My chatbot is dead – Why yours should probably be too
azumbrunnen.me
azumbrunnen.me
That attention disruption got so bad I blocked Intercom and their ilk via injected CSS rules.
I really want to take it a step further and write a plug in to interface a chat bot with the human on the other end and waste their time so sites begin to learn to stop using that BS.
Actually, the ones I really despise with a burning passion are the ones like above, but which follow you around the whole site like some kind of demented Clippy 2.0 - you close the popup and navigate to another page and... BAM! It's back, again!
https://coworker.com has this annoying banner at the moment advertising their global pass.
Nice X to close. I refresh page and it's back. If I clear my cookies or my adblock causes a javascript error, then fair enough. Otherwise, persist my decisions please.
#hello_banner { display:none !important; }
.grecaptcha-badge { display: none !important; }
For a next version of this sketch they could borrow some inspiration here: https://darkpatterns.org/
As developers who never need help, we tend to forget that some people might do, and that a Chatbox that annoys us can help them.
Anything that pops up and interrupts me while I'm reading something else is pretty much a deal breaker. Having to stop what I'm doing to hit a little X and get back to it has cost a bunch of companies business from me.
Having said all that, there are a couple of companies that use them in the correct way. The one that springs to mind is xolo.io. Xolo is a one stop shop for my business (expenses, invoicing, tax accounting) and has chat staffed by humans and it's a fantastic way to interact with them (as an existing customer).
I message them, and assuming it's within business hours*, I get a rapid response and answer to my question / solution to my problem.
The supermajority here does, or so I would guess.
We aren’t representative, and that’s important to remember when designing user interfaces.
A favourite story[1] (and I can't remember where I read it first) is the WWII bombers that the US military were trying to understand how to protect by analyzing the bullet holes. But the bullet holes only told part of the story. Any bomber that made it home to be analyzed had only been hit in a non-critical place.
I'm not saying there isn't a place for chat, because I believe that they're absolutely is. A good, direct to the person chat, or unobtrusive ability to reach out is great.
Shoving it in my face and breaking the flow of what I was doing is not.
While I agree that HN readers aren't representative, I have a hard time believing that other people don't behave the same way when presented with annoying website behaviours (even if they later want the contact with the human).
I've seen behind the scene of businesses that get customers asking stupid stupid questions, so I also understand the where the zendesk 'are you sure this doesn't answer your question' comes from. Doesn't make it any less annoying when applied badly to expert audiences.
I like the WW2 story as well, but when chat increases total interactions and conversions, it’s hard to make a case for survivorship bias.
For example, chat popups that ask you to interact before you've had a chance to even read the title of the page.
Every customer who hasn't purchased yet is part of a funnel, and you absolutely CAN test with an A/B test (or equivalent) whether you're losing people when you annoy them by popping something up on the screen.
And perhaps I'm an outlier. If the A/B test says you get more sales when you annoy the customer, who am I to argue (or it annoys me and not them). But it does mean you will have them as a customer and not me, and that's fine.
I don't think I'm wrong to use this as one of many indicators of how annoying a SaaS is going to be to deal with over the long term.
In context of UI annoyances and regular people, it's also worth remembering that the representative (i.e. not tech-savvy) population accepts what it's given. They don't have the mental models to conceptualize what exactly is it that annoys them, and they don't have a voice to express that annoyance. But they are annoyed, and you can hear that when you actually talk to them. The frequently heard things like "Why did they have to rearrange everything again?", "Google 'xyz contact form' because it's easier than finding it on the xyz site", or "My computer is slow it's probably viruses" are symptoms of that.
this. chatbots are significantly less than what they're hyped up to be.
I think one of the most regularly overlooked areas in which people may need help is through accessibility features. I'm not neurodiverse myself, but things that pop-up and animate can be problematic for those that are.
Having the option there may be useful, but the intrusive pop-ups that the previous comment referred to do not seem useful to me, and could even greater confound the experience for groups of people. I'd expect the intrusive features are mostly driven by conversion rates rather than usability.
