TADSummit 2022 8/9 Nov Aveiro Portugal
CPaaS Conversational Platforms and Conversational Customer Service – The Experience Gap”?
Ben Waymark, Chief Technology Officer, Webio.
CPaaS players are doing the low hanging, simple conversations via their conversational design and plug in’s to the messenger layer, but what are they really hoping to achieve, and should they be aimed at the developer community?
No-code low-code configurable conversational customer support have done really well by integrating with customer ticketing, and integrating other platforms into their workflows. Kustomer.com was bought for a billion, something is going right there.
Conversational experiences are becoming part of the digital customer experience. What does this look like and why might this be important for other companies to understand?
to make a truly conversational platform are:
Restful APIs
Good programming/development tools that facilitate rapid development
Solid basis in conversational Text then move on to voice
Get and Keep context
Great UX driven conversational design
Nomenclature, vocabulary and accents right
Self-learning/Self-correcting training so it can grow
Webio builds conversational interfaces for the credit and collections industry (at the moment) with emphasis on good customer experiences
20+ years experience in innovation including telephone trunk monitoring systems in call centres (20+ years ago), interactive tv, SMS bulk messaging for horse racing tips and automated chatbots
Backgrounds as a software engineer, and have product management, technical delivery, strategic planning with strong leanings towards marketing and usability
Pre webio was building chatbot for travel industry as part of their innovation lab
Accent is Canadian, vocabulary is British and Irish as I live in England and work in Ireland
a Jesuit who wrote “Orality and Literacy” 1982, in it he spoke about Oral societies where transformed by literacy
Pre oral we memorise things and tell stories laden with pneumonic devices
One of the big transformations is how we go from a memory based society to a written-law, written holy books, and physical boundaries, the fluidity of culture can change because we have ideas and rules set it stone
He talk of oral societies, literate societies then secondary oral societies which is an oral society based on the written word (plays, radio, tv, and now internet) and this makes its orality more contextual, nuanced, and sophisticated
The interesting progression here is how we don’t stop at literacy, the urge is to come back to orality
Computer engineering similarly going “secondarily oral” conversationally
Its how we think and its what everyone wanted
Everyone bought Alexa because they wanted it to work
But it doesn’t do what we want, it does single commands, not turn by turn conversations because it doesn’t manage context, its not secondary orality
When Ong spoke of oral societies he spoke of the importance of memory and pneumonic devices and literacy erodes this: Conversation isn’t barking specific commands, its saying those commands in infinite variety of ways
Alexa and the like only work if we remember all the commands, which is rubbish
But first let me start with a quick history
Punch cards to joysticks keyboards to mouse to fancy mice …. But we’ve always wanted to speak!
Speech was held back until recently by voice recognition
Conversational held back by the sequential logic of scripted programming languages
with programming languages we went from logic gates to assembly to BASIC TO every sophisticated 3rd, 4th, 5th generation languages, OO etc all trying to make programming more conversation
But as we get into pattern matching, AI and Data Science the sophistication grows but some of the core problem remain
Until AI & DS context has always been supplied by logical flow. But context has also been limited by interoperability, because it can only get context from one application on one device, and without context we struggle,
But first a history interoperability
Connecting TRS80-100 modem to Apple SE, setting parity and NBIT, 9600 baud and away it went!
Then Apple started reading DOS disks
Then Usenet and the alt groups, Archie and Veronica
Then Netscape and https
Which leads us to ….
Interoperability between all devices: IoT, phones, laptops, servers, containers etc all possible through restful services
You can use any programming language or operating system
You can setup your Alexa to connect to the RestFul API for the Dublin Bus service thanks to Resftul API
When you want to get your CRM to connect to an SMS aggregator or to WhatsApp or to WeChat or Telegram you do that through Restful APIs
But how do we get context
To ask alexa what the bus time is, I need to know you are in dublin
Cookies
Data and data science tools like digital twinning (give examples), total life time value (recently, frequency, value and spend)
CRMs
Channel (Facebook? LinkedIn? SMS? WhatsApp?)
