Website support chatbot
A chatbot on your website that answers customers from your own help docs, product pages and policies, says where the answer came from, and hands the rest to your team.
Customers try to help themselves, fail, and then cost you a human contact.
Human support is expensive
Live channels (phone, chat, email) cost about $8.01 per contact, against about $0.10 for self-service.
Gartner, 2019Old self-service rarely works
In the same survey, only 9% of customers fully solved their issue through self-service. FAQ pages and search bars make people hunt for the answer; most give up and email.
Gartner, 2019Generic AI chatbots are risky
A chatbot that makes up a refund policy or a price creates more tickets, and more angry customers, than it saves.
What the agent does
- 01
Reads only what you give it
Help articles, product specs, shipping and returns policies and pricing are added as Markdown documents or a searchable index. The agent finds the right page, answers from it, and says so when it cannot find an answer.
- 02
Sits on your website
An embed puts the chat on your site with two lines of code, with suggested first questions such as “Track my order” and “Return an item”. You change its label and colour without redeploying.
- 03
Can look things up
With a custom tool it can check an order status or an account detail, not just read docs.
- 04
Learns what customers ask
Every answer comes with a topic and keywords, so you can see which questions come up most and which docs are missing.
Checked before anyone sees an answer
The guardrail
Off-topic, abusive or risky questions, such as legal matters or a refund over your limit, are stopped by guardrails and answered with your own wording.
Where a person steps in
When a customer needs a person, a workflow creates the ticket in your helpdesk with the conversation and a summary already written.
What is recorded
Monitoring and Analytics show every conversation, what it cost, and which answers were blocked.
The steps, in order
This is the fastest Studio use case: no integrations are needed to go live with answers.
| Step | What you do in Studio | Typical time |
|---|---|---|
| 1Add your content | Create a knowledge base and add your top help articles and policies, starting with the 20 that cover most tickets. | 30–60 min |
| 2Create the agent | Give it a name, a goal and a model; attach the knowledge base; add starters. | 10 min |
| 3Add guardrails | Add “off-topic”, “refund over $X” and “abusive message” rules with your replies. | 10 min |
| 4Test it | Paste 20 real tickets into the Playground, and add a test suite so future changes cannot break good answers. | 30–45 min |
| 5Go live | Create an embed, set the label and accent colour, and paste two lines into your site. | 10 min |
| 6Later | Add the helpdesk hand-off workflow, an order-lookup tool and more content. | As you grow |
A worked example
An online business with 2,000 support contacts a month. The inputs are assumptions about a typical business; replace them with your own numbers.
| Line | Assumption | Per year |
|---|---|---|
| Contacts the agent resolves | 40% of 2,000 a month, which is 800 a month | 9,600 contacts |
| Human cost avoided | 9,600 contacts at $8.01 each | ≈ $76,900 |
| Team time freed | 800 a month × 6 minutes each, which is 80 hours a month | ≈ 960 hours |
| Total value | ≈ $76,900 | |
| Krazimo Studio cost | ≈ $1,500–$2,000 a year: the Starter plan at $99 a month (about 2,400 runs a month at 3 turns per chat) plus model usage | |
About 80 hours a month, which the team spends on the conversations that actually need a person, while customers get answers in seconds instead of hours.
See it working
on your own information.
Tell us about your business and we will walk through this use case with you in Studio.