This steak is delicious, uhhh… computer. Thanks for listening, sorry about being snippy.
- Chuck, our AI User Persona

Key Takeaways

  • Natural Language Interfaces in commerce are not a fad, they’re an inevitability
  • Conversational Interface Design requires Agile methodology with developer and copywriter teams
  • Connected services, properly orchestrated, increase the quality and ROI of user engagement, especially cognitive API’s

Note: This post is the second in a series around Caffeinated Commerce, our voice interface coffee shop demo powered by Alexa. You can read the first post by Chris Vafiadis here. Ironically as we built ours, Starbucks launched their version (WSJ article).

Designing invisible user interfaces
One of the most interesting fields that is emerging for creatives, commerce experts, and corporations is the notion of designing “invisible” user interfaces. Entertainment, as well as practicality, has made the notion of talking to computers a cultural and technological inevitability. Just as Star Trek predicted the iPad, this favorite clip with Scotty demonstrates the chasm we’re now in the process of crossing.

 

With an estimated 5 million Alexa Echo devices sold, and just over 10,000 Alexa Skills in the store, it is clearly at the ground floor, but has the fabled hockey stick growth pattern every investor dreams of. Investment firm Mizuho projects that Amazon will make about $4 billion from Alexa Echo hardware by 2020, and all of the tech giants are racing to deliver on the promise of conversational interfaces with tools like Siri, Cortana and Google Home.

Just as the Web disrupted businesses everywhere by connecting everything, and mobile further spun the globe by putting 90’s Cray supercomputers in our pockets, the combination of Artificial Intelligence and Natural Language Interfaces is going to create a similarly disruptive situation. 

The sky is falling, again? 
Every new technology under the sun is the “next big thing.” In this case, the situation is different. So many converging technologies are assembling at once that it’s hard to imagine a world without great change. Especially if you subscribe to the notion of the singularity. 

Exploring that notion in a day-to-day scenario drove our Caffeinated Commerce demonstration with Episerver. Connecting Alexa to a Storefront “Coffee Shop” and allowing users to conversationally get their Starbucks fix at the Ascend Conference in Las Vegas by delivering a digital gift card to their phone. 

In this article I’m going to explore the non-technical perspective. The role of the Cognitive Copywriter, AI User Experience Orchestrator, or whatever term ends up being used, for those who work to bridge the gap between logic trees, brand value, and human experience. 

Coder + Copywriter: Natural Language Interpretation and Detecting Intents
The most important element of conversational UI is the ability to determine intents and objects, and correctly act. Anyone who has used Speech-to-Text software over the years is familiar with the, sometimes hilarious, misunderstandings of spoken words that occur as humans and machines interact. 

In designing a user experience, the “engine work” is handled by the developer in the code-level training of the skill to recognize particular terms, their context, what they trigger, and the development of logic trees, but the copywriter also has a critical role in collaborating in development of conversation and logic planning. The “Cognitive Copywriter” needs to leverage their creative, human intuition in building the “what if” scenarios around conversation trees, like writing multiple scripts for a single story. 

In a previous example, we built a text-driven transaction chatbot “AI Santa” using Microsoft’s LUIS natural language API and Bot Framework for Skype and Facebook Messenger, connected to Amazon’s Marketing API. 

Regardless of how intelligent a system is at understanding intents and objects, the interface design is an interactive story in which a user can say literally anything, and requires both technical expertise and creative storytelling and intuition. Scenarios must be mapped-out, tested, and reimagined. The design is as inherently creative as it is technical.

The overarching implication is that for successful voice interface implementations, a developer and copywriter need to work closely together in an Agile workflow with significant, ongoing, iterative user feedback to build the best possible experience. The AI creative team is developer and copywriter. 

Alexa as Barista: Personalizing an AI Coffee Shop Interface

Our goal was to, with the absolute minimum of data, a first name and zip code, deliver a personalized coffee buying conversation with an iPad Mini and an Alexa Echo, running on an Episerver storefront.

Once the account had been created, Alexa took over and the interface became conversational. 

Without exploring the many conversation flows that the system created, here are the key personalization elements in the experience: 

  • Addressed them by name
  • Welcomed them back and asked if they wanted the same thing on a second visit
  • Made a personalized welcome joke based on their state or country of origin to build familiarity
  • Used their ZIP Code to identify product purchase trends for Starbucks coffees in those areas and recommend popular items to better align with what they’re more likely to enjoy
  • Used their ZIP Code to identify differences in temperature between their home and Las Vegas (the location of the demo) to recommend a drink to cool down or warm up, depending
  • Used their ZIP Code to present their local news to read after they checked out
  • Used their ZIP Code to present a song related to their state after they checked out
  • Made a flirty, positive comment about them based on their product choice and reinforce the personal relationship

Net-net, customers entered their basic information, and Alexa spoke to them with a remarkable level of personalized insight, before completing the transaction and delivering the order. 

