Falling For The Wrong GUI · Meydad Marzan

Winners and Losers in The Conversational UI Landscape

Two hands frame a city skyline in a phone camera while another hand holds up an instant Polaroid photo of a person standing in the same park.

In the past year or so, we have witnessed the emergence of Conversational UI. It was always there, but recently more and more companies are adopting it as if it were the next best thing. Content is The King, but it is nothing without its Context Queen. Channeling services and content carelessly through a trendy UI might hurt businesses a lot in the cruel app-world of no-second-impression-to-make. It happened with apps “tinderized” for no good reason, and it might happen again with “conversationalized” apps.

New to Conversational UI? Here’s a quick intro

Conversational User Interface is an interface that uses a back and forth conversation as the model for interaction. Human types or says something to a machine, and the machine responds back with actionable information in a conversational manner in text, voice or both. This post deals with the visual side of Conversational UI and the way it is mis/used by different apps and services.

Messaging Apps As Infrastructure

We all use messaging apps as Whatsapp, Slack, Telegram and many more, to talk to our human friends and co-workers (C2C). In the past year, many messaging apps opened their gates to companies, establishing person-to-company communication (B2C).

The idea behind it is clear — reach people in the channels they already massively use most of the time, instead of fighting a hopeless battle to drive them to niche, standalone apps they infrequently visit. For some companies this transition works beautifully, but for some it’s a hollow, superfluous and dangerous gimmick.

Matching GUI to Channel Type and Mindset

Diagram (1): a two-by-two map. The horizontal axis runs from Standalone to Parasite apps, the vertical axis from a “Select-one-of-many” mindset to a “Deal-with-one at-a-time” mindset. Meekan Scheduler via Slack and Uber via Facebook Messenger sit in the Probable Success corner (parasite, one at a time); Quartz News, Hyper Travel and Operator Ecommerce sit in the Probable Fail corner (standalone, one of many); Payso Conversational Banking, and Buzzfeed on Line, Blynk on Kik and Forbes Bot on Telegram, sit in the two question-mark corners.

Matching GUI to Channel Type

We can distinguish between two types of apps on the channel type dimension in this landscape:

Standalone apps that require an individual download (e.g Quartz news standalone app)
Parasite apps that appear as “channels” or “bots” within existing messaging apps (e.g Forbes Bot on Telegram)

As discussed before and as can be inferred from the diagram below (2) — going with the parasite model is inevitable as 71% of US Smartphone users use only 1–6 apps on a daily basis. Setting standalone apps in the risky part of the diagram above (1).

Statista bar chart: average number of apps used per day by US smartphone users, August 2015 (2,011 respondents; source: Millward Brown). None: about 1%. 1–3 apps: about 28%. 4–6 apps: about 43%. 7–10 apps: about 20%. More than 10 apps: about 8%. — Got an idea for an app? Think again (Diagram 2)

Matching GUI to Mindset

On the mindset dimension the range is somewhat wider and more diverse, but I chose to divide apps using Conversational UI into two as well:

Select one of many mindset — Where the app is serving the user with big sets of information to scan through, and enables drill-down when the user sees something relevant. This model can be found in informational news apps and ecommerce apps, where our minds are trained for quick “sort and find” in a rich and complex stream of data items.
Deal with one at a time mindset — Where the user deals with one item at a time and usually tries to complete a single task every use cycle. This functional model can be found in apps like Uber, Google Calendar and Navigation apps, where our entire attention and cognitive capacity are focused in the task completion.

Side by side. Left: the New York Times app with a scrolling list of headlines, labeled “New York Times App: Consume a lot at once. Quick scan and select.” Right: Waze turn-by-turn navigation showing the next turn onto Portland St, labeled “Waze Navigation App: Complete one task at a time. Richness of data is measured and allowed only if assists the mother task.”

Conversational UI has two inherent characteristics that need to be taken into consideration before implementing:

Information from both sides flows in small bits, one bit at a time. Unlike search results for example, where one bit is followed by a long tail of items as response.
“Action-Reaction” model that collects and filters data through casual conversation and not through visual controllers with bigger and faster impact on the data shown to the user.

These two characteristics make Conversational UI perfect for products like Meekan that aims at solving the unwanted back and forth we experience when trying to set a meeting with other people. Meekan is essentially a bot with access to the team members calendars, from which it can infer what is the best time for all participants trying to set a meeting. Genius.

Fits perfectly to the mindset of a conversational interaction, and operates by riding on other apps as Slack, Hipchat and more. Setting Meekan in the probable winners section in diagram (1).

Slack on desktop and mobile with the Meekan Scheduling Assistant bot posting meeting invitations: it notes that your calendar shows you're available, offers to RSVP if you say accept, decline or maybe, and then confirms “Invitation accepted.” Labeled “Meekan: One of many friends and co-workers; Uses the conversation interaction naturally.”

Common Is Common For a Reason

It simply works. This is me trying to buy a white shirt on “Spring”, a regular ecommerce app vs “Operator”, a conversational commerce app:

Side by side. Left: the Spring app showing a grid of white shirts with prices and filters, labeled “Spring App: Plethora of options in a heart beat.” Right: a long Operator chat in which an agent asks for size, fit and price range before suggesting a single Topman shirt, labeled “Operator App: Sweating through the process.” — Conversational commerce is exhausting

And trying to book a flight with “Kayak” vs “Hyper”:

Left: Kayak's flight search form (Boston to San Francisco, dates, one traveler) and a results list of United flights at $824, labeled “Kayak App: Although the required input is the same, the GUI controllers feel faster than manually typing the trip description.” Right: a Hyper concierge chat that asks for travel dates and preferred departure times over several days, labeled “Hyper App: Conversational trip planning.”

I admit — I started this micro research biased against Conversational UI, but it makes a whole lot of sense in some cases. I am super curious to see how this space evolves. Especially for vocal conversational UI (e.g Siri, Cortana, Alexa, Viv). If you are interested too, or have anything to contribute or new apps to share, put it in the comments to this post or in the (awesome) Twitter hashtag #convcomm created by Chris Messina.