I've found that most of these chatbots will use sound alerts to draw attention, and Cape's environmental toolkit definitely references these. While a lot of the content in this toolkit is for physical environments, a lot of it will transfer to digital product UX in terms of stimuli etc.
I have an ASD diagnosis. This is the first I've heard of "condition". Neat! Is this to move away from "disorder" because of stigma/negative connotation?
Popups of all kinds, I despise. There should NEVER be any unsolicited modals. That includes chat boxes, "please subscribe to our newsletter", GDPR warnings, and all of them. They all interrupt my attention flow severely. If you're lucky, I will inspect your DOM and will block your modal with injected CSS/JS rules, reload, and keep browsing your site. If you're unlucky, I'll leave and look for other competing sites that are a more pleasant experience. I have clicked out of several sites before that had a stupid "can i help you" Intercom popup. On a few of those occasions I actually replied "no go away" before leaving the site, hoping that they would take the feedback for the future.
I find that stores which do this tend to be more interested in making a sale than ensuring you have something you're happy with and want/need.
What should we expect from sites which behave like this?
Even then, too many sites replaced their point of contact with these trivial keyword kind of chat bot interfaces and it's annoying as heck.
I tried to report a phishing page hosted on some random SaaS thing yesterday. Instead of a contact form / abuse contact e-mail in the page footer I ended up on a FAQ page telling me to report it using their chat thing. Of course that chat was not staffed without any indication until you typed up a message (another frequent factor of annoyance for those not in US-adjacent time zones) and the keyword matcher asked me if the FAQ page I was on would be of any help. When I put the data in the thing to go on with my day it asked me for an e-mail to follow-up with me, something this interaction did not require.
Sure, I get when people get more customer interaction from these things but please folks, make sure it's not replacing functionality for other parts of your (potential) customer base.
And I can agree with you. There is actually nearly always a customer segment that finds chatbots and popups totally annoying. This usually ends up in this segment leaving the site disproportionately due to the modals, which leads to a loss of revenue.
At the same time there is a segment (usually much larger) that appreciates chatbots and FAQ modals - even on the first visit. Here these functionalities lead to a significant increase in sales.
I think that it is quite difficult to automatically assign the user to the appropriate segment and adjust the behavior of the website accordingly. Unfortunately.
Du such popups annoy me and I belong in the same segment as you? Absolutely. Am I an empiricist and do I have to advise most of my clients on such functionalities after appropriate tests because they boost sales? Also absolutely.
And on the question of GDPR modals, I'm afraid I have to say that they are an absolute legal necessity. Unfortunately, the way they have been implemented in most cases I know of is designed to put the user behind the scenes and make him accept more tracking/cookies than he would like to. These things annoy me so much and should be banned from the net.
I block cookies from all sites by default, and make exceptions for the few that I need to log into. The result is the GDPR modals keep popping up because the modals can't set cookies to remember whether or not I accepted.
So the only thing I can do is inject JS/CSS to block them.
(Also by de-modalizing it and hiding it, I never actually need to accept anything, and nobody gets my permission to track, hee hee.)
They absolutely are not a legal necessity if you don't do things you shouldn't be doing. That's the point of GDPR :).
But of course doing these things, putting up the popups, and then trying to direct users' anger at Brussels is empirically the revenue-maximizing strategy, so here we are.
Hah. You expect sites to not track you just because you are there for the first time? If that were the case there’d probably be no GDPR.
The unfortunate result is that the goddamn GDPR warnings keep popping up, because ironically, the way those websites remember that you either "accepted" or "denied" the tracking request is to leave a cookie, and I don't allow or want to allow that cookie.
EDIT: Separately, there are many websites that only have an "accept" button and no "decline" button. By blocking the modal dialog boxes, I never actually "accepted", and therefore they never got my permission to track. Ha!
GDPR is about a lot more than just cookies.
>By blocking the modal dialog boxes, I never actually "accepted", and therefore they never got my permission to track. Ha!
Unfortunately they probably treat this as if you accepted the modal (which is against GDPR).
Sure, but they can't track me anyway :)
GDPR is about all tracking - cookies are just one implementation.