Intent recognition in AI (give an example)
But it is hard
“How much is a flight to Paris?”
“What is the weather like there?”
“Do I need a VISA?”
“Is it going to be sunny this weekend?”
--play video--
They key will be to confirm what is being said, and not be afraid to get it wrong (Googles biggest innovation was accepting close enough rather than perfect)
Big usability mountain to climb with rubbish design tools that build rubbish conversations
Let me tell you about my Alexa, because it shows the problem with the interfaces.
Alexa – playing radio is great, driving is great, turning lights off is great
Bedroom not recognised as bedroom light, have to remember what to call everything. Still have to code; still need those tools we lost with literacy!
Voice is always the goal
WeChat is incredible and can translate
Voice clips
Voice searches (WeChat example)
Pizza talking
Alexa & Spelling
Android Auto & Driving
There hasn’t been a runway app
WeChat style search filtering hasn’t arrived
Accents still a problem – polish English fine, Hiberno-English less so
It’s not a natural conversation … it’s commands by voice
KLM did huge effort, had agents train, had all the data. Still isn’t great.
UX and ‘calm technology’
Good UX starts with good interfaces to build; most of the awful interfaces that have been produced start with rubbish programming interfaces (Remember MS FrontPage?)
IBM Watson or Amazon Alex and building a bot in there and how hard it is.
If this then that interface
Perfecting “Bricks” then stringing bricks together – ID&V is one example, Payment, Promise to Pay etc
AI/ML and DS play a big role, but the solution is fundamentally UX not AI … buttons vs intent example
When we get the bricks right, we need to get conversational right – thinking about conversation UX like any other interface
This means not repeating the same thing over and over (variation)
Erica Hall has a great book on conversational design, we need to do that: we need thought User Experience lead conversational design that looks at the lessons learned by great usability be that iPhones, Amazon “One Click” pay button, or googles search interface, Tinder’s profile importing or NEST smoke alarms
This means having a great interface for the user, but also for the people developing, testing and wireframing conversations
Understanding “Aye-up” is hello, “Youalright” is a statement answered with the same and fifty bucks is different when talking about currency or hunting licences
Some of this is understanding WHO is speaking (which is difficult in a family) and WHY they are speaking (what do they want, intents in ML)
Also means learning emerging words and vocabulary and learning the changes as they comes into usage … “wasssaaaappppppp”
Self-correcting –unguided learning in AI nomenclature-- conversational interface will be key to both making and keeping the interface good.
This will need to be a variety of mechanisms for understanding how words change, new words and saying arise and when the “bricks” change
It will also mean understanding how the context of the same speaker may change when they age and when their interests, goals or circumstances change
And this really is what the job of the conversational platform will be to do: take the different components:
Interoperability
Voice recognition
Natural Language Generation with variety and tone
The ability to build conversations through ‘bricks’ of functionality with good usability – photoshop of conversation!
Store context and confirm context (including knowing who the speaker is)
Design conversations with skill and attention to detail
Get the accents, variations, and vocabulary correct (and keep it correct when language changes!)
Conversational customer service is likely thoing to be a big driver because:
Starting with an ID&V means we can definitively ID customer
We can get contextual information through CRM
The use cases and outcomes are specific
There are clear commercial goals and aims: reduce call wait time, give customer care agents meaningful rather than repetitive tasks, upsell and cross sell
So when we are building a Conversational Platform as a Service these are the elements/experience gaps are bridging to make a truly conversational platform are:
Restful APIs
Good programming/development tools that facilitate rapid development
Solid basis in conversational Text then move on to voice
Get and Keep context
Great UX driven conversational design
Nomenclature, vocabulary and accents right
Self-learning/Self-correcting training so it can grow
I expect that will start to see each of these element come together over the next years or so