Thread by @mredstone

Welcome to day 2 of the 5th World Chatbots & Voice Summit, hosted the the beautiful city of #Berlin

I'll be keeping this thread going to cover all the presentations and learnings of the day.

As co-chair, I'll be looking after the chatbot track, today.

#chatbotvoice2023

First up, Alin from @SapienceS2P

Talking about how to deliver value and a competitive edge through customised chatbot solutions

SapienceS2P is an #SAP procurement consultancy

Taking us through their #procurement chatbot, Sapience Satori, which can integrate with #ChatGPT, most SAP APIs and also communication channels - #teams #slack #whatsapp

For a #cpo

Streamline procurement
Find savings
Engage with supply chain

Live demo time (always a brave thing to do)

Showing the #msteams interface

Asking for a #purchaseorder report, the interface asks for dates for reports.

Delivered all POs for the dates requested.

Alin now talking through proactive notifications designed to ensure that users are comfortable with the interactions.

Demonstrating the #ChatGPT integration. Sending a prompt into teams to query the procurement data:

"Calculate number of PRs waiting for PO release"

#copilot

Using the #ChatGPT integration, can uncover massive insights on procurement data - finding saving opportunities, calculating total spend, understanding statuses. All conversational in #MSteams

Another benefit is summarising procurement contracts - asking for payment terms and other important parts of the contract, all using #ChatGPT

Now a live demo of calling the Bot from within #MSTeams

Using voice to ask the bot to query the data and getting a voice response of the answer to the query.

Now demonstrating the agent dashboard in #MSdynamics - smart assist, sentiment analysis and CRM details all in one screen.

Some great questions from the audience on training data for #ChatGPT and also demonstrating value in procurement processes

Next up. A conversation to remember from Leonid from Big Picture big-picture.com

We'll be doing a hands on workshop to build a chatbot persona

Get your best @MiroHQ hat on because we are about to jump into a very interactive session.

Quick reminder of what a conversational agent is - the interface between a human and a computer #HCI

Though, the evolution of humans and computers are running at very different paces.

I'm pretty sure that Leonid just said that we are building a chatbot for a contract killer agency.

Definitely not a serious usecase.

Looks like our team is building a persona for "slasher sloth"

And we're off at the workshop.

Team 'Slayer Sloth' featuring:
Co-chair Michelle from @BOTfriendsAI
@emiellangeberg

First task is persona creation and design:
Brand name, tasks/services and values
#Chatbot character and tone of voice

We've got a 35 year old sloth that's the love child of Danny Devito and Marge Simpson. Ernie Sanchez, the slowest killer in the west.

(I have far too much fun with persona design)

Now it's time to inject all this design into a prompt before testing out the chatbot.

With the ability to provide example dialogue in order to shape the chatbot even further.

Randomly selecting two groups to present their personas

Copying the persona into the chatbot prompt. And chatting with a sheep about dispatching the users neighbour.

Great fun interacting with a couple of different personas. But some important lessons to remember

Next up, Joachim Jonkers from @wearesinch talking about Chatbot adoption in smaller companies.

Bots tend to be adopted by larger enterprises, so what can be done to get adoption going in smaller businesses.

Let's hear about how @wearesinch can help.

We have reached the early majority phase of the innovation adoption lifecycle.

There's still huge potential in the market for more bot adoption.

How do we get the late majority and laggards onboard?

What does the market say about why the time is right for wider adoption?

1) Increased digitalisation of custimer acquisition
2) marco-economics for cost cutting
3) shifts in consumer expectations
4) momentum created by #ChatGPT

@salesforce report on the state of the connected customer (2021) shows chat messaging apps growing at a fast rate. Email and phone are still the most popular channels, but they can be automated with bots as well.

But what is preventing companies from building bots?

1) complexity
2) lack of scale
3) lack of resources
4) chatbot prejudice
5) lack of understanding

Complexity:

There's a fear that chatbot development is complicated and requires technical skills

Lack of scale:

companies feel that a chatbot isn't worth the effort for a smaller size business

Lack of resources:

Usually a lack of budget or internal expertise

Prejudice

People have had bad experiences with poor chatbots so there's a negative connotation

Lack of understanding:

Do not understand the benefits of chatbots

So what can be done to handle these challenges to adoption?

Use no/low code builder for faster ROI, makes it easy to build and allows for scale of complexity and customisation

Focus marketing messages on success stories for smaller businesses

Leverage #ChatGPT momentum

So how do different business help with adoption?

There's a role to play for #conversationdesign agencies, enterprises and bot platforms.

Up next, Navninder Singh - Digital Innovation Manager from @AkzoNobel

Powering home decor with intelligence. Mixing together digital innovations across #chatbot and #visionai

This is an interesting challenge. Most products you can pick up and feel etc. With your home decoration, you can't do that and if you're not happy, you can't take the product back.

