Chatbot platforms fall into three groups that solve genuinely different problems, and the most common selection error is buying from the wrong group entirely. A customer service suite, a developer framework, and a self-hosted open-source platform are not competing products. This page sorts them by what they are for, covers the selection criteria that outlast any specific vendor, and is direct about where each stops fitting.
The Criteria That Decide Fit
Platform capability lists converge quickly and tell you little. These four questions separate the options meaningfully and remain relevant regardless of which vendor is currently ahead on features.
Where Does the Conversation Start?
A widget on your website, inside your product, on a messaging channel, or through voice. Platforms differ sharply in channel strength and this is frequently the deciding factor.
How Does It Access Your Knowledge?
Grounding responses in your own documentation is what determines accuracy. Retrieval quality matters more than which model sits behind it.
What Happens on Handover?
Escalation to a human is where most deployments actually succeed or fail. Our AI chatbot development work treats handover design as a primary requirement.
Where Does Conversation Data Go?
Transcripts frequently contain personal or sensitive information. Data residency and provider terms constrain selection more than capability comparisons do.
Customer Service Platforms
These are support suites with conversational capability added, and they suit organisations whose primary need is deflecting and routing support volume.
What This Category Covers
Ticketing, agent workspace, knowledge base, and conversational handling in one product, with the bot integrated into the same workflow agents already use.
The Established Options
Intercom, Zendesk, Freshworks, and Ada are the widely deployed examples, with the major CRM vendors offering comparable capability inside their own suites.
Why Integration Wins Here
The value is the bot and the agent sharing one system, so handover carries context and resolution data lands in the same place.
Where This Category Stops
Product-embedded assistants doing something other than support, and any requirement for unusual conversational logic, sit outside what these handle well.
Developer Frameworks and APIs
Where the assistant is part of your product rather than your support function, you build rather than buy, and this category provides the components.
What This Category Covers
Model APIs, retrieval infrastructure, orchestration frameworks, and conversation state management assembled into something specific to your product.
When Building Is Correct
When the assistant does something proprietary, needs deep product integration, or where conversation is a product feature rather than a support channel.
What You Take On
Evaluation, monitoring, cost management, and safety handling all become yours. Our generative AI development work covers that operational layer.
The Honest Threshold
If your requirement is deflecting support tickets, building is rarely justified. If the assistant is the product, buying rarely fits.
Open Source and Self-Hosted
Where data cannot leave your infrastructure, or where cost at volume makes hosted platforms unviable, self-hosting becomes the answer rather than a preference.
What This Category Covers
Platforms you run yourself, giving full control over data, model selection, and customisation, in exchange for operating them.
The Established Options
Rasa is the long-established framework for controlled conversational design, with Botpress and similar platforms offering more visual building.
Why Organisations Choose This
Regulatory constraints on data processing, cost at high conversation volume, or a requirement to use specific models including ones you host yourself.
The Operational Reality
Hosting, updates, model serving, and monitoring become your responsibility. Our cloud consulting work covers whether that capability exists before recommending it.
Deciding and Avoiding the Common Mistakes
Most disappointing chatbot deployments trace to a handful of avoidable decisions made before any platform was selected.
Ground It or Do Not Build It
A bot answering from general knowledge rather than your own documentation will be confidently wrong. Retrieval grounding is not optional.
Design Escalation as a Feature
Optimising for containment pushes the system toward answering things it should hand over. Escalation is a success path, not a failure.
Start With Your Highest-Volume Questions
The top handful of repeated questions usually represent most of the volume. Handling those well beats broad shallow coverage.
Measure Resolution, Not Deflection
Whether the customer’s problem was solved, not whether a human was avoided. Our AI consulting services engagements set that measurement before deployment.
Review the Page Quarterly
This category moves faster than any other in software. Treat any platform recommendation, including this one, as needing periodic re-checking.
FAQs
What is the best AI chatbot platform?
It depends on category. Intercom, Zendesk, Freshworks, and Ada for customer service. Model APIs with retrieval and orchestration frameworks when the assistant is part of your product. Rasa or Botpress where self-hosting is required.
Should I build or buy a chatbot?
Buy if the requirement is deflecting and routing support volume, since service suites handle that with agent handover built in. Build if the assistant is a product feature doing something proprietary or needing deep product integration.
What makes a chatbot accurate?
Grounding responses in your own documentation through retrieval rather than letting the model answer from general knowledge. Retrieval quality determines accuracy more than which model sits behind it.
Why do chatbot deployments disappoint?
Usually poor grounding, producing confidently wrong answers, or optimising for containment, which pushes the system toward answering things it should escalate. Both are decisions made before any platform is selected.
When is self-hosting a chatbot platform worth it?
When regulation prevents conversation data leaving your infrastructure, when volume makes hosted pricing unviable, or when you must use specific models. It requires capability to host, update, and monitor the platform.
What should I measure?
Resolution rate, meaning whether the customer’s problem was actually solved, plus escalation appropriateness in both directions. Deflection rate alone rewards a system that avoids handover when it should hand over.
