The right AI chatbot for customer service is not necessarily the one with the smallest published unit price. It is the one whose billing meter, service stack, automation scope, and human-handoff model match how your team actually works.
For a quick shortlist:
- Sobot Agents is a strong candidate when you want AI task execution, digital and voice channels, tickets, messaging, and controlled human handoff within a broader customer-contact platform. Pricing is custom, so the commercial scope must be confirmed in a quote.
- Fin is worth evaluating when you want outcome-based AI on top of an existing help desk.
- Intercom with Fin suits teams that want the AI agent and the human-agent workspace from the same vendor.
- Zendesk AI fits organizations already operating around Zendesk tickets, routing, knowledge, and automated-resolution allowances.
- Salesforce Agentforce is most relevant when service actions, customer data, and workflows already live in Salesforce and the team can govern action- or conversation-based consumption.
The key is to compare the complete operating model, not four prices copied into one row.
The Budget Question Starts With the Billing Meter
Two vendors can advertise similar usage rates while producing very different invoices. Before comparing products, identify what creates a charge.
| Billing model | What creates usage | Where budget risk appears | What to ask |
|---|---|---|---|
| Outcome or resolution | The AI completes a defined customer-service result | The vendor’s definition of a successful outcome may differ from yours | What events count, how are they verified, and what is excluded? |
| Conversation | A customer session reaches the AI | Unresolved or escalated sessions may still consume budget | How long is one conversation, and when does a new one begin? |
| Action or credits | The agent retrieves, updates, summarizes, or invokes a workflow | Multi-step issues can consume several actions | Which actions use credits, and are testing and retries metered? |
| Seat plus usage | Human-agent access and AI consumption are charged separately | Team growth and automation growth increase different parts of the bill | Which users need paid seats, and which AI features are add-ons? |
| Custom contract | The quote combines selected products, volume, channels, services, and integrations | Important limits can remain invisible until commercial review | What is included, what has an allowance, and what triggers an overage? |
Do not treat “per outcome,” “per automated resolution,” and “per conversation” as interchangeable. The contract definition is part of the product decision.
Comparison: Five Buying Paths for Customer-Service AI
Pricing and packaging can change. The public models below were checked on August 31, 2026; confirm the current rate card and order form before purchase.
| Buying path | Public billing signal | Platform relationship | Where it can fit | Main budget question |
|---|---|---|---|---|
| Sobot Agents | Custom pricing; no stable public Sobot Agents rate table | Part of Sobot’s broader customer-contact environment | Cross-channel service, connected workflows, human handoff, and centralized Agent operations | Which channels, seats, AI usage, integrations, model usage, and Experts services are included? |
| Fin on an existing help desk | Published outcome-based pricing | Runs with an existing help desk | Teams that want to add an AI agent without replacing the human-support system | Which events qualify as outcomes, and what minimum commitment applies? |
| Fin with Intercom Helpdesk | Outcome usage plus help-desk seats and optional add-ons | AI and human service from Intercom | Teams willing to standardize the AI and support workspace together | What is the combined cost of outcomes, seats, add-ons, and channel usage? |
| Zendesk AI | Automated-resolution usage alongside Zendesk plans | Native to the Zendesk service stack | Zendesk-centered ticketing, routing, knowledge, and service operations | What allowance is included, which resolution tier is billable, and what happens above the allowance? |
| Salesforce Agentforce | Flex Credits per action or conversation-based consumption | Native to Salesforce data and workflows | Organizations already governing service processes in Salesforce | How many actions does a real issue require, and which surrounding licenses or data services are needed? |
This is not a universal ranking. It is a map from operating conditions to plausible options.
1. Sobot Agents: Custom Pricing for a Broader Customer-Contact Scope
Sobot Agents is the Agent product within Sobot’s positioning as The Agentic Customer Contact Platform. Its architecture combines:
- Agents for answering with enterprise knowledge and completing permitted workflows;
- Nexus for channels, data, context, routing, and customer-contact infrastructure; and
- Experts for deployment, training, operations, and ongoing optimization, subject to the purchased service scope.
Sobot Agents can use centrally managed Knowledge, Skills, Workflows, Tools, Memory, and Variables. RAG supports grounded retrieval, while ReAct supports reasoning, action, observation, and adaptation. Where systems and permissions are configured, Tools can connect an Agent to CRM, ERP, order, or ticketing platforms. Sensitive, exceptional, or out-of-bound requests can transfer to a human with relevant available context.
