1 of 54

Paul Sweeney �Co-Founder / Chief Strategy Officer

What’s changed and how it matters

From Communications to �Conversations

2 of 54

3 of 54

4 of 54

Everybody

Has a Plan

Until they get punched in the face

Mike Tyson

5 of 54

6 of 54

Conversational AI �is directionally correct….

7 of 54

LLM’s are great �disambiguators and solve the discovery problem

�

Moving from connecting channels to connecting to what happens in channels.

���

�

​

Access to broad and narrow 3rd party services via plug ins will be in layers, from selected, to approved, to called up, to created on the fly��

​

The search query becomes �the prompt chain

��

�

​

Text-first experience today, �but then multi modal��

�

​

The intelligent assistant becomes the new UX

Well, yes. Kinda.

�

�

​

8 of 54

LLM

NLU

DM

Keeping Conversations on the Rails...

9 of 54

Custom LLMs in Context �Gets Complicated

LLM

DM

Channels

Knowledge Bases

NLU

API

UI

Storage

Vector DB

Models

Model Mgt

Micro/ services

MLOPs

LLMOPs

Creator Tools

Admin

Security

Audit

Reports

10 of 54

The Language Model Stack

David R. Oliver https://medium.com/future-architecture/the-language-model-stack-d38c8de880ec

11 of 54

What We Learned �From Above ….

Information is matched; interactions enabled; transactions made, all orchestrated in the one conversation flow. ��This is how value gets released. But you need it all.

​

The LLM’s will be open source, customized, trained. That takes time and effort.

​

Not one LLM, but multiple LM’s being trained. That requires LLMOPs.

​

Model accuracy, intent fit, outcomes all have to be assured and controlled for. Custom data, values, numbers etc.

​

All this has to be performative, low latency, and super low cost. Scale a suboptimal architecture and see what happens to your AWS bill.

​

12 of 54

Conversational AI Today….

Most companies don’t have access to the API to bring in the data

​

​

Evolving use of chain of thought, chain of verification, chain of… prompt engineering. ��Results still vary. Might still get performed by future versions of the LLM (GPT6)

​

​

​

Co-pilots are the UI, the experiences still must be designed. Chat is customer co-pilot, agent assist is the employee assist. ��Still, pretty generic. Work is deeply contextual.

​

Copy generation, smart replies, summarization, conversation labelling for intent-based routing are features here today.

​

But not yet “transformative”.

​

ChatGPT Enterprise v Custom LLM Models �is TBD.

​

Enterprises very concerned with integrity of own data, and value of own data.

​

13 of 54

Conversational AI Further Questions….

Voice Interaction is still input, instruction based. If that. Keep an eye on Voice interaction on ChatGPT Mobile App. Behavior will follow engagement.

14 of 54

Conversational AI Further Questions….

IVR versus voice assistant on a website? ��Sounds great but what's the use case? ��No evidence its driving down inbound calls for instance

15 of 54

Conversational AI Further Questions….

10x conversational data in voice: performance for one, then more biometric type data. ��Yet, no permissions, ethical AI rules.

16 of 54

Conversational AI Further Questions….

Voice cloning, voice ID, avatar experiences. �Sure, for entertainment. ��Might be more of a fraud and security issue today.

​

17 of 54

Conversational AI Further Questions….

Conversational and metaverse would be synergistic, but nobody gets customers service in the metaverse IMHO.

18 of 54

CPaaS Provocations

19 of 54

CPaaS Provocations

Personal assistant co-pilot �as future customer interface

���

�

​

20 of 54

CPaaS Provocations

Communications data, and �knowledge base data, �are both required to facilitate deeply personalized customer conversations�

21 of 54

CPaaS Provocations

Deep data driven insights into “everything” that could matter

��

�

​

22 of 54

CPaaS Provocations

Need for customer intimacy �to deliver all the above.

��

�

​

23 of 54

CPaaS Provocations

Communications data, and �knowledge base data, are both required to facilitate deeply personalized customer conversations�

Personal assistant co-pilot �as future customer interface

���

�

​

Technology �stack performance at scale �for real time processing��

�

​

Deep data driven insights into “everything” that could matter

��

�

​

24 of 54

CPaaS Provocations

Vertical have own �workflows, data, integrations, and models

�

�

​

25 of 54

CPaaS Provocations

Where do you partner, �with who, and for what reason?

�

�

​

26 of 54

Insider news…

  • Nobody gets automated out of their job. Unaddressed work pours in.
  • Focus on JTBD and see how LLM’s etc. can be used.
  • The LLM’s are getting charged for storage. There is a lesson in that.
  • LLM driven services need zero latency. There is a lesson in that.
  • Voice LLM’s are hard. They won’t always be. Then they will very profitable.
  • Automate entire processes to deliver solid value proposition.

