Who wrote this, and what is our interest?
SpiderHunts wrote this article. We build custom conversational AI systems for UK businesses, so we have an interest in the question of when a platform is enough and when a build is worth it. We have tried to be fair: no vendor paid to be included or reviewed this article, and we say plainly where a platform is the better choice. Product details and prices were checked on each vendor's own website and are correct as of October 2026. They change often, so check the linked pages before you decide.
How do the main platforms compare?
The platforms fall into four groups: cloud developer platforms, low-code builders, an open-source framework and helpdesk add-ons. The right group usually matters more than the right product within it.
| Platform | Type | Best suited to | Pricing model (as of October 2026) |
|---|---|---|---|
| Google Dialogflow CX | Cloud developer platform | Teams on Google Cloud, text and voice | Pay per request or audio minute |
| Microsoft Copilot Studio | Low-code builder | Microsoft 365 and Teams organisations | Credit packs or pay-as-you-go |
| Amazon Lex | Cloud developer platform | Teams on AWS, often with Amazon Connect | Pay per text or speech request |
| Rasa | Code-first framework | Teams that must self-host or own the stack | Free developer edition, enterprise on request |
| Voiceflow | Visual builder | CX teams designing agents together | On request |
| Botpress | Visual builder plus code | Smaller teams that want to start quickly | Free tier and monthly plans, AI usage extra |
| Intercom Fin | Helpdesk AI agent | Support teams on Intercom or another helpdesk | Per outcome |
| Zendesk AI agents | Helpdesk AI agent | Support teams already on Zendesk | Included in plans, with resolution allowances |
What are the cloud developer platforms like?
Google Dialogflow CX. Google's platform for building text and voice agents from structured flows and, in its newer Conversational Agents console, generative playbooks. It runs on Google Cloud only, and London (europe-west2) is listed as an agent region. Google's pricing documentation lists flow-based agents at $0.007 per text request and $0.06 per audio minute as of October 2026; generative features are priced separately. It suits teams with developers who are comfortable on Google Cloud.
Amazon Lex. AWS's service for building conversational interfaces with voice and text, built in the console or with SDKs, and commonly paired with Amazon Connect for contact centres. Optional generative features run on Amazon Bedrock, which is billed separately. AWS's Lex pricing page lists $0.004 per speech request and $0.00075 per text request as of October 2026. Lex is available in the London region.
Both are powerful and inexpensive per request, but they are developer tools. Someone still has to design the conversation, connect your systems, write the content and test it.
What are the low-code builders like?
Microsoft Copilot Studio. Microsoft describes it as a graphical, low-code studio for building AI agents, which can be published to Teams, Microsoft 365 Copilot, websites and other channels. Its natural home is an organisation already on Microsoft 365, especially for internal assistants. Microsoft's UK pricing page lists £153.80 per pack per month for 25,000 Copilot Credits, with a pay-as-you-go option, as of October 2026. The UK is a documented data location.
Voiceflow. A collaborative, visual platform for designing and running AI agents, with code steps when you need them, across web chat, WhatsApp, SMS and voice. It is aimed at customer experience teams. Pricing is on request for businesses as of October 2026.
Botpress. A visual drag-and-drop builder with a code toolkit for developers. Its pricing page shows a free plan and paid plans from $150 a month billed annually as of October 2026, with AI usage, meaning the cost of calling the language model, billed on top.
When does an open-source framework make sense?
Rasa is a code-first framework for building conversational agents. Its current approach, called CALM, uses a language model to interpret what people want while structured flows enforce the business logic, which is a sensible split for regulated work. Rasa can be self-managed on your own infrastructure or private cloud, which matters when data must stay inside your environment. Note that the original Rasa Open Source project is described by Rasa as being in maintenance mode, and current development is in Rasa Pro, a licensed product with a free developer edition. You need engineers to run it.
Are helpdesk AI agents good enough on their own?
If your support team already works in a helpdesk, its AI agent is often the fastest route.
Intercom Fin is an AI agent for chat, email, voice and other channels that works on Intercom or alongside some other helpdesks. Intercom's Fin pricing page lists $0.99 per outcome as of October 2026, with a monthly minimum when used with another helpdesk, and states that Intercom bills in US dollars only. Data hosting is offered in the US, EU or Australia.
Zendesk AI agents answer across messaging, email and voice using generative AI and Zendesk's own intent models. Zendesk says they are included in every Suite and Support plan, with an allowance of automated resolutions and the option to buy more.
Both are strong at answering from your help centre. They are weaker when the answer depends on data that lives outside the helpdesk, such as a bespoke booking system or an internal database, or when the conversation happens on the phone or WhatsApp rather than in support tickets.
When is a platform enough, and when is a custom build worth it?
A platform is usually enough when:
- Most questions can be answered from a help centre or a set of documents.
- You use one main channel, and the platform supports it well.
- You have someone in-house who will own the configuration and keep the content current.
- Per-conversation or per-outcome pricing works out sensibly at your volumes.
A custom build starts to earn its cost when:
- Answers depend on live data in your own systems, such as stock, bookings, cases or accounts.
- The assistant must act, not just answer: book, rebook, update the CRM, raise a job.
- You need several channels, such as web, WhatsApp and phone, from one knowledge base.
- You have specific data requirements, such as UK processing or redaction before data reaches a model.
- Per-resolution pricing would become expensive at your volumes, and you would rather own the system.
This is where SpiderHunts fits. We are not a platform; we design and build custom AI chatbots, AI voice agents and AI agents, built on whichever platform or model API suits the job. As the authors of this article, we would rather tell you a platform is enough than sell you a build you do not need. For the decision process in more depth, see how to choose an AI chatbot platform, and for budgets, what an AI chatbot costs in the UK.
How should you trial a platform before committing?
- Collect 50 to 100 real questions from your inbox, chat logs or call notes.
- Load your actual content, not a demo knowledge base.
- Ask the real questions, including awkward ones and ones it should refuse.
- Count how many it answers correctly, how many it admits it cannot, and how many it gets wrong.
- Check the handover: does a person receive the full context?
- Work out the monthly cost at your real volumes, including model usage and channel fees.
A wrong answer delivered confidently is worse than no answer, so weigh that third number most heavily.