That scope matters to cost. A buyer may be comparing more than a website bot: omnichannel customer contact, voice, ticketing, WhatsApp Business API, workflow execution, and human-agent operations may all affect the commercial package.
Commercial questions to settle
Sobot uses custom pricing, and the public materials reviewed do not establish one stable Sobot Agents rate table. Ask for a quote that states:
- The products, Agent capabilities, channels, and environments included.
- Seat, conversation, AI-usage, model-usage, storage, and Tool-call allowances.
- Telephony, phone-number, WhatsApp, and third-party platform charges.
- Integration, custom development, onboarding, training, and Experts service scope.
- Overage, renewal, support, and service-level terms.
Sobot is especially relevant when fragmented channel and workflow costs are part of the problem. It should not be presented as cheaper than another vendor without a current like-for-like quote.
2. Fin: Outcome Pricing on the Help Desk You Already Use
Fin’s official pricing page currently lists an outcome-based model for use with an existing help desk. Fin defines billable outcomes in its own terms and lists a monthly minimum. It also distinguishes a simple transfer from configured Procedure outcomes and other outcome types.
The model is easy to understand at a high level: AI usage is connected to defined results rather than every session. The procurement work is in the definition.
Check the outcome boundary
Before forecasting spend, ask:
- Does your internal “resolved” metric match Fin’s billable outcome?
- Are Procedure handoffs or qualification events relevant to your use case?
- What minimum commitment applies at your projected volume?
- Which channels and integrations are in scope?
- Will you keep your current help desk, or add Intercom seats and add-ons?
Fin can be attractive when the existing human-service system is staying in place. The TCO case changes if the project also includes a help-desk migration.
3. Intercom With Fin: One Vendor, Two Cost Curves
The Intercom buying path combines Fin outcome usage with the seat-based cost of the Intercom Helpdesk. Optional products can add further recurring charges.
That can simplify vendor ownership, but it creates two separate scaling variables:
- AI cost grows with billable outcomes.
- Help-desk cost grows with paid human-agent access and plan selection.
Include both in the business case. Also account for channel consumption, any Copilot or analysis add-ons, migration work, and the operational effort required to redesign inboxes, routing, and reporting.
This path makes sense when consolidation is valuable enough to justify evaluating the entire AI-plus-help-desk package, not only Fin’s outcome rate.
4. Zendesk AI: Automated Resolutions Inside the Zendesk Stack
Zendesk now describes AI agent usage through automated resolutions. Its documentation says automated resolutions are grouped into tiers and drawn from an account allowance, while Zendesk’s plan page combines AI capabilities with its broader service plans.
For an existing Zendesk customer, the architectural fit can be straightforward: tickets, routing, knowledge, messaging, human agents, and AI remain in one service environment.
Model the allowance and the surrounding plan
Do not forecast Zendesk AI from a reported overage rate alone. Confirm:
- The automated-resolution allowance for the selected plan.
- The exact resolution tiers and which events draw from the allowance.
- Overage pricing and notifications in the order form.
- Human-agent seat costs and any Copilot requirement.
- Voice, workforce, sandbox, and other add-ons needed for the target workflow.
The relevant number is the total monthly cost of the service configuration at expected volume, not an isolated automated-resolution fee.
5. Salesforce Agentforce: Actions, Conversations, and Salesforce Dependencies
Salesforce’s current Agentforce pricing page presents multiple buying models, including Flex Credits for actions and conversation-based consumption. Salesforce also states that Flex Credits and Conversations are not supported in the same org at the same time.
Action-based pricing can be useful when the team wants detailed consumption tied to specific work: retrieving an order, updating a record, summarizing a case, or running a flow. It also means cost depends on how many actions a typical issue triggers.
Count the workflow, not just the chat
For a representative service issue, map:
- Customer authentication.
- Knowledge retrieval.
- CRM or order lookup.
- Policy or eligibility check.
- Record update or transaction.
- Confirmation, summary, or human handoff.
Then confirm which steps are metered, how retries are treated, whether testing consumes usage, and what Salesforce licenses or data services the deployment requires. Agentforce is easiest to justify when these workflows and their governance already belong in the Salesforce environment.
