Facebook’s new chatbots still need work

99 views Leave a comment

Facebook debuted chatbots for Messenger this week to many fanfare, earnest consumers a new approach to correlate with their favorite businesses over a company’s renouned messaging service. The suspicion is that Messenger users could simply “chat” with these automatic program programs to do things like review a latest news from CNN, get a weather news or even go shopping. In practice, however, a bots — during slightest during launch — unsuccessful to live adult to a hype.

Trying to use a bots for elementary tasks — like anticipating out if it would sleet or selling a black shirt — was frustrating, unsatisfactory and eventually distant reduction fit than simply visiting a company’s website itself.

That’s not to downplay a intensity for bots in a prolonged term, or a possibilities supposing by bringing Facebook’s vast bottom of businesses to Messenger where they could better offer business who increasingly use mobile messaging apps while on their smartphones.

However, as a chatbots for Messenger height launches, a bot knowledge leaves many to be desired.

To get a good feel for Facebook’s chatbots, we demoed a 3 “Featured” bots that Facebook is compelling around a Messenger height website: CNN, selling app Spring and continue app Poncho.

Each bot had a “Try it” couple provided, which opens directly on Messenger a discuss interface with a business.

For starters, clicking this couple — a new brief URLs directed during Facebook Page owners (in a m.me format) — merely launched a discuss window; there was no nod content from a business in question. In other words, yet we know that there’s a bot to correlate with, we don’t know how to begin.

Does it need a trigger word or phrase? Can we usually contend “hi?”

Facebook will reportedly shortly residence this problem when it finalizes a rollout of “Messenger Greetings,” that will concede businesses to pass along a note to business when they flog off a discuss session. These greetings could indoctrinate users how to get started regulating a bot. It’s misleading since Facebook wouldn’t have this enabled for a chatbots during launch, though. After all, this whole judgment is new to so many of today’s mobile messaging users who are not aged adequate to remember chatting with IM bots like SmarterChild from behind in their PC days.

As it turns out, not all a bots work a same way.

For example, simply observant “hi” to CNN and Poncho generated an automatic response, though Spring’s bot abandoned me. (I had to Google to find out that a approach to speak to Spring was to contend “go shopping” to it.)

Unfortunately, that was a slightest of my problems with regulating a bots.


Here’s a elementary problem a chatbot could assistance me solve, we thought: I’m in a marketplace for a new shirt. we like a tone black and we don’t wish to spend a lot. Couldn’t Spring’s new selling bot assistance indicate me to some good items?

I didn’t design it to attain in assisting me slight tip shirts by sum like sleeve length or fabric form during this early point, though we during slightest believed it could lift adult a few possibilities.

I was wrong.

After removing a selling event started by a keyword trigger, Spring’s bot interface is easy adequate to use. It asks we a array of questions to slight down what you’re looking for — Men’s or Women’s items?, “Clothing, Shoes, or Accessories?”, etc. — and we click on your response.

After we narrowed it down to tops underneath $75, Spring returned 5 equipment it suspicion we would like.

Why would we like these sold items? we don’t know. Spring knows zero about my selling history, what character of shirt I’m looking for, a color, a arise (work or casual) or anything else.

It returned 3 white shirts, one t-shirt, and a sweater.

Uh, OK.

  1. Screen Shot 2016-04-13 during 3.29.50 PM

  2. Screen Shot 2016-04-13 during 3.30.00 PM

At this point, I’d like to ask a bot for some-more options. But how do we do that?

I attempted “chatting” with a bot, presumption during this point, anything we pronounced would during slightest trigger a new “help” outline that explains how to continue regulating a service.

“I don’t like these” we told it, and was met with silence.

“hi?” we spoke into a void. Nothing happened.

“can we uncover me more?”

Finally, a bot answered, informing me that we could lapse to a options above and click them again to start a routine over. Hopeful that a 5 equipment it returned were a randomized organisation and I’d see 5 new shirts if we steady things, we did usually that.

The bot returned a same five shirts.

Well, maybe we need to be some-more specific, we thought.

“Please uncover some black shirts,” we said.

The bot gave me a couple to a website.

I know what you’re thinking. Well maybe Spring didn’t have any black shirts underneath $75? But it did! The couple forked to Women’s black shirts on a Spring site, and there are some-more than a few from that to choose.

The thing is, it usually would have been easier to revisit Spring’s website directly, rather than dancing by this talkative interface.


Poncho’s snarky continue bot was even worse. Though we got it articulate with a elementary “hi,” a continue reports unsuccessful to broach even a many simple information, like when it would rain.