What a strange and petty complaint.
Agree. Over the last few years, I started running into those more and more often as well. From an UX standpoint I would categorize them somewhere between marque/blink tags and rotating GIF icons.
Actually, that was roughly the time from which I recall first running into those things on corporate websites (IIRC Yello Strom in Germany had one in the early 2000s, widely know for a particular Easter egg).
I find it kind of funny that those are having a revival in recent years, along with animated GIFs (abused as a video format this time round) and annoying overuse of the word "cyber".
Snark: Apparently anyone can call themselves UX/UI experts nowadays. I guess I shouldn't worry about getting qualifications and just slap that label onto my CV...
It's quite disappointing that it took a Google UX design lead to come up with something like this.
But now when I realize that this was made from someone at Google's UX team, it all makes sense. Interacting with this site feels exactly like interacting with the typical Google property, where all standards and expectations gets thrown out for something that probably looks pretty on the designers screen but is 80% off for everyone else.
Or better, the click would cast a vote for the candidate the developer prefers...
That's the problem, though. Motivated humans can weaponize anything. We spend too much time trying to stave off the impossible, instead of going after the abusers - which today often are the presumably legitimate businesses, as a lot of feature misuse became standard business practice.
It's really weird that people actually spend hours and days on implementations like this because they and are seemingly the only people that actually find enjoyment in it. If you were to talk to your friends about it they would probably say "it's cool" but many of us are still going to say "Why?"
On an iPad, the site shows it's own context sensitive cursor below the iPadOS cursor, for extra weirdness.
Although hiding the real mouse cursor probably shouldn't be allowed without a permission dialog imo.
I would have thought it to be OK for a website to pick from a few standard system cursors (pointer, hand, I-bar, move, resize, etc.) but not no cursor.
If a website made your cursor disappear immediately upon full screen that would be freaky, as if you lost control of your machine, thus, I don't think the websites should have this capability.
It's almost as bad as websites customizing the CSS or scroll bars. Don't do that.
Everyone dreads call menus / phone trees - chatbots are largely the same except there are easier & better UX alternatives in an online medium.
I ended up googling and found out a secret keyword to talk to an agent who one can directly say after choosing a language. I never have been connected faster to a human when calling a large companys hotline.
The audio from those calls is fed to a team of people who are responsible for teaching the system, so that the next model can be more accurate in directing calls.
I once met someone who had been working that gig while in university. Apparently the vast majority of the recordings are boring and easy enough for a human to classify, but some are just people rattling off swear words, screaming, etc.
USAA, for example, makes it abundantly clear that you’re talking to a robot, before beginning the conversation and provides multiple choice options for common support flows. I’m actually a big fan of theirs for things like travel notifications or ATM withdrawal limit increases.
Instead of navigating through 3 links to reach the form, having to search the newly loaded page each time, I answer 3 questions.
In truth, I just type “ATM withdrawal limit” and the form begins, because it’s so obvious what I’m trying to do.
Chat dialog interfaces can be cool as a way to piggyback a service through existing chat services. I used to love goog411 back in the SMS era. But I don’t want everything I do with a computer reduced down to a fake chat with a robot.
The point is to eliminate the need to work to find stuff.
We have sophisticated data inputs that we've developed for decades. Tons of usability and accessibility thought has gone into things like how someone can select a date, or input a number. Throwing all that away for a dumb NLP chatbot because "Wow! AI!" is a shame. Tack on a glorified search interface that looks like a support rep with an IQ of like 20 if you like, but please don't funnel me into it when I'm already on your site or app and I've purposefully selected a menu item like "close my account."
I can't entirely agree with the points mentioned in this article.
Yes, a chatbot is not ready to replace human support.
However, there are other areas where they show good results.
We have been helping customers with lead generation chatbots for the last four years.
What we have learnt is customers would instead engage with a chatbot than filling up lengthy forms. Because chatbot nudges users to give information.
Also in websites where you have a lot of information to convey, ex: banks, mortgages, etc. again chatbot makes sense. Because it helps give information that's selective to the customer. And they don't have to struggle through a lot of information presented to them at once.
