Creating a Successful CUI Strategy

If you are waiting at an airport lounge - the most common phrase you hear is "Hey Siri". As CUI (Conversational UI) is gaining tremendous popularity via the messaging, chat platforms and chat-bots, the voice assistants like - Siri/Cortana/Echo/Home are taking this to a whole new level. I have a feeling that a few more innovations will come on CUI very soon. As ML/AI/NLP and related technologies mature and Big Data churns out even bigger data - the CUI will gain strength. Enterprises are looking at CUI to take over support, buying, selling, recommendations, directions – pretty much every possible use case. Many Apps and services are looking into integrating CUI. The drivers are mostly both cost and accuracy improvements. As an enterprise takes its journey towards a CUI enabled interaction – it gets deeply integrated with AI/ML/NLP and Big Data. Since all these technologies are still evolving – choosing the right strategy for implementing a CUI is critical. At the same time - it’s very challenging to create a comprehensive and future-ready CUI game-plan. The biggest issue I see is - most enterprises are looking at CUI implementations just as a traditional UI layer sitting on top or the usual service/data layers - that's a big mistake.

Any enterprise that is looking into implementing a CUI needs to ensure that all the data providers and the acting systems are ready. This might need a certain amount of redesigning for the line-of-business and supporting systems as well. Here are a few recommendations for a winning CUI strategy:

1.     Choose the use cases carefully

Most of the implementations of CUIs I see currently are around task management. They have been limited to automation of mundane low-value human tasks - 'find this', 'do that' type things. CUI is much more than that and has to mature (rather quickly) beyond these use cases. CUI is not a glorified speech-to-keyboard mechanism - it’s more than that. Asking my digital assistant to "tell me where my package delivery is" –or “tell me when my next meeting is” saves me just 1 tap/click - not a big deal for me. But, if I can tell the App “Please take care of my morning schedule” and the App then looks at the morning traffic and realizes that my first meeting is at risk and hence updates that meeting to a 'tentative' status with an explanation sent to the meeting organizer (and subsequently find the next free slot and book it) – that will be a real "assistant". But, implementing this complex actions driven use cases which interface with multiple systems is challenging and expensive. When you implement CUI to your App/services – you will typically find a lot of possible use cases with varying complexity and cost. Map all your potential CUI use cases in a 4-quadrant value/impact vs cost analysis and then decide your priority.

Fig 1: Analyzing the Use Cases

2.     Review the Data Strategy thoroughly

Most of the enterprises today have a data store at the bottom of the stack. The services which are built on top of the data layer manipulate that data and feeds to all the upstream systems. Then a GUI works by aggregating the data from the services and then presenting it (after some rearranging). But a CUI deals with information (and decision, action) rather than data and that has to have much deeper integration within the entire stack. As Data moves through the layers of Big Data/Analytics - it transforms into insight, then decisions and then to actions and thus become more ‘intelligent’ – hence Artificial Intelligence. Keep in mind this evolution and transformation of data and then design the CUI strategy around the insights, decisions and actions rather than the old-style data.

Fig 2: Transformation of data with AI/ML

3.     Redesign services to make them Conversation ready

In majority of the CUI implementations, only the UI is being converted to CUI and the rest of the stack still stays old style. Today most CUIs break down the 'conversation' into smaller tasks - fire services and upon getting the structured data back - they reply back. The data that comes back from most of the services in an enterprise are way too structured than a normal conversational interaction will ever be. Rather than structured services returning data from relational or no-sql type databases, if the 'conversation' mode stays across the stack - that will make things more accurate. Most CUIs force me to ask kindergarten style questions – example - "find me the nearest pizza place" - which is a very unnatural way of restaurant discovery. They do this because the query that goes to the service (and then to the database) and the data structure that comes back is by design very structured at the moment. This is probably also the reason why most of the time my Cortana just does a Bing search and displays the results - that in my view is not CUI.

On the other hand, a human - while feeling hungry - would see the nearby restaurants - go back and find out his/her previous experiences with each of them, find out if any friends ever recommended it and then look for discount coupons - this process is iterative and well, "conversational". I believe some of them (like Siri) does combine multiple data streams, takes into account the users' preference/history etc. - but, still the experience is yet to be 'human like'. So, a full review of the source and format of all data and their services that feed the CUIs is definitely needed. Do recognize the importance in this and be open to redesign some of the service stacks.

 4.     Recognize the Machine in the middle

The CUI has a very unique problem when it comes to its most important aspect - the conversation. In a natural human interaction - it's a human to human (H2H) conversation. But, in case of a CUI - there is a machine in the middle (rather a lot of them). So, the conversation breaks down like this:

• Human to machine
• Machine to another machine (repeated a few times)
• Machine to human

To the previous point that the 'conversation' does not stay across the stack - makes this more complex as the sentiment of the conversation can get lost as it travels across the stacks and in most cases what finally comes out is - pure data and tasks. With every layer there is some degree of breaking down, translation and transformation of the conversation that happens. This makes the CUI with a machine way more complex than a human interaction. I think this area will be the slowest to evolve and catch up with the rest of the technology stack.

Fig 3: This is how the conversation gets lost (found on the Internet)

5.     AI is still too artificial to replace human intelligence

 CUI is also deeply related to the AI/ML/NLP and related algorithms and technologies. Let's analyze one use case to see how CUI and AI need to work together to process a lot of inputs/data. The use case of ‘getting direction’ is a very old and common scenario - In the 'human way' I stop my car at a roadside shop or a gas station and ask for directions. First, there are usually multiple people there - so, I get a collective input. Plus, I get to find out the nuances - how wide is the road, is that road drivable if it's raining, is there a restaurant somewhere nearby? Well, in the machine world - I use a maps CUI I have to ask 10 or more specific questions to get enough intel. So, when the Maps CUI is trying to have a 'conversation' with me to give me directions with a CUI it has to pay attention to these details. Designing the little nuances around the interactions, and conversations decide the effectiveness of a CUI. This is probably the most important factor that makes a CUI successful.

And finally, in a CUI – the ‘C’ (conversation) is more important than the UI. Do not create complex UIs – just keep it simple – let the conversation be the hero – the UI is there just to support it. CUI is definitely the future of human-to-machine interaction and will slowly replace a lot of traditional UIs and use cases. But, while implementing a CUI strategy – enterprise architects need to be careful and take note of the pitfalls. A successful CUI strategy will no doubt have a lot of positive impacts on the business.