After removing my plcae and seeking if it could outline me daily continue reports (NOPE!, we said), it told me we could still discuss with it for other information, specifically:

“Feel giveaway to ask me if it’s going to rain, if we need sunglasses or a jacket, or usually contend hello from time to time!”

So we asked: “is it going to rain?”

Poncho answered: “Wet. Warm. Yuck.”

  1. Screen Shot 2016-04-13 during 3.32.53 PM

  2. Screen Shot 2016-04-13 during 3.33.08 PM

I’m sorry, though what….?

Is that ostensible to be a genuine answer?

I attempted to get Poncho to explain — did “wet” meant sleet was coming, then?

“Sorry, you’re going to have to contend that again in Cat. Meow?,” it said.

You’ve got to be teasing me.

“When will it rain?,” we asked. It didn’t understand.

“What time will it rain?,” we attempted again. No go.

“What’s a commission possibility of rain?,” we attempted awkwardly.

“Wet. Warm. Yuck,” a bot repeated.

I threw my phone opposite a room.

No thanks.


Of a 3 Featured bots, CNN’s did better, though still had some issues. It responded to “hi” as well, thankfully, as we still have no idea what you’re ostensible to contend to these things.

It afterwards leads we to collect a news we wish to hear: Top Stories, Stories for we or Ask CNN.

Top Stories is a curated preference of a tip news, and can offer we a outline or a couple to a CNN site to review further. The outline underline is a bit suggestive of Quartz’s standalone news messaging app, and is a good approach to get briefed on a many critical aspects to a day’s news.

Ask CNN lets we mention a subject you’d like to review about, while Stories for we earnings suggestions formed on what you’ve been reading. (Presumably, this will urge over time.)

While CNN’s bot achieved these tasks sincerely well, it still unsuccessful in other areas. For example, since it’s been automatic to demeanour for topics by keyword, apparently, it didn’t seem to grasp a shade of a query seeking for “U.S. news.”

It’s trustworthy that someone competence cruise to ask for U.S. news around this bot, in sequence to get a filtered preference of a day’s tip news. However, CNN’s bot usually seemed to demeanour for a word “U.S.” in a title when returning a suggestions. Instead of “top” stories, a bot returned accurately dual articles:

CNN’s “U.S. mobster’s heirs to Cuba: You owe us — so compensate up” and a CNN Money video entitled “Ken Burns: Trump divides us. Sanders does too.”

Not usually are these not what one would cruise a tip U.S. stories, a second was returned since it had a word “us” in a headline.

Maybe we didn’t word it correctly?

Oddly enough, seeking usually for “us news?” returned a opposite set of (now three) stories with a word “us” in a headline, including an op-ed called “What Easter teaches us about heaven.”

Similarly, requests for a “latest news” stumbled — again, usually looking for a keyword in a headline.

This was not useful during all.

Do bots have a future?

While these tests were brief, it goes to uncover that Messenger’s chatbots are still a prolonged approach from fulfilling their intensity to be a useful approach to correlate with businesses and brands by chat.

These bots clearly need to be means to interact with their tellurian business some-more conversationally, and have a softened bargain of what people are indeed observant to them. They need to have some-more functionality — like a ability to lift adult specific equipment when selling — and they need to be softened automatic from a start.

Some bots are softened than others, of course.

The Guardian gave different chatbots a dry run and had similarly churned results. Simple bots like The WSJ and CNN’s news bots did better, while Operator’s selling bot — that claims to respond to emoji queries — didn’t really work. Meanwhile, Sequel Stories’ “choose your possess adventure”-style story bot was some-more fun, their news found.

The doubt for Facebook’s business users is either or not it’s value a time and bid to build one of these bots, or if a chatbot proviso will finish adult being a fad.

For now, it’s still wholly unproven that business will want to discuss with bots — generally when they work so poorly. The experience, if not softened in brief order, could turn a mobile homogeneous of punching your approach by a business’s automatic patron use hotline.

At a finish of a day, people might wish to discuss with a person, not a machine, when perplexing to get patron service.

They may think that regulating a company’s website or local app is easier and faster than perplexing to get a bot to respond properly, so they simply omit a bots.

And if people try bots and destroy to find them useful right out of a gate, there’s a good possibility businesses won’t get a second chance to win behind those same customers in a future.

“SIGH” we wrote to Spring’s chatbot, finale my knowledge today.

The bot pronounced it got my note and would be behind with me shortly. (Spring says a bots are partly human-powered. Maybe my outline was being sent to staff?, we wondered.)

Over an hour later, Spring asked me if we wanted to keep shopping.

Not around bot, we don’t think.