We have customers who use our chatbot link directly on google ads.
We have also noticed our customers maturing over time and demanding more functionalities on the chatbots.
To summarize, no chatbots are not dead. They are only beginning. However, you need to narrow down on a specific problem and solve it.
Yes, there will likely be more chatbots in the future. And yes, THEN they will be usable. But no, it's not a good user experience yet. If a website needs a chatbot they:
- Didn't put enough effort in organizing the website
- Don't want to spend money on real support
- Hate their customers
So there is always the temptation to save costs of your support with a pre-filter, so that your support can talk to people who actually have a question or problem that needs to be talked about.
It's the equivalent of "have you tried turning it off and on again?". This question must be asked, because it actually resolves a substantial portion of the issues - but it does seem quite wasteful to employ people to whom 30% of the calls ends with asking that question. Or, for other scenarios, reading verbatim one of the top answers from your FAQ page.
Nothing wrong with this. You can say this about practically any company using software for anything. Not spending money on people is the point of enterprise software.
Yep, I agree. However, even in the wake of GTP-3, maybe it is time to abandon the term Chatbot. Essentially "chatbots" are a tool to deal with unstructured data/content. At least this is how I see them: a better search interface.
As opposed to a call center in an impoverished country staffed with subsistence wage workers with an average tenure in the job of something like 6 months or less?
Companies talk a lot about CX and CRM until it's time to spend money on the front line workers who actually interact with the customers.
It's the digital version of the "automated phone guidance thing". The "press 1 for sales, press 2 for accounts, press 9 to kill yourself" !
I think many of the "smaller" websites has been sold a lie ! I see "silly/useless bots" that only makes the customer mad instead of helping him.
Just to show I'm not all hate and brimstone :) I found adding a "manned whatsApp" to your website is gold !
Are you running the Dark Reader extension by chance?
I think the author didn't test with Dark Reader, which is fair, but they override the cursor, which is not fair. Please don't do that, it's not as cool as you think it is and it breaks things for your users.
look at this,
:root {
...
cursor: none !important
}
html {
...
cursor: none !important
}
* {
cursor: url(cursor.19bbb4e3.png), none !important [0]
}
#cursor {
background: rgba(200, 200, 255, .2);
transition-property: width, height, border-radius, transform;
transition-timing-function: ease-out;
transition-duration: .1s;
pointer-events: none;
transform: translate(-50%, -50%);
border-radius: 100%;
position: absolute;
left: -1000px;
z-index: 99999;
width: 36px;
height: 36px
}
He kinda messed up, mayb?
ChatBOXES on the other hand are amazing, a few reasons why:
- You can't accidentally hang up and have to re-dial and wait on hold
- There's an incentive for the company to respond while the customer is there, because they might disappear at any moment (unlike email)
- There's a record in writing
- It works with people with poor language skills
- No hold music & you can see where you are in the queue
- Support teams can scale by handling multiple conversations at once and using copy/paste from scripts, terms etc.
- You can include hyperlinks, screenshots etc. to make things super clear or open up other tools (e.g. screen share)
- Contact details, email address etc. can't be misheard
- No bad signal / fuzzy line
- You can data mine the requests to figure out common requests and optimize for them
Long live the chatbox!
The tools to make good chatbots are complex, but they exist, and in the hands of someone who is trained in all of the requisite areas can be wielded to create something useful. It just costs money most companies don't believe is worth spending.
As I travel to the appropriate link, there is none such ability or option. I relay the appropriate message to their customer service. Fast forward a day later, they were unable to reply to my message in time, they have already processed the order, and are actually begging me to accept the purchase because as a small company they are incapable of managing their customer service department.
All in all, I felt bad that they would have lost a substantial amount on my cancellation or refund. But I can't help but feel the entire situation could have been avoided if they had eschewed the "benefit" of "automating" their customer service.
I think a good reason Salesforce has been so successful is that they force users to abide by the one-reply system of following up on customers.
Fully agreed with that. While for sure "boring" for many and not good as marketing tools, the best "chat bots" IMHO basically are inline commandline tools that help specific communities.
