The Message Is The Medium: Reasons ‘Assistants-as-App’ Work

On a typical day, I’ll chat with colleagues on Slack. Later, I’m sure to receive a message from a friend on WhatsApp or Facebook Messenger. Then, on my way home, I’ll use good old SMS to let my wife know I’m on my way.

What’s been absent from these conversations is commerce. Although messaging is the way users communicate with each other, it’s not how they interact with businesses. That is, until now.

A new wave of startups are radically simplifying the way they communicate with users and the trend is just getting started. In my last post I discussed how I’ve personally used some of these services. In this post I’ll go deeper into what this trend is all about.


Although terms like “conversational commerce” and “invisible apps” have floated around the web recently, neither is quite right for describing what I’m seeing.

Instead, I propose “assistant-as-app” to mean: an interface designed to enable users to accomplish complex tasks through a natural dialogue with an assistant.

Note that the assistant does not have to be a real person per se. The assistant could be an artificial intelligence or a human sending automated messages interspersed in the stream of the conversation. It could also be a group of people interacting with customers through an online persona.

For example, Mindy is there for me when I use Vida Health to track what I eat and Tim helps me book travel on Native. However, as far as I know, Mindy and Tim could just be avatars I message when I need something.

The Conversation is the Interface

The interface used by assistant-as-app services is an effective way to help people get more out of the Internet. Since it looks like a chat, interactions feel familiar. As I mentioned in a previous post, people don’t want something truly new, they want the familiar done differently (see the California Roll Rule).

The power of the conversational interface is that it shields the end user from having to learn anything new. We already know how to chat, so making requests is easy. An assistant-as-app leverages well-trained human assistants to use complex technology behind the scenes. The assistant can process requests that would otherwise require several steps, complex analysis, or pro tools the layperson is unlikely to have the knowledge or patience to use.

Better Than Bots

But who needs humans? Isn’t voice recognition powered by artificial intelligence enough? Not really and not yet.

Just last week I found myself desperately trying to talk to a human when calling my credit card company. After several failed attempts to get through the automated voice response, I called out my request in a mechanical, over-articulated, robot-sounding voice. “REP-REE-ZENT-A-TIV,” I said, sounding like Robby the Robot.

“OK, let me get someone to help you with that,” the automated voice finally responded, although I wasn’t sure if by “that” she meant my credit card problem or my weirdo-who-speaks-like-a-robot problem. Although talking in a robot voice is perhaps a humorous example, it makes the point that users still have to structure their requests in a way machines understand.

Today, there are primarily three solutions. Either route users to a fully automated prompts (like an annoying call routing robot), send them to human helpers, or (the most common solution) ask them to fend for themselves on the company’s website.

However, as A.I. becomes more capable, human-curated technology delivered through a conversation interface will provide the best of both worlds. Though the cost of hiring a human assistant has been out of reach to-date, the costs will soon plummet as the number of users served per assistant increases.

What is it Good For?

An assistant-as-app is suited for certain scenarios. Here are three criteria where I expect assistant-as-app services to excel:

1. When There’s One Goal and Too Many Options

According to a recent survey by Carat, a global media agency, “41% of people feel overwhelmed by the wealth of choices on the web, making it hard for them to make purchase decisions. Meanwhile 26% of people feel there is so much information on the net that it is hard for them to find what they are looking for when shopping online.”

When booking a flight for example, customers have only one objective — to find the best deal. They don’t need so many dizzying options, they just need one, as long as it’s the right one. Before travelers started booking by themselves online, a good travel agent could help narrow down the choices. But today, we’re on our own.

When we stopped calling travel agents, we implicitly chose cheaper tickets over better service. However, with an assistants-as-app, the consumer no longer has to compromise.

It should be noted that an assistant-as-app is not ideal for situations when the user enjoys browsing or where the best option is subjective. For example, for many people, clothes shopping itself is part of the fun. Being told the best choice may not be all that helpful.

However, there are plenty of opportunities to use an assistant-as-app where the user has to weed through too many choices, particularly in enterprise applications. I envision a future where complex tasks, like running a marketing campaign to increase site traffic or launching a coupon offer, are executed by an assistant. Instead of asking the busy user to navigate a complicated interface, an assistant-as-app will do the heavy lifting and offer up just a few of the best options.

2. When Data Collection is Easy but Analysis is Hard

An assistant-as-app is particularly good at off-loading the burden of analysis. For example, the Vida app leverages dietitians to help users diagnose food sensitivities and allergies. With Vida, the user simply has to take pictures of their food before each meal. Then the assistant compares what was eaten to how the user felt, looking for what may be causing the adverse reaction. This sort of analysis is a burden for the user but is easy for a skilled agent.

The assistant becomes even more powerful when she, he, or it, can access disparate sources of information. Processing data is a headache for users but is an opportunity for assistant-as-app companies.

For enterprises swimming in data, an assistant-as-app could be a godsend. Imagine an assistant continually optimizing your website. Instead of hiring someone with these rare talents, specially trained assistants could use the latest tools to continually run tests to increase conversion rates. All the testing and number crunching would get handled by the assistant. The site owner would just need to provide access, approve the tests, and ok subsequent roll-outs of successful changes.

Imagine the collective sigh of relief from busy workers who no longer need to learn how to use yet another vendor’s optimization tools. With an assistant-as-app, just ask and ye shall receive.

In addition to complex interfaces, assistant-as-app services are particularly well-suited for small screens. Texting is fine on phones, but as smart watches cross the chasm into mainstream adoption, users need a better way to interact with companies through their wrist-worn devices. Reciting your request to an assistant-as-app through the watch in plain English is easy.

3. When it Feels Like a Friend

Working with an assistant through a conversational interface should feel like interacting with a friend. These apps work best when the user trusts the assistant’s unbiased recommendations.

However, if a friend was chatting you up to make a buck, you’d quickly see through the scheme. Similarly, an assistant-as-app is best suited for subscription models where the value lies in being an objective filter. If Native started recommending specific hotels over others based on earning a commission or if Vida began hawking vitamins, I’d quickly lose trust.

Another friend-like characteristic of an assistant-as-app has to do with the pace of the interaction. When sending a friend a text, you’d expect to wait a while before they respond. Likewise, conversational apps are for when you need something soon but not immediately. Since culling the right options, running test, or analyzing data takes time, assistants are not well suited to provide instant feedback.

Google and Apple are already working on perfecting virtual assistants like Siri, made of 100% computer code. However, fully automated voice technologies are good for specific scenarios — namely, when the user needs immediate information for simple queries. An assistant-as-app excels when a request still requires a human touch.

Although improvements in artificial intelligence will help assistant-as-app services reduce costs over time, a highly adept human assistant working with increasingly sophisticated technology, will be the way consumers interact with an array of services in the years to come.

Note: Also, thank you to Jonathan Libov for commenting on previous versions of this essay.