Investor’s Guide to Chatbots
By Christopher Steiner
More than one experienced tech observer has crowned some form of chat the next great technology platform. There have been many slide decks that have portrayed the evolution of tech thusly: PCs -> the web -> mobile + apps -> chat + bots. While it remains to be seen if bots will become the next consequential technology platform, the excitement around the space is based on the reality that billions of people have migrated much of their mobile-based attention to chat applications such asWhatsApp, WeChat, Line, Slack, and Facebook Messenger. If people are going to live on chat platforms, the thinking goes, then developers — and investors — will follow them. There has already been a scrum to get teams funded and applications up and ready for a shift that some people expect to be as big as that of PC-based web to mobile.
With this momentum established, startup investors may have already seen an influx of pitches from teams who are focused on the chat space. Due to its relatively short stint in the spotlight, not all investors may be familiar with the space, its virtues and its pitfalls. With that in mind, we’ve prepared this guide on things investors should understand about chatbots and the realm of applications that they may eventually enable.
That anatomy of a bot that chats
Chatbots are applications whose user interface is largely based in chat, and, in most cases, listen in the background of popular chat programs such as WeChat and Facebook Messenger, popping in when they’re called by name, hash, or some other activator. One simple scenario of how that could work: A user would simply query a travel bot named TribBob by typing, “@tripbob, what is the price of a round-trip ticket from Chicago to Boston Sept. 23 to 28?” The TripBob bot is queried with the user’s request when the chat platform recognizes the bot’s handle. TripBob’s backend then parses the user’s text to pull out the travel request, returning results to the user right inside of the chat application in the form of text, a card or even a webview.
The perception among chatbot enthusiasts is that most people would prefer to quickly dash off a note to an @Instacart bot — “Please deliver orange LaCroix and Claussen pickles to house now” — than pulling up the app, searching for LaCroix and pickles, and submitting the order.
The point of any application like this is to find a route to users that offers less friction than the usual on-boarding path through a traditional mobile application, which requires people to download an app, a major conversion at the beginning of the product-user relationship that must happen before any other interactions can take place. And even following that conversion, 90% of downloaded apps only get used once. With a chatbot, there is no app that needs to be installed, no major commitment required from the user before the application can be queried — and it can stay relevant by living within a chat application that most users open half a dozen times a day or more.
“If consumers prefer to use messaging and have the ability to use messaging, why not enable them to do so?” says Will Dawoodi, cofounder at Init.ai, which builds AI chat technology used in bots.
Investors approaching this space should understand some simple tenets, and know that the chat marketplace and ecosystem remains largely undeveloped and, as of yet, undetermined.
Chatbots aren’t new — only the idea that they can do everything is new.
Many people, although not many millennials, will remember chatbots from the 1980s that ran in the MS-DOS command line and referenced a collection of ‘if else’ statements to settle on a response to random human requests and musings. Similar variations on that kind of chatbot manifested throughout the last few decades on all manners of computing platforms. The responses they produced may have become snappier and more varied, but the guts of these applications largely worked in the same rote fashion.
What is rather new is the idea that chatbots can field and actually understand a wide range of words and grammar construction using natural language processing, and in turn use that ability to abet all manners of transactions. The growing pool of developers with natural language processing and machine learning experience, and the advances in the field during the last several years have helped intensify the focus on chatbots and the perceived simplicity of interaction they may bring.
The chatbot concept appeals to us because people generally prefer interactions that mimic the cadence and rapport of human speech. Just as the best writers capture a conversational tone in their prose, the best applications find a way to cross raw utility with intuitive interfaces. Chatbots, it can then be argued, offer the most intuitive interface of all, giving users the chance to type or simply say exactly what it is they seek, be it information, a product, a service, or a channel of communication with another human.
At the same time, some human behavioral experts warn that chatbots, with limited language understanding and range of feedback, won’t provide a satisfactory UI for many people, as human agency, our ability to exercise free will and make choices from available options, often demands more. Many like browsing their options in as informative a way as possible. Instacart co-founder Max Mullen points out that his company started as mobile-only, as that was the trend at the time, but due to customer demand Instacart quickly built a web interface so consumers could browse groceries on their PCs.
“Humans take agency in many ways, and they will not always want to select the only option available through a computerized decision tree,” explains Sally Applin, a Silicon Valley-based Ph.D. at University of Kent, Canterbury in the Centre for Social Anthropology and Computing. “Anytime systems are too rigid, they will break. If a chatbot does not have a very flexible range of options for humans, they won’t use it, and won’t like it.”
The Platform Battle
One of the largest questions with chatbots: what will ultimately become the dominant environment in which they’re used? Chatbots need a place to live. They need a front-end interface supplied by a platform with that already reaches a critical mass of users. Facebook’s Messenger platform, at least in the United States, has a good positioning in this space right now. But Google and Apple, because they control the operating systems of the mobile world, could put their own solutions forward, building on the relative successes of Siri for Apple, and Google Now for Google. Slack may have an edge in the realm of the nascent field of enterprise bots, as Slack has a ready platform upon which thousands of companies share and define proprietary missions and communications between their employees.
In some ways the coming chatbot wars, if they are indeed coming, may mimic that of content distribution, where key social platforms: Facebook and Twitter, along with Google’s Pagerank algorithm, have a firm grip on who and what gets a voice on the web. In the battle of chatbot platforms, the winners won’t only decide the content that you see, but also how your words will be parsed, and who can answer your words, spoken or typed, and in what ways.
