Subscribe
Sign in
Talklab
p/talklabAI-Powered Chat Analytics for Customer Insight
Start new thread
Willian Valle

Talklab - AI powered chat analytics for customer insight

by•

Say goodbye to generic customer service metrics and hello to AI-driven insights with Talklab. Our platform analyzes customer chats to offer detailed, actionable reports, from sentiment scores to behavioral tags. Lower churn rates, elevate customer satisfaction

Add a comment

Replies

Best
Willian Valle
Maker
📌
Hello Community! 🖖 I've been working with software development for almost 20 years now, having founded a software house (named SofaCoding) that operated from 2014 to 2018. In recent years, I've been working a lot with customer support and messaging markets, which alerted me to the deficit of information that companies have about what their customers are demanding in support channels. With the recent popularization of large language models, I decided to start a product that could use the companies’ chats and tickets to answer some key questions: 1. What are the main topics that are leading customers to support channels and how the appearance of these topics is evolving over time? 2. Who are the customers that have demonstrated negative sentiments lately, such as frustration or anger? Thanks to the PaLM 2 and GPT3.5 language models, the first public version of Talklab can answer these questions and many more. Currently, our software only integrates with the Zendesk platform, enabling Zendesk users to import chats and tickets. Then, Talklab will automatically generate data about the content of each chat. Using the generated tags, it's possible to create reports based on groups of tags and for a specific time period. We're launching a preview version today that enables the Product Hunt community to explore a demo workspace and create prompt-generated filters, which we are calling 'smart tags.' To access the demo, please head to https://talklab.tech Additionally, we're opening a free-beta whitelist registration that you can join here: https://forms.gle/z4LR2dMop66CAZBq6 -- Willian Valle 🇧🇷 willian@talklab.tech
Anna Kasumova
Hi Willian! Congrats on the launch. In my first business project over 10 years we control customer support every day be reading all messages and listening all voice records. It works, we really keep the qulity of our customer support service. But it's a human-factor. The QA who checks it, with time can be not so strong with the mistakes. Also they start friend and the quality of testing becomes weaker in time. Also we have the exam system which analyze the work of every agent and gives rating. But it's not AI driven and takes a lot of resources. Your tool can help us and a lot of B2B companies who cares about the customer support. I think I can start test it the nearest time. What about your monetization scheme?
Willian Valle
Maker
@anna_kasumova Hi Anna. Thanks a lot for your feedback!
I believe that, whether using humans or ai, it’s important for the customer support market develop an objective set of parameters that define the quality of a service. So that we can avoid the bias and randomness that both ai and humans have. During Talklab development, I tried a lot of prompts to achieve a fair score of quality, but currently talklab just presents a score that identifies the mood of the customer, which is not suitable to identify the service quality. But, I haven't given up yet 😂 and I’m constantly looking for works or discussions in this field. About monetization: It's the decision that is keeping me awake at night. 😅
It seems clear that I'll pursue a SaaS, subscription-based model, but I haven't determined the pricing yet. I have significant variable costs, primarily based on the number of characters sent to language models. Therefore, the plans should feature different thresholds for character counts per billing period. Do you think it would be acceptable to charge based on the number of characters processed per month?
Anna Kasumova
@vallewillian I think that easier - to make monthly subscription based pricing, without symbol limits. Why? Because pricing MUST be clear for customers. And even I with a lot of experience, I really can't understand the quantity of symbols. It's hard to calculate and understand. Better if you will calculate yourself, and based on your understanding, you make pricing which cover all expenses.
André J
Now this is cool! What are some of the most common actionable insights your customer finds with this tool?
Willian Valle
Maker
@sentry_co Hi André :) Thanks for the feedback! (are you a brazilian, btw? 😆) Generally speaking, I believe these are the greatest value deliveries that the tool can bring: 1. What are the main topics that are leading customers to support channels and how the appearance of these topics is evolving over time? 2. Who are the customers that have demonstrated negative sentiments lately, such as frustration or anger? The insights brought by the first question can help service teams identify and respond faster to customer demands, while the second question can help reduce churns.
