AI support bot
It closes the repeating tickets out of your own documentation. Everything else goes to a person with the history: the question, what has been tried, what the bot checked.
- Buildfrom $1,700
- Timeline2–4 weeks
- Supportfrom $570/mo
Who this suits: Support queues made of the same questions about access, payment and «it is not working», where the hard tickets drown between them.
What is in the first version
- answers from documentation
- escalation with history
- handover into tickets
- customer rating
The figures above are the package this sits on: AI agent or process automation. Anything past its edges the calculator adds before the work starts, not after.
How the path goes
- 01
Every ticket has three routes
What the bot takes is a written list of topics rather than a guess from the model’s confidence.
- 02
The customer checks the answer
One question follows the answer, and a no is not the end but an escalation with the thread assembled.
- 03
The operator gets a case
The ticket carries the customer, the thread and what the bot already checked, so the operator starts where it stopped.
What decides whether this works or annoys people
The word «operator» always works
A request for a human is a command rather than a topic: it works the first time, at any moment.
The question is about one account
Either the bot reads the real status from your system or it hands over: reciting the general procedure annoys more.
Closure share is not the goal
Reward the share it closes and it closes what it does not solve, so solved and repeat tickets are counted instead.
What to measure once it is live
- tickets the customer says were solved
- repeat tickets on the same topic
- time to a human after an escalation
- the topics that escalate most often
Why this comes out faster
The shape of this one is known: the states, the edge cases and the things that usually go wrong have been decided before. Nothing here is a template, and the saving is not in your half of the work. It goes into your process, your content and the systems this has to talk to, which is the part nobody can have solved in advance.
A likely stack for this
Picked against the task when we scope it, not decided in advance. This is the shelf it usually comes off.
- Bot and serverPythonFastAPIPostgreSQLRedisTelegram Bot API
- Models and documentationOpenAIAnthropicpgvector
- Queues and operationsn8nTemporalDockerGrafana
Price it yourself, right here
Five steps, and you can see the number without leaving a contact. The estimate accounts for the kind of work, what you already have and what it will need inside.
These bills do not come from us
- AI model calls at the provider’s rate
- Licences and seats in your ticket system
- Channel plans charged per conversation
- Hosting and the server with thread history
Asked before the first call
- Will the customer know it is a bot?
- We recommend saying so in the first message and showing how to reach a person there. A hidden bot buys minutes and loses trust when it is worked out.
- Does it run inside our helpdesk or beside it?
- Inside yours, if you have one: it answers within the ticket and hands it over in the same place, so the history stays in one system.
- Can we edit the answers ourselves?
- Yes, the answers are your documentation rather than text in the code, and the list of topics the bot never takes is a setting too. New data sources are code.