So how do you do this:

Guided experiences with #chatbots

#visionai to power decoration experiences

So guided experiences with #chatbots

How do we empower the consumer conversationally? Start them off on one of two journeys.

Empower them to make inspired and bold decisions

Handhold them through the journey

Generate a wishlist

Different experience for professional Vs #DIY

This also allowed for a large amount of learning opportunities

Conversation asks for email/telephone number to enable #ecommerce opportunities like back to basket nudges and promotional marketing

Allows any warranty issues and claims to be expedited from weeks to days.

Since 2009, they have been building and iterating their immersive experiences

From photos and manual creation through to disrupting that with #visionai with #startup partners

This allows to provide for a more personalised offerings, an enhanced customer journey and enables staff to provide expert help, rather than spending time doing the work manually.

From a business perspective:

Faster turnaround of decor ideas etc.

Improved brand lock-in

Opened up more consumer channels

Significant savings - no more outsourcing to manual effort

Human capital is concentrating on QA rather than mundane

So what is next for the future of empowering decor

3d room modelling with #generativeai

More business successes:

Enhanced top of funnel

Multi language support

Further integrations with e-commerce partners

Integration with back end systems

#GenerativeAI for imaging and conversations

Next up - Stefan Schaeffer from the @DFKI on harvesting low hanging fruit with minimal effort using #LLMs

So first, a quick recap of different #LLMs

Commercial and #opensource (which the DFKI researchers prefer)

For many tasks, you may not need closed source large language models as open source are performing good enough for most uses.

#BERT not an #LLM is used for a museum experience chatbot

CHiM - chatbot in the museum, funded by the BMBF and a joint project between London, DFKI and Städel

Partners have major knowledge repositories about the museum exhibits

Chatbot intros itself, allows for object selection for more details as well as #multimodal museum map

As a #cognitive assistant, it has to deal with open and closed questions. Inside and outside the museum.

When was it painted? Who painted it? How big is the painting in real life?

All structured data that can be dealt with by intent mapping.

So how to deal with open questions...

Initially trained on 15 selected exhibits from the museum and offered the public to ask questions.

Then labelled questions across 12 content types.

Testing in the real world...

Initial testing produced a 62% result in providing correct answers.

The testing allowed for all possible questions, some of which were not mapped against the data available.

Next project being worked on is a toolbox for hybrid events.

Experimenting with the #vicuna #LLM which is not trained in their context, to use as a conversational concierge for hybrid events.

The Vicuna LLM is being used for intent classification and answer generation.

Wrapping up an amazing and informative talk.

Up next, Dr evangelis from @SwissRe discussing how we can start pushing the boundaries of empathy and efficiency in the digital world.

We've seen AI fitness coaches, health and nutrition coaches.

Who wants these coaches?

The fitness one lots of hands, the nutrition coach from the film Jexi, nobody.

So what do people want AI to do for them?

Service robots in hospital and restaurants have zero empathy.

Does empathy become a luxury that is only accessed through a premium?

We've seen lots of robots in sci-fi, but how do they score from an empathy perspective?

But what is empathy?

Vs compassion

As humans we perceive orangutan as more empathetic than humans.

Message here is that we perceive empathy the closer the thing is to a human.

If an AI model can perceive human emotion then they can respond in an empathetic way through movement, language etc.

We're now looking at #digitalhumans and how we can apply emotional reactions and movement to create empathy.

Can we do this with voice only?

With artificial voice synthesis.

Do consumers care whether the voices they are interacting with are real or not?

Can we use #GenerativeAI to create empathetic experiences?

We're seeing a picture of the speaker with a fake voice and 3d movement created very simply.

Reality is no longer binary

Biggest opportunities for mixing virtual with real:

Health coaching
Wellbeing
Retail

Empathetic bot interactions outperform human interactions.

Now, can you take everything written by a therapist and get a bit trained on this data.

The therapist (obviously) disagreed as the bot cannot feel human emotion.

Now seeing a demo of an AR health coach, plus some gamification of workout partner.

Welcoming the final talk of the day and summit - Grant Ronald from @Oracle

Discussing his experience of the challenges of building conversational assistants in the real world

The good, bad and ugly.

Let's start with this.

It's not difficult to build a digital assistant, but it is harder than you think.

So why is it harder than we think?

First major challenge.

The business.

They usually come up with crap usecases

Does the business understand the lifecycle and effort...

Training the team, training the models, UX efforts, retraining NLP, measuring success etc.

Business understands business, not #CAI

It helps to ensure that everybody is aligned to the project plan and timeline.

So what does the team look like for a successful project?