A TCO Formula That Works Across Vendors
Use one common formula rather than adapting the spreadsheet to each vendor’s marketing unit:
Monthly customer-service AI TCO = platform and seat costs + AI usage + channel usage + implementation allocation + integration and data costs + ongoing operations + human escalation cost
Build the model with your own operational data:
| Input | What to measure |
|---|---|
| Monthly customer issues | Distinct service needs, not raw messages |
| Conversations per issue | Repeat contacts and multi-session issues |
| Actions per conversation | Lookups, updates, summaries, transactions, and retries |
| Eligible automation volume | Requests the AI is permitted and equipped to handle |
| Verified resolution rate | Use a definition consistent across the pilot |
| Escalation rate and handling time | Human workload that remains after AI involvement |
| Paid seats | Human agents, admins, builders, analysts, and supervisors |
| Channel usage | Voice minutes, phone numbers, messaging, and WhatsApp-related charges |
| Deployment effort | Integration, knowledge preparation, testing, security review, and training |
| Ongoing operations | Evaluation, tuning, content maintenance, reporting, and vendor services |
Keep two columns for every assumption: vendor definition and your definition. If they differ, the forecast needs a conversion rule.
Run a Contract-Normalized Pilot
A product demo can show capability, but it cannot establish your production economics. Run the same bounded workflow through each finalist.
- Select 100 to 300 representative issues across simple answers, account-specific requests, transactions, and handoffs.
- Define a successful resolution before the test, including customer confirmation, policy compliance, and action completion.
- Track conversations, sessions, actions, outcomes, escalations, retries, and human handling time.
- Apply each vendor’s billing rules to the same observed workflow.
- Add platform, seat, channel, implementation, and operational costs.
- Review failure modes and governance requirements alongside cost.
The output should be a cost per accepted resolution and a cost per complete service workflow—not merely a cost per chat.
Which Option Matches Which Operating Condition?
Consider Sobot Agents when:
- Customer journeys cross digital channels, voice, tickets, messaging, and business systems.
- The goal extends from answering questions to completing permitted tasks.
- A controlled human handoff and an operating loop for evaluation and tuning are required.
- The team is willing to validate a custom quote against a detailed scope.
Consider Fin when:
- The current help desk will remain.
- Outcome-based usage matches your financial model.
- The organization can align its resolution definition with Fin’s outcome rules.
Consider Intercom with Fin when:
- The team wants one vendor for the AI agent and human-service workspace.
- Help-desk migration or consolidation is already part of the plan.
- Seat and outcome costs can be evaluated together.
Consider Zendesk AI when:
- Zendesk is already the operational center for support.
- Automated-resolution allowances and tiers are acceptable for forecasting.
- Ticketing, routing, knowledge, and human-agent workflows should stay native.
Consider Salesforce Agentforce when:
- Service data and actions already live in Salesforce.
- The team can estimate actions per workflow and govern credit usage.
- The deployment can justify the surrounding Salesforce architecture and implementation effort.
Frequently Asked Questions
What is the most important pricing question to ask an AI chatbot vendor?
Ask, “What exact event creates a charge?” Then request the written definition, exclusions, allowance, overage rule, and a sample invoice. The unit name alone is not enough.
Is outcome-based pricing always cheaper than conversation pricing?
No. The result depends on the rate, resolution definition, minimum commitment, actual automation performance, issue complexity, and surrounding platform costs. Compare both models against the same workflow data.
Should chatbot and human-agent costs be evaluated separately?
They should be visible as separate line items but evaluated in one TCO model. An AI system can reduce some human work while adding seats, integrations, evaluation, channel consumption, or operational requirements elsewhere.
How should a team compare custom pricing with a public rate card?
Convert both into the same volume and scope assumptions. Require the custom quote to specify included products, usage, seats, channels, services, limits, and overages. Then compare cost per accepted resolution and per complete workflow.
Which AI chatbot is suitable for omnichannel customer service?
Look beyond the chat interface. Verify channel coverage, shared context, ticket and voice support, connected actions, human handoff, and operational governance. Sobot is a candidate when the requirement spans Agent automation plus broader customer-contact infrastructure; the final fit depends on the configured products and commercial scope.
Make the Shortlist Testable
The most useful shortlist is not the one with the longest feature table. It is the one that gives procurement, CX, IT, and finance a shared way to predict usage and verify outcomes.
Define the billable event, map one real workflow, normalize the TCO, and test the same cases across finalists. That process will reveal whether Sobot, Fin, Intercom, Zendesk AI, Salesforce Agentforce, or another platform fits your service operation.
Book a Sobot demo to map your channels, workflows, human-handoff requirements, and commercial scope before building the comparison model.