André J
@vallewillian I added it to my PH collection. Will revisit when we start rolling out a bit more widely. And no, I'm Norwegian.
Naveed Rehman
Hey there! Congrats on the awesome launch of Talklab! The AI-powered chat analytics for customer insight sounds super intriguing. One suggestion to consider for future improvements would be to explore integrating real-time sentiment analysis. It could offer even more valuable insights into customer experiences. Have you considered any unique use cases for Talklab in specific industries?
Willian Valle
Maker
@naveed_rehman Hi Naveed. Thanks for your Feedback! About the real time analysis: currently, the platform is able to process zendesk conversations within a few minutes after the ticket or chat is finished. However, initially I think about working with a limit of up to 12 hours for conversations to enter the platform. This while operating in beta. In the future, I hope to implement an email alert feature, in case a service appears that has mentioned a specific subject or sentiment. About specific industries: There are several markets that I imagine could benefit from this type of service (such as BI areas, or companies with specific compliance needs). Today I have only implemented the basic functions, but I hope that the beta phase will help me to better understand these market demands and which use cases I should focus on first.
Philipp Shay
It appears to be a fantastic tool. Congratulations and best of luck! I'm curious about how you generate the topics you associate with each customer conversation. Are you using ChatGPT for this?
Willian Valle
Maker
@kinzarra Hi Philipp. Thank you for your feedback :) Currently, I employ two major language models for processing each chat: Google's PaLM 2 and OpenAI's GPT-3.5. (the base model of chat gpt) Both are supported by the excellent Python framework, Langchain. If you are interested, I have written an article on Medium about this framework: https://medium.com/artificial-in...
Valeriia Dziubenko
Wow, Talklab sounds like an incredible tool for customer insights! I'm excited about the AI-powered chat analytics feature. Can you tell me more about how it analyzes customer chats? Does it use natural language processing to understand sentiment and behavioral patterns? Also, I recently came across a study highlighting the importance of personalized customer service in reducing churn rates. Have you considered incorporating personalization features into Talklab? It could be a game-changer in terms of boosting customer satisfaction. I look forward to hearing more about your product and how it can revolutionize the customer service industry!
Willian Valle
Maker
@valeriia_dziubenko Hello Valeriia! Thank you for your feedback. Certainly, I can elaborate on the technology I am using. Currently, I employ two major language models for processing each chat: Google's PaLM 2 and OpenAI's GPT-3.5. Both are supported by the excellent Python framework, Langchain. If you are interested, I have written an article on Medium about this framework: https://medium.com/artificial-in... In this setup, the content of each chat is transformed into a vector and sent to the language models, along with natural language commands to request the primary topics and sentiments. Talklab users can create personalized filters using prompts. For instance, a prompt like "chats where the customer was rude" will generate a tag for each conversation fitting this specific pattern. I am quite intrigued by the study you mentioned. Do you have any references to share? Additionally, what kinds of personalization features do you think would synergize well with Talklab? Thank you once again for your support and feedback.
Valeriia Dziubenko
@vallewillian Thank you! Sounds reasonable! Here's an article I've mentioned: https://www.idomoo.com/blog/how-... Regarding the possible personalization in Talklab, to start with, I assume it would be great to quickly get the information when specific customers are disappointed or concerned about something so that we can instantly react and try to find possible satisfactory solutions for them before they leave.
sumit sharma
Cool Thumbnail!
Willian Valle
Maker
@sumit_sharma29 Thanks, Sumit! This character was created using Midjourney V5 (and adapted later using illustrator).
Lucas Pilzen
Nice idea! Gotta give it a shot! :)
Willian Valle
Maker
@lucas_pilzen Thanks, Lucas!
Chalie Clark
TalkLab looks like an interesting platform for team communication. How does it enhance collaboration within teams? 💬
Willian Valle
Maker
@chalie_clark Hi Charlie! Thank you for your feedback :) The current version of the software always delivers the same view, for all accounts associated with the same workspace (total view of conversations). However, I thought of bringing a feature to allow users to only see their own conversations in a future version. This would allow agents to use talklab to prioritize customer service or retrieve past issues.