In my experience that's exactly what they are; you end up second-guessing the machine and thinking "what do I need to type to get where I need to be". It's like playing one of those old text based adventure games.
I can't remember a single encounter with a chatbot where a good search interface wouldn't have got me there faster.
I still think that chatbots provide an entirely new way of interacting with software. We're very used to dividing our desktop up into windows, and giving each process it's own two-dimensional real estate.
Conversation-oriented UI design fundamentally is different to that: you are escaping the 2D arrangement and allowing applications to collaborate with teams of people in a totally new way.
Here is something I wrote about that: https://riskfirst.org/complexity/Betting-On-RPA
The base commands are things like !help, !about, !coin (simulates a coin toss), and then I added a little NLP on top to translate to one of those commands. I feel like that works a lot better than a general conversational bot that most likely isn't gonna respond in meaningful ways to most queries.
It would be fair to rule those out and look at more conversational chatbots that provide good guidance through form filling, understand natural language and maybe are even hybrids that can be taken over by humans when they fail themselves.
A simple form of the hybrids is some pre-processing. The bot says "hello" and "whats your problem" or identifies your account and than human support staff jumps right in to the action.
But a chatbot which actually answers my questions would be great. I hope what we are seeing is the Through of Disillusionment[1] and soon will get support chatbots that are actually helpful.
[1] https://en.wikipedia.org/wiki/Hype_cycle#/media/File:Gartner...
To sum up my experience with chatbots: I've had deeper conversations with clippy than with most chatbots.
Still, I strongly believe that guidance is needed online, especially in E-Commerce. That's why I've recently (in fact, we are signing our seed round later today) created Neocom [1], which merges the good parts about chatbots with real expert knowledge.
Till date I haven’t seen one chatbot that has been really helpful. It’s like the Internet version of a phone tree where you struggle to get to a human who may be able to help.
Fast forward to 2020, and you cannot throw a rock with closed eyes without hitting some kind of "computer". Smartphones, tablets, laptops, everywhere. There is not just one factor that made this change come about but a lot: arguably, computers are generally more useful today than they were back in the day, and this is part capability and part usability.
Both seem to still lack quite a bit in today's chatbots. I think a big hurdle has already been taken, though, by turning to voice rather than typing as the primary input modality. This wasn't possible just a decade ago because automatic speech recognition just didn't work well enough. But there's still lots to be desired today.
Using a conversational UI where a simple button or a few clicks would suffice is utterly useless. As is, say, having your bot read out to you a long list of movies when you ask "what's showing in the theater tonight"? Traditional GUIs are much superior here.
No, voice based interactions shine when they are about complex matters. Stuff that is too difficult or too cumbersome to express with other means. Perhaps because there a lot of different options combined in your query. Or because some inference is required on the chatbot side. Or because it's not a single query at all but -- shock! -- an actual dialog that will get you to your goal.
A small and completely made-up example:
User: I'm meeting Dana for dinner at 7, and perhaps we'd like to go and see a movie afterwards. What's on tonight?
System: Where are you having dinner?
User: In the West End.
System: Super Marvels is playing in the Broadway at 9.
User: Oh, no, Dana hates super hero movies.
System: I see. What does she like?
User: I don't know - comedies, I guess. Thrillers, maybe.
System: Well, the Broadway is also showing Dark Waters at 9.
Compared to a dialog like the above, today's conversational UI's are very simplistic. We're still with at least one leg in the stone ages. But there is real potential here but there's also a real risk that we'll never get there because once conversational UIs get a bad rep, then it's over.
Like with any technology, if it is over-hyped with respect to what it can actually deliver (currently), it won't last. I think that is what the author of the above article has experiences. High expectations, low delivery.
I, for one, hope that we can hold on and that conversational UI's won't go the way of 3D TV (goodbye...) or VR (niche), but at this point, I wouldn't know which way to put my money on.
Information is power and these chat interfaces will sooner or later prey on the less informed.
Why is replacing badly structured content with low quality NLP an improvement to the user experience?