Chuck the Truck Driver: Two Imagined User Journeys with Natural Language Processing 
But what goes into creating this kind of interface design? In our prototype, we had just two effective variables for a first time user experience – name and zip code. We also had only minimal secondary API’s, none of them cognitive. So, as a thought experiment, here is a scenario playing out with two different approaches to AI interface design with the same imagined user. 

In the first, the design is purely functional. It is what you would get if you plugged an ecommerce platform into a basic Q&A system with natural language processing and basic intent detection on a voice interface like Alexa.

In the second, we’ll see how better design and the benefit of cognitive APIs and CRM enhance customer experience and revenue. I’ve italicized my commentary on both experiences.

Design 1: Purely Functional
Chuck is going to his favorite steak place, where he interacts with an essentially functional voice interface AI that has been placed at his table as an experiment in streamlining the ordering process. 

Alexa: “Welcome to Steak Town how can I help you today?” – Too open ended.

Chuck: “Give me a steak, medium rare.” – “Give me” is a very clear intent. Steak is a very clear product category. Medium rare is a very clear product customization.

Alexa: “What type of steak would you like?” – Narrowing from product category to catalogue item.

Chuck: “T BONE.”

Alexa: “OK, you’d like one T-Bone steak, medium rare. Is that correct?” – Stored “Medium rare” in cache.

Chuck: “Yeah… And give me some fries and don’t skip on the ketchup.” – “Don’t skip” is a colloquialism, logically read it means “include” but to North Americans it means “extra.”

Alexa: “OK, you’d like one T-Bone steak, medium rare, with fries and ketchup. Is that correct?” – Natural Language failure in not understanding colloquialism.

Chuck: “Extra ketchup!” – User frustration.

Alexa: “OK, you’d like one steak, medium rare, with fries and extra ketchup. Is that correct?” - No sentiment analysis has been processed to detect the customer’s emotional state and alter response.

Chuck: “YES!” – User frustration increased.

Alexa: “Wonderful. Your order will be out shortly.” – Intent fully understood and order processed to kitchen. 

Alright! Successful order completed, and look at that increase in productivity! Alas, Chuck is now pissed off at this stupid computer who replaced his friendly human. The computer had no idea he was having a bad day, or that he was a loyal customer. All sorts of bad UX. 

Design 2: Designed Emotionally with Cognitive API’s
Let’s assume that we’ve tricked-out the AI platform with a suite of currently commercially available Cognitive API’s, and have been more thoughtful in how we mapped out potential scenarios with the richer data we have access to. For example, we know what he ordered last time and implemented sentiment analysis.

Alexa: “Welcome back to Steak Town Chuck! Your usual healthy Chicken Salad today?” – Uses Microsoft’s Facial Recognition API, associates image to database, pulls previous order.

Chuck: “Ah heck. No I feel like something different.” – Microsoft emotion detection notes that he looks sad. Can also be detected through voice sentiment analysis.

Alexa: “Well we’re glad you’re here Chuck! Sometimes you need to treat yourself! 15% off, on us. Remember to tell me your whole order, since I’m not always the quickest.” – Adapts statement with positive reinforcement based on “sad” variable. Presents promotional offer in context of helping build relationship with positive emotional connection and associates the discount with “us” including brand and human staff. Reminds user to be specific in stating the order in advance to elicit a result that captures the entire product detail set. Humanizes with self depreciating comment.

Chuck: “Thanks… Uh… Computer. OK, I’ll get this T-Bone steak medium rare with fries and lots of ketchup. Maybe that apple pie too.” – Responds to reminder for detailed product order. Responds to combination of positive interaction and discount with additional item in cart, increasing total order value.

Alexa: “Coming right up boss! The chef is on it. Denise is going to come by in a minute.” – Has an adaptive set of personalized statements based on user personas, in this case, informal. Is reinforcing that humans are still in charge of this place, and has alerted a staff member via the restaurant staff screen system that this customer could use a human touch.

OK, NOW we’ve had an efficient customer service experience:

  • Loyalty maintained through customer identification and CRM
  • Smarter design by priming the customer to fully state the order led to faster intent determination
  • Sentiment analysis and dynamic promotions led to larger cart size
  • Recognized to escalate next interaction to a human
  • We should have given the AI a name

Implications
We are only scratching the surface of conversational interfaces. As technology improves, with increased accessibility of interfaces, intelligence, and connectivity to services, we are on track to a future where the primary interface with our devices is spoken, In its infancy, designing user interfaces requires both development and creative expertise to maximize the benefits for brand and customer experience.  

Sean MacPhedran is a Senior Content Strategist at SMITH

Tags: AI, Caffeinated Commerce, Alexa, Experience Design, Cognitive APIs, User Journeys