​

27 of 54

SuperPrompt≈

  • As a senior analyst in the CPaaS or customer experience SaaS industry, working for a large CPaaS organization, your task is to describe the potential impacts that large language models (LLMs) will have on different functional areas of your organization. Start by analyzing the potential effects of LLM technologies on each functional area and then focus on the cross-functional impacts. Once you have completed this analysis, use your insights to evaluate the functionality of your products and the way you deliver services to your customers. Rate these different effects in order of their potential disruptive impact on your existing business model. Explain how these effects can be used in combination to create a new type of customer engagement service that would disrupt Twilio. Next, imagine yourself as Jeff Lawson, CEO of Twilio, and critique the disruptive service description in ten conversational interactions with the senior analyst. Extract the main points from these interactions and create a four by four diagram that summarizes the results. Your response should provide a thorough analysis of the potential impacts of LLMs on different functional areas, considering both individual and cross-functional effects. It should also include an evaluation of the disruptive potential of these impacts on Twilio's existing business model. Your analysis should be detailed and well-supported, providing clear explanations and examples. Please note that you should consider the specific context of Twilio and the CPaaS industry when analysing the potential impacts and evaluating the disruptive service description.

28 of 54

TAKEAWAYS

Customer service getting disrupted

1

Internal enterprise �silos remain

3

Conversations contain events and data

2

29 of 54

paul.sweeney@webio.com

Thank you

30 of 54

31 of 54

32 of 54

Learnings from the past year

The rise of �digital collections

1

Internal enterprise �silos remain

3

Automation works and it’s very powerful

2

33 of 54

CPaaS Provocations

  • Personal assistant co-pilot as future customer interface
  • Communications data, and knowledge base data, are both required to facilitate deeply personalized customer conversations.
  • Deep data driven insights into “everything” that could matter.
  • Need for customer intimacy to deliver all the above.
  • Technology stack performance at scale for real time processing
  • Vertical have own workflows, data, integrations, and models
  • Where do you partner, with who, and for what reason?

​

34 of 54

Insider news…

Nobody gets automated out of their job. Unaddressed work pours in.

Focus on JTBD and see how LLM’s etc. can be used.

The LLM’s are getting charged for storage. There is a lesson in that.

LLM driven services need zero latency. There is a lesson in that.

Voice LLM’s are hard. They won’t always be. Then they will very profitable.

Automate entire processes to deliver solid value proposition.

​

35 of 54

SuperPrompt≈

  • As a senior analyst in the CPaaS or customer experience SaaS industry, working for a large CPaaS organization, your task is to describe the potential impacts that large language models (LLMs) will have on different functional areas of your organization. Start by analyzing the potential effects of LLM technologies on each functional area and then focus on the cross-functional impacts. Once you have completed this analysis, use your insights to evaluate the functionality of your products and the way you deliver services to your customers. Rate these different effects in order of their potential disruptive impact on your existing business model. Explain how these effects can be used in combination to create a new type of customer engagement service that would disrupt Twilio. Next, imagine yourself as Jeff Lawson, CEO of Twilio, and critique the disruptive service description in ten conversational interactions with the senior analyst. Extract the main points from these interactions and create a four by four diagram that summarizes the results. Your response should provide a thorough analysis of the potential impacts of LLMs on different functional areas, considering both individual and cross-functional effects. It should also include an evaluation of the disruptive potential of these impacts on Twilio's existing business model. Your analysis should be detailed and well-supported, providing clear explanations and examples. Please note that you should consider the specific context of Twilio and the CPaaS industry when analyzing the potential impacts and evaluating the disruptive service description.

36 of 54

�Take aways���Customer service getting disrupted�Conversations contain events and data�Co-pilots for everyone, even you. �

37 of 54

END Original Slides

38 of 54

"From communications to conversations – �what’s changed and how it matters”�

39 of 54

40 of 54

41 of 54

42 of 54

43 of 54

Conversational AI is directionally correct….

  • Moving from connecting channels to connecting to what happens in channels.
  • LLM’s are great disambiguators and solve the discovery problem
  • Access to broad and narrow 3rd party services via plug ins will be in layers, from selected, to approved, to called up, to created on the fly.
  • Text-first experience today, but then multi modal.
  • The search query becomes the prompt chain.
  • The intelligent assistant becomes the new UX ? Well, yes. Kinda.

44 of 54

So, how to keep conversations on the rails..