Acquiring Users Takes A Different path
One fundamental question that startups focused on chatbots should have to answer: how are you going to acquire users? This is of course the real crux for any startup in any business, but most of the standard solutions that startups employ against this problem don’t work as well in chat. Unlike standard web advertising, ads driving users to websites or through deep links to app stores won’t necessarily push them closer to conversion when that process has to take place within chat. And startups don’t only have to acquire the users, they also have to wrest them from interfaces belonging to competitors — web, mobile apps — and get them onto a new medium, chat, with a new vendor. For a chat startup in the travel space, a user who is used to booking plane tickets on the web with Expedia will have to be convinced to book tickets in chat, and with a company other than Expedia.
An existing large company like Walmart can easily tell its millions of users about its new chatbot, available on Messenger, by flashing alerts on its website, in emails and on its mobile applications. There exists an organic path for established ecommerce companies like this to push users toward its bots. But for new startups whose primary product exists in a chatbot, the challenge of acquisition and conversion is far different.
Chat platforms are aware of the conundrum. Facebook, when it opened up Messenger to developers’ chatbots, it also released what it calls Messenger Links and Messenger Codes, which allow deep links that can be used on the web, on email, or in SMS that route the user directly to a conversation in Messenger with, one would assume, the entity that created the code. Facebook is trying to move many of the notifications that currently live on SMS — like when your pizza guy arrives at your door or when your Uber is pulling up outside — to its own platform, so it’s also supplied developers with APIs that match user profiles to existing data that companies might possess to reach these customers through Messenger.
So startups do have some tools at hand to get users directly in front of their chatbots — and that’s something that investors should ensure that founders clearly understand. But innovation and creativity on this front will be key. In marketing their bots, startups need a comprehensive plan for acquisition that plies new techniques.
The additional danger for chatbots is the lack of branding available in what is largely a text-only interface. This may make re-activating and retaining users harder than normal, as there’s less to stop somebody from querying the TripJoe bot instead of the TripBob bot.
Chatbots probably won’t work for everything
The best applications of chatbots will be those that feature interactions without a lot of ambiguity. Booking a flight has become the classic example that many people use when describing chatbot-borne commerce, which makes sense, because it’s a larger transaction that only requires a few data points to fully describe it: origin, destination, airline, dates and times. But, as Felicia Schneiderhan, CEO of 30SecondsToFly, points out, booking a hotel room, something that normal travel sites like Hipmunk and Orbitz try to do in tandem with flight purchases, is much harder in a chat format because hotel rooms are subject to a whole range of subjective differentiators, such as location, interior design and cuisine and because people usually pore over pictures of the place before they book.
Schneiderhan’s team has created a chatbot named Claire that can book flights, hotels, cars and anything else a business traveler might need while on the road — but even she sees the difficulty chatbots will have in tackling more nuanced commerce exchanges and tasks that a human can easily dispatch. “If variability of user preferences and options is substantial, it is very difficult to accurately predict what the user wants, even with large amounts of data,” she says. “The other question is about the complexity of the task that the chatbot is supposed to perform. How many different intents does the chatbot need to recognize? The more limited the number of intents, the more sense it makes to transfer a task to a chatbot. Many people underestimate how many intents seemingly simple tasks require. Scheduling meetings, for instance, involves approximately 20 intents.”
By that, Schneiderhan means that chat offers a narrow conduit for information. On a mobile phone, where much of users’ interactions with chat takes place, the screen is small, and the platform in use, be it Facebook’s Messenger or Slack, doesn’t give the user the same richness and intensity of information that the web can give. Shopping for shoes, where the shopper requires many inputs and likely prefers to see many different options at once, would be a challenging transaction to carry out through a chatbot. Not impossible, but not intuitive at this point for most users. If the buyer knows exactly what shoes he desires, say a replacement pair of Bass boat shoes in light brown, then a chatbot can work well. It’s the process of discovery, which requires something of a information firehose for users, that can be awkward within a chat format.
Investor Outlook Summary
As with any space, investors looking at chatbots can employ different investment strategies. Rather than trying to pick a winning app, for instance, investors could look for startups that have focused on building backends, services and tools for chatbots — much like a Heroku or a Digital Ocean enables applications to be deployed to the web with little focus on servers or stack management.
Investors examining startups in this space may also consider looking for companies that have built — or seek to build — backends that are largely platform-agnostic, so that the startup can easily deploy its bot on mediums to which users migrate in the future. This is a fairly simple concept in computer science, and REST APIs are platform agnostic by default, but investors should be sure that founders aren’t betting too heavily on one platform over another, as the space remains ill-defined.
Startup investors who know the space and feel strongly about one platform, be it Facebook’s Messenger, Slack, or another, might look for opportunities to go all-in on that platform, perhaps by backing a company that is building an analytics suite for apps on that platform, or one that has decided to focus its app-building efforts on that one platform.
In all cases, investors should seek out founders who grasp the virtues and limitations of chat and have developed novel solutions built with those understandings in mind. If a startup says they’re going to be a dominant player in the chatbot space, they should be prepared to offer up plans that show their understanding of the current state of the art in chat, show that they understand human agency, the challenges of platform uncertainty, and how they will acquire and keep users.
Originally published at fundersclub.com on August 1, 2016.