A few years ago I designed a product like this, but more focused on sales calls. We called it DiscoveryCall.ai and and my friend built a proto-type. It was cool, but the sales cycle was something I didn't want to tackle.
A few ideas & learnings if they can help.
1. If you focus on a specific domain of conversation, you can pull conversation data and "summary" into a more valuable data structure. For sales it might be identifying the Decision Maker, budget or launch date. Project meetings will have budgets, timelines, approvers etc as well.
2. My thesis was that getting data into a structured format was more interesting than summary. Sales people need to enter their data into Salesforce. Sales people hate doing this and do it poorly. It's valuable if an application can process the recorded conversation into a completed SFDC record/object.
3. It might be possible to rely on the user to prompt the application correctly in the conversation. In a project management scenario, the user might want to collect launch_date as a field. The user would ask, "When is the launch date?"
The response might be, "I don't know, we're thinking about March or maybe April or even May."
The user of the application can then repeat. "OK, let's pencil in the launch_date for April 15th."
During onboarding or setup for this type of conversation and data extraction, the user would set these keywords like "Pencil In" to then save the date mentioned in the conversation.
Anyways, that's what I remember. If any of that is interested or you want to see any old notes and wireframes, hit my email.
A few ideas & learnings if they can help.
1. If you focus on a specific domain of conversation, you can pull conversation data and "summary" into a more valuable data structure. For sales it might be identifying the Decision Maker, budget or launch date. Project meetings will have budgets, timelines, approvers etc as well.
2. My thesis was that getting data into a structured format was more interesting than summary. Sales people need to enter their data into Salesforce. Sales people hate doing this and do it poorly. It's valuable if an application can process the recorded conversation into a completed SFDC record/object.
3. It might be possible to rely on the user to prompt the application correctly in the conversation. In a project management scenario, the user might want to collect launch_date as a field. The user would ask, "When is the launch date?"
The response might be, "I don't know, we're thinking about March or maybe April or even May."
The user of the application can then repeat. "OK, let's pencil in the launch_date for April 15th."
During onboarding or setup for this type of conversation and data extraction, the user would set these keywords like "Pencil In" to then save the date mentioned in the conversation.
Anyways, that's what I remember. If any of that is interested or you want to see any old notes and wireframes, hit my email.