ChatGPT is very good at data extraction

Learn how we’re automatically turning conversations into spreadsheets with AI.

Marketers have tried for years to implement marketing chatbots that would collect valuable data, but the approach was always flawed with marketers using traditional automation workflow tools to try and design a conversation. Here are three common problems I encountered having helped 20,000+ marketers implement Facebook chatbots.

1. Marketers would try and design a conversation, creating steps or “Funnels” that would ask users to input a response and hope that users got to the end. Hope is not a strategy, and conversational funnels rarely work as end users almost always went off script by asking questions, and in frustration, not wanting to chat with a bot at all but with a human.
2. With or without automated responses, extracting data from a chat transcript was a huge pain. It’s tedious and prone to error.
3. The end goal for many Facebook Marketers was to collect lead data in Google Sheets because many were “Agencies” and they could just share the lead data with their clients without involving them in the complexity of a chatbot or how they collected the data. Every business already uses spreadsheets, a universal format for unstructured data.

How ChatGPT solves the problem

Conversations are messy collections of unstructured data, but ChatGPT is very good at generating structured data from unstructured conversations. In this prototype application we developed, we connected Facebook Messenger to OpenAI’s API and tell ChatGPT (using the API) that we want it to extract data from messages and write the data to Google Sheets.
We simply use the Google Sheet columns headers to prompt AI for the data we want to collect and when the data can be extracted it’s written to the Google Sheet in real time.

Let’s Look At Some Use Cases!

In this first scenario, a real estate agent for example, might have a number of data points they want to collect from prospects.

In this next example, an order number might be embedded inside a longer message, but with Botsheets you’ll instantly get the order number just by adding “Order Number” as a column header. You can add custom headers like this using natural language and in ANY LANGUAGE!

Here I added “Summary” as column header and Botsheets will summarize the entire conversation in a single cell. Want to track customer sentiment for all your conversations? Just add “Sentiment” as a column header. That can help you identify trends and opportunities for improvement.

The use-cases are infinite and because the data lives in Google Sheets it’s easy to manage, collaborate on, or share, and you can feed stored data into Google Data Studio, or any 3rd party analytics or CRM tool using Zapier (or Integromat/Make.com)

How would use this tool?

While this is not a commercially accessible tool yet, here are just a few use cases we’re already testing

Sales & Lead Generation

• Detect promising leads
• Enrich your contact database
• Improve CRM processes
• Generate segmented email and marketing lists
• Generate custom audience data for Facebook ads

Customer Support

• Analyze responses of customer surveys
• Prioritize support tickets
• Manage ecommerce order details in an organized way.
• Detect negative sentiment
• Get a complete transcript summary without scanning through a lengthy transcript.

It’s still early stages for Botsheets and it’s OpenAI integration, but the time-to-value from this early version of an AI writer is promosing.

See our latest AI work over at botsheets.com