But these rants are a critical source of learning. Opportunities are where your complaints are.
So, with that in mind, we’ve collated some of the worst chatbot feedback to be found on Twitter. Brands, pay attention to the lessons they hold.
1. If the user wants a human, give them a human
Chatbots are supposed to ease user frustration by offering instant first-line answers and 24/7 support. But when they’re presented as the only service option, they can have the exact opposite effect.
For anything outside of quickfire FAQ or routine processes, your chatbot (usually) can’t help. And, as this tweet highlights, forcing it on users as their only choice is a great way to trigger frustration.
2. Make your bot easy to bypass
This links to the next lesson: make sure chat users can bypass your bot. This kind of chatbot feedback from @BeckyACampbell (we’ve used one example of many in this vein) is proof that users want a bot opt-out option.
For example, if the user types “transfer”, then transfer the chat to the next available human operator. The same applies to “transfer me”, “human operator”, “speak to a human”, or any other combination of words indicating that the user isn’t happy to proceed with a bot.
You might also consider adding a preferred routing option in your pre-chat survey form. For instance:
|Talk to a bot||Talk to a human:|
|Wait time: 0 minutes||Wait time: 2.5 minutes|
3. Test that your chatbot can cope with volume
Unlike human operators, a chatbot can easily juggle dozens of queries without pause for thought. (Or a need for breaks.) But you still need to test how well your chatbot copes under unusual chat volumes — and set your queue limits and offline rules accordingly.
Otherwise, you open yourself up to a fail like the one above.
4. And test it can handle routine queries, too
This tweet documents Twitter user @Lullykaty asking for a pretty straightforward request: the status of a tracked delivery. Yet the chatbot in question is seemingly incapable of handling that routine query.
Few users would expect a complex, nuanced conversation with a chatbot operator. But if your chatbot can’t handle the routine queries that it’s specifically deployed to handle, then why deploy one at all?
5. Use your chatbot sparingly
A chatbot isn’t suitable for every interaction. They’re highly convenient when used for a specific purpose, in a specific context. (For example, taking/cancelling bookings, checking the status of tickets, answering common help queries, and so on.)
But if you try to wedge a bot into every touchpoint, users will quickly develop fatigue. As Twitter user @Lakshana_95 demonstrates, it’s not a great idea to block every contact option with a bot gatekeeper.
6. Use helpful error messages
Your bot won’t always get it right. It won’t always understand. And when that happens, your error messages should be more illuminating than “sorry”.
As the chatbot feedback from Twitter user @sourabhkt points out, your chatbot should give something more helpful than a repetitive apology. So:
- – Clarify the problem / misunderstanding
- – Remind the user of bot capabilities
- – Offer an escape route
7. Monitor your chatbot service
This should go without saying: but your chatbot should never go mute. There is no excuse for a chatbot not to reply — it’s an automated service.
So, make sure you have monitoring in place for your bot service. You need to know as soon as possible if the bot has become unresponsive, or if there’s been a system failure.
Otherwise, as in this instance tweeted by @Neilmc83, you get frustrated users firing increasingly annoyed messages into a void.
8. Intervene as and when needed
Twitter’s worst chatbot feedback often centres on the issue of standalone chatbots. Simply, standalone chatbots don’t work. (Well.)
Rather, your chatbot should tag-team with your human operators. For example, the bot can fact-find and route the user to the relevant operator. The human employee can then deal with the query quickly and competently.
Your chatbot is a handy tool, but it can’t replace human service. Bots will always need human operators to train them, and to take over for them when conversations are complex.
10. Don’t oversell your bot
As Twitter user @meinardi points out, chatbots aren’t “intelligent”. They’re essentially looking to match keywords or clicked buttons to the most useful pre-defined response.
Which is fine, and which remains a handy and time-saving service for both the user and for support desks. But don’t try to present that simple flow-based chatbot as a flexible AI assistant.
Overstated hype leads to disappointment, disillusionment, and negative chatbot feedback like the above.
11. Don’t spam your bot
Websites are already increasingly annoying to navigate. (Think cookie settings, permission requests, full page pop-ups, a barrage of adverts, auto-playing videos, etc.)
Don’t add your chatbot into that mix.
Forcing your bot down users’ throats makes it seem like just another annoying pop-up, rather than a convenient support option. As this tweet from @anuragsaharoy proves, users don’t want bot spam.
12. Use meaningful triggers
This links to the next lesson from annoyed chatbot feedback via Twitter: use meaningful live chat triggers.
For example, a bot message popping up on every page is unnecessary. (And annoying.) A bot offering proactive assistance to a visitor stuck on a help page, meanwhile, is not.
Other useful times to dynamically trigger a bot invitation include:
- – If the user is repeatedly encountering an error message
- – When the user has missed an available offer or discount
- – If the user has scrolled to the bottom of your ‘contact’ page
13. Use chatbots in your queue
Twitter user @jlengrand spotted a series of messages from “Vance”, purporting to apologise for the wait and claiming to be working the case. Unfortunately, the messages came at every 3 minute interval and were obviously not sent by Vance himself. Cue the annoyed online complaint.
While queue messages are a necessary part of the chat wait experience, and they can (and should) involve bots, you need to do better than empty talk at timed intervals.
For example, your bot could keep users in the loop of their current queue position and estimated wait times. It could take in user details while they wait, so the experience feels faster and the operator can enter the chat with information already to hand. Or, it might even offer some form of queue entertainment.
Just don’t bother with pointless platitudes that are evidently automated.
14. Don’t pass your bot off as a human
Our next piece of bad chatbot feedback shares a similar theme. Namely, don’t try to present a bot as a real human chat operator. Your users won’t appreciate being lied to.
Making a full chatbot disclosure is transparent. It promotes trust. And, importantly, it also sets reasonable expectations as to what the user can achieve from the session.
15. Make other support channels accessible
Last but not least, make sure that your chatbot exists within a broader help desk ecosystem. As this tweet from @hbkahn points out, customers want choice as to how they can access support.
That includes telephone support. It includes live chat (manned by actual humans). It includes email, self-service options, social media, and so on.
Customer service should be accessible, versatile, and integrated into an omnichannel help desk. You’re not going to achieve that by hiding all your support options and forcing users into a bot funnel.
Learning from chatbot feedback
Try searching for “bad chatbot” on Twitter. Or “dumb chatbot”, or “stupid chatbot”, or “chatbot sucks”, or just about any bot slur you can think of.
Customers are fed up with poor bot experiences, and they’re airing that frustration.
This doesn’t mean that bots are a bad idea — far from it, in fact. Bots drive speed, convenience and efficiency across any customer service front line.
But don’t deploy one as a silver bullet. Learn from these common pain points, and proceed with due care.