How to build your AI support agent with no code
Build AI support agent with zero coding
I built an AI support agent for my company that answers inbound support emails automatically. It knows the details of our product and how to handle frequent questions. I built it with zero coding in one hour. It answers > 1000 emails per month. It costs us $50. Here is how.
I have an online business where customers can contact us via email. The questions are usually about pricing, product features, bugs, or payment issues. They are not complex, but to answer you do need to know the details of our product. You also need to understand a user's specific case. The requests are in different languages.
My solution uses 3 platforms: OpenAI, Gmail, and Zappier to stitch them together.
When I got tired of repetitive support work and thought of hiring our first support agent it hit me that there should be a better way. The support requests are somewhat typical and my guess was chatGPT-4 should be able to handle them if given the right context upfront.
I checked some SaaS solutions available on the market. The one that might work is Intercom. But the pricing is crazy. They want $1 per "resolution", i.e. a dollar per chat. No, thank you Intercom.
So I decided to build it myself. It appeared to be easy, with no coding required. You can do the same for your business in an hour.
Here are the exact steps.
Part 1 - create an Assistant in OpenAI.
Sign up for an OpenAI API account.
Navigate to the Playground and select Assistant.
AI support agent using OpenAI’s Assistants Create an assistant. That's where the magic happens. We need the AI to know the details of our business. The best replies to the standard questions, frequent issues, details of the pricing plans, etc.
First, type in the prompt with basic instructions like "You are a support agent named Anna for our company providing so and so service. Please be polite and concise, and always greet users.” Ask to reply in the same language the question is if you have international clients.
Provide instructions for your custom support agent Then provide the specific details on how to handle standard requests. You can provide instructions as plain text in the same prompt field. But you can also attach them as a text or PDF file. Using a file can be easier so you can keep a copy of the file that you update with new information and product changes, and update this file from time to time in OpenAI interface.
Test the assistant. Go and ask the questions in that same interface. Just copy the last 5-10 questions you got from your customers and see if your assistant provides the expected replies or if you need to tweak your instructions.
Part 2 - connect your Assistant to gmail using Zappier and setup the automation
Create a Zappier account and click "Create" to create new automation.
You will need to create only free steps in that automation. The first step is to launch automation based on new inbound email. Chose Gmail, connect your account, and select "new email" as event triggering the automation.
AI support agent with Gmail and Gmail in Zapier The second step in the automation is ChatGPT - a conversation with an assistant. Connect ChatGPT, the event is: Conversation With Assistant. Provide the required login details. In the configuration of this task, you should use the "body plain" field from the previous step's email in the message field. You must select your assistant from the list of assistants.
The third step is again Gmail. The event is "create draft reply" if you want to review the replies manually. But you can also set it to send emails automatically. In the configuration of this event, you need to select the thread ID from the first step in the Thread field. You need to select the email address from the first email in the field "To". And you add the "from" field. At the end, you add the response from the ChatGPT step as the body of the email.
That's it. You are ready to go. Test the automation and see how it works. You can add additional steps like automatically classifying each inbound request to collect statistics and understand better what needs to be done to reduce inbound inquiries. This can be done as an additional step that sends the message to ChatGPT to add the category to the request and then save it into Google Sheets where you can gather the statistics.
The whole thing cost me around $50/month to handle around 1000 monthly inbound support emails. The cost is mostly for Zappier, it's cheap to handle that number of short emails with ChatGPT. So, kids, build your own AI agents, and don't get your business bankrupt by Intercom.
If you have any questions you can reach me at @mstysin on X.