Hurdles to Overcome in AI

Hurdles to Overcome in AI

Last year, I stumbled upon two successful business models—Chorus and Gong— which provide suggestions to salespeople while they are having conversations. These products listen into conversations a salesperson has with a potential customer. They identify pivotal moments in conversations and allow salespeople to easily follow up with customers and close more deals.

Within 10 minutes of first finding these products, it was obvious to me that is how people will interact one day—that is, we'll use conversational intelligence (an industry within artificial intelligence) to better connect with one another.

More recently, I spent 25 minutes brainstorming with my friend and AR expert Tony Morales. Below is the results of our  short brainstorm.

[:00-:05] Top problems facing conversational intelligence today

Lexical categorization
Speech to text not 100%
Speaker recognition in early phases
Microphones in loud environments
Relatively quiet environment is required
People don't want to be recorded all the time
Augmented reality is not mainstream
Most AR devices do not have microphones
Trouble recognizing niche keywords, terms, abbreviations
Concern about privacy, data, storage, and how it will be used
Schools/healthcare systems might not adapt if accuracy is low

[:05-:10] Prioritize the largest problems & turn into an opportunity

Trouble recognizing niche keywords, terms, abbreviations → How Might We build a conversational intelligence system in a way that's valuable while not being 100% correct and not having 100% coverage?

[:10-:15] Identify solutions to the challenge/opportunity

Pick an niche industry with lower barriers to entry
Pick an industry that CIQ is a "nice to have"
Help people better their relationships with others by providing generic suggestions
Use existing lexical analysis tools to find one that's great
Take an existing AI tool and use it for a different industry
Save audio for later analysis and backfilling
Target fun, frivilous, or ancillary use cases
Prompt users at the start and end of a conversation
Focus on transcribing smaller sections but at higher accuracy
Build a professional development tool that addresses people's learning skills
Listen into Teams/Zoom calls and provide feedback based on call
Create a single script from multiple devices

[:15-:20] Prioritize the best solutions

[:20-:25] Make an action plan

Lastly, we designed a quick 2 week experiment for the low-effort, high-impact solutions. Feel free to reach out if you're interested in our findings!

  1. Review Zoom/Teams API for transcription strategy
  2. Build 2-3 sustainable business models for Zoom/Teams listen-in product
  3. Test with 5-7 potential customers and log feedback