LLM

DM

NLU

45 of 54

Custom LLM’s in context gets complicated

LLM

DM

Channels

Knowledge Bases

NLU

API

UI

Storage

Vector DB

Models

Model Mgt

Micro/ services

MLOPs

LLMOPs

Creator Tools

Admin

Security

Audit

Reports

46 of 54

The Language Model Stack

David R. Oliver https://medium.com/future-architecture/the-language-model-stack-d38c8de880ec

47 of 54

What we learned from above…

  • Information is matched; interactions enabled; transactions made, all orchestrated in the one conversation flow. This is how value gets released. But you need it all.
  • The LLM’s will be open source, customized, trained. That takes time and effort. Not one LLM, but multiple LM’s being trained. That requires LLMOPs.
  • Model accuracy, intent fit, outcomes all have to be assured and controlled for. Custom data, values, numbers etc.
  • All this has to be performative, low latency, and super low cost. Scale a suboptimal architecture and see what happens to your AWS bill.

48 of 54

Conversational AI Today

  • Most companies don’t have access to the API to bring in the data
  • Evolving use of chain of thought, chain of verification, chain of… prompt engineering. Results still vary. Might still get performed by future versions of the LLM (GPT6).
  • Co-pilots are the UI, the experiences still must be designed. Chat is customer co-pilot, agent assist is the employee assist. Still, pretty generic. Work is deeply contextual.
  • Copy generation, smart replies, summarization, conversation labelling for intent based routing are features here today. But not yet “transformative”.
  • ChatGPT Enterprise v Custom LLM Models is TBD. Enterprises very concerned with integrity of own data, and value of own data.

49 of 54

Conversational AI further questions

  • Voice Interaction is still input, instruction based. If that. Keep an eye on Voice interaction on ChatGPT Mobile App. Behavior will follow engagement.
  • IVR versus voice assistant on a website? Sounds great but what's the use case? No evidence its driving down inbound calls for instance.
  • 10x conversational data in voice: performance for one, then more biometric type data. Yet, no permissions, ethical AI rules.
  • Voice cloning, voice ID, avatar experiences. Sure, for entertainment. Might be more of a fraud and security issue today.
  • Conversational and metaverse would be synergistic, but nobody gets customers service in the metaverse IMHO.

50 of 54

CPaaS Provocations

  • Personal assistant co-pilot as future customer interface
  • Communications data, and knowledge base data, are both required to facilitate deeply personalized customer conversations.
  • Deep data driven insights into “everything” that could matter.
  • Need for customer intimacy to deliver all the above.
  • Technology stack performance at scale for real time processing
  • Vertical have own workflows, data, integrations, and models
  • Where do you partner, with who, and for what reason?

​

51 of 54

Insider news…

Nobody gets automated out of their job. Unaddressed work pours in.

Focus on JTBD and see how LLM’s etc. can be used.

The LLM’s are getting charged for storage. There is a lesson in that.

LLM driven services need zero latency. There is a lesson in that.

Voice LLM’s are hard. They won’t always be. Then they will very profitable.

Automate entire processes to deliver solid value proposition.

​

52 of 54

SuperPrompt≈

  • As a senior analyst in the CPaaS or customer experience SaaS industry, working for a large CPaaS organization, your task is to describe the potential impacts that large language models (LLMs) will have on different functional areas of your organization. Start by analyzing the potential effects of LLM technologies on each functional area and then focus on the cross-functional impacts. Once you have completed this analysis, use your insights to evaluate the functionality of your products and the way you deliver services to your customers. Rate these different effects in order of their potential disruptive impact on your existing business model. Explain how these effects can be used in combination to create a new type of customer engagement service that would disrupt Twilio. Next, imagine yourself as Jeff Lawson, CEO of Twilio, and critique the disruptive service description in ten conversational interactions with the senior analyst. Extract the main points from these interactions and create a four by four diagram that summarizes the results. Your response should provide a thorough analysis of the potential impacts of LLMs on different functional areas, considering both individual and cross-functional effects. It should also include an evaluation of the disruptive potential of these impacts on Twilio's existing business model. Your analysis should be detailed and well-supported, providing clear explanations and examples. Please note that you should consider the specific context of Twilio and the CPaaS industry when analyzing the potential impacts and evaluating the disruptive service description.

53 of 54

�Take aways���Customer service getting disrupted�Conversations contain events and data�Co-pilots for everyone, even you. �

54 of 54

What We Learned �From Above ….

Information is matched; interactions enabled; transactions made, all orchestrated in the one conversation flow. ��This is how value gets released. But you need it all.

​

The LLM’s will be open source, customized, trained. That takes time and effort.

​

Not one LLM, but multiple LM’s being trained. That requires LLMOPs.

​

Model accuracy, intent fit, outcomes all have to be assured and controlled for. Custom data, values, numbers etc.

​

All this has to be performative, low latency, and super low cost. Scale a suboptimal architecture and see what happens to your AWS bill.

​