots and five seats. Agentkit's usage guide describes its unit as a visitor message, with one allowance shared across chatbots in a workspace. Chatbase credits and Agentkit messages are different units, so neither the 10,000-credit example nor its price can be compared with Agentkit's 12,000-message tier without a measured credits-per-reply rate and the same visitor workload. If you need two chatbots, Hobby is ineligible. At the message limit, the website agent shows visitors a limit message and cannot answer again until the billing cycle resets or the owner upgrades. Check the pricing page for model access and integrations; do not assume unused allowance carries forward.
A tier is predictable only within its limits. For other vendors, ask whether the bot stops, the account buys credits, the vendor bills overages, or the plan changes when usage reaches the ceiling. For seasonal traffic, compare twelve realistic monthly bills, not one peak month multiplied by twelve.
Compare two offers with the same workload
Use one set of inputs for every finalist: monthly conversations, replies per conversation, model mix, required integrations, staff seats and a busy-month multiplier. Request a worked bill from each vendor with these inputs. Mark missing answers as unknown, not zero.
A useful comparison line is:
monthly total = base subscription + required seats + billable usage beyond allowances + required add-ons + taxes
For credits sold only in packs, round the shortage up to a whole pack before multiplying by the pack price. For outcomes, calculate each billable outcome type separately. For annual offers, show both the monthly equivalent and the actual cash commitment; an annual discount is not a monthly cancellation option.
Then ask what the bill omits. Does one quote include human support, training-source refresh, export, extra bots and production API access while the other charges for them? Can you audit individual units against a usage report? The implementation-cost guide covers staff and setup work beyond the software bill. The chatbot cost guide covers the broader operating budget.
Run the worksheet for your normal and busy months. Use the Chatbase planner if you need to compare its credit packs, or Agentkit's pricing page for current message allowances and features. Then ask each finalist for a quote using those inputs. If a vendor cannot show which events produce each line, hold the purchase until it can.
