This is exactly the kind of thing I was thinking of. Use the data channel to help support better understand exactly what the issue is from the first message. By knowing what transaction triggered the request, give the support agent hints with relevant documentation on merchant issues or quirks. If a user taps “I don’t recognise this”, support should be given the exact transaction details up front and told that the user is suspecting fraud along with possibly quick links to relevant fraud policies and procedures.
Coincidentally Slack’s most recent blog post was about the importance of the human touch when dealing with customer queries. If anyone was going to implement bot support, on the face of it, Slack seems like one of the most obvious candidates but
When you’re receiving similar inquiries, it’s tempting to create canned responses. DON’T DO IT. Unless you’re capturing a set of instructions that are difficult to remember, I wouldn’t recommend substituting macros for your own writing. It’s obvious when you’ve copied an automated response, rather than writing the note yourself — canned responses lack personality, empathy, and individuality.
At Slack, we’ve created a plethora of macros! However, they all consist of the bare bones information agents need to address issues. None of them are intended to be copied and immediately sent as a response. We’ll use a macro as a starting point, which will then be crafted into a personalized response. Craftsmanship reflects that you’re a human who cares about their work, so be sure to craft each response for your customers.
Canned support responses when you think you’re talking to a human are the worst.
That doesn’t mean all automated support should be off the table though, just that expectations should be set and met. If a customer is clear that they are using a self-serve support system, that’s fine but if you ask to be put through to a human, you better be put through to one who can genuinely chat to you, understand the issue and have the power to resolve or escalate potentially complex problems.
Haha! Richard, you just beat me to this…
Thanks for sharing, Alex. Nice addition to this thread. The blog post highlights the importance of customer expectation.
If you contact customer support expecting a human response but receive a canned reply from a bot that is definitely bad. But if you’re waiting in a queue due to scalability issues, an immediate reply would be better as long as it’s clear that it’s automated. Additionally, adding personality / humour to the automated response can help to increase customer satisfaction.
Plain, one optioned and non personal canned response are one the single most off putting methods of customer support I have ever come across, especially when up to that point you think you’re chatting to a human.
Your reply did give me the idea though, that it could be quite effective to generate a variety of responses (the same answer worded differently) for the most common questions. Also have what I call variable fields which essentially allow you, or the bot, to pass in a variable that is personalised to the user from their data. For instance, throw in a remark about or an example of one of their transactions relevant to a hypothetical query of theirs; as well their name. This would make it feel a lot more personal, even though it’s still informing you it’s a bot.
You could have them worded in different tones (always positive, but some empathetic, some funny and some generally polite) depending on their attitude leading up to the response. If we wanna go really super sci-fi, then have the bot guess the appropriate tone of response for you by gauging their attitude. Not sure on the current level of intelligence of natural language and emotional interpretation by machine learning but it’s interesting. I’d expect that’s a lot less likely though.
Hey Ben, sorry I meant to say earlier I agree with you on this
Also a good point. Repetitiveness can be annoying, I recently watched an Amazon Echo demo in which a user was demonstrating Alexa’s home automation skills in different rooms of a house. All very impressive, except that Alexa replied with the exact same word “okay” in the exact same tone every time and after a few times it started to feel impersonal. Definitely yes to personalisation!
Check out the BBC article above about Luvo (powered by IBM Watson Conversation):
In future, Luvo may be able to understand if a customer was feeling frustrated or unhappy and change its tone and actions accordingly, IBM said.
I’ve just come across this story -
which has a few more interesting results in it, which feed into the discussion about bots & how Monzo manages it’s customer service -
More than four in five (83%) believe speaking with a person will always be an important part of the customer service equation.
(Shouldn’t be a problem, I’m sure Monzo will always have phone support), particularly because -
As customer service requests become complex, reliance on human interaction increases. More than a third of customers prefer to go in-store (34%) for complex enquiries, while another third prefer to connect by phone (33%).
In terms of preferred digital customer service channels, 22% of consumers want access to an online account … and 9% cited that they prefer to connect using mobile apps.
I expect the favorability of mobile apps will increase as tools like Intercom become more deeply integrated into apps (push notifications with new messages ftw).
And lastly, this is useful context for any future Monzo customer satisfaction surveys -
A quarter of respondents would give a positive review, and almost a fifth (18%) would renew products or services, even if they aren’t the least expensive option. This compares to 21% of those who would write a positive review and just 13% who would renew products or services following good customer service on digital channels.
It’s also worth noting the implications of the 4 in 5 humans prefer humans to bots response for Atom since -
though the quality of support that the bot is able to provide will obviously be a big factor in overall satisfaction.
Another downside of almost every chatbot is this,
Your customers don’t want to tailor their text to fit your bot’s limited capabilities. To stand out in an increasingly crowded field, modern chatbots need to synthesize language like a human — that means understanding slang and idioms, correctly interpreting ambiguous phrases, and connecting related concepts to deliver helpful responses.
* Taken from the marketing website of a company that has developed a chatbot. You can see them here.
Adding to the momentum behind chatbots…
To add some context, I think it’s fairly significant that Twitter - currently under a lot of pressure to speed up user growth / increase revenue - has been investing it’s time developing this product.
They must think there’s some real demand for this, although their track record for product development based on demand isn’t exactly perfect.
Folks should checkout https://getchip.uk really nice on-boarding via chat, seamless handoff into live help chat, and a good example of how a conversational UI can be useful and fun. It’s a shame it’s wrapped in an app framework for now - but the potential is there to leverage via other channels as people start to trust the channels more. If you sign-up use my code Q103AA
We’ve been aware of Chip for a while now and good news is they’re integrating with Monzo, but not the chat feature.
Cool - what does their integration with Monzo currently stretch to though? Currently they use Saltedge to scrape the last 90days of your transactions to perform the analytics. If Monzo had a stable API I guess they could offer that as a link option in their ‘connect’ service (which would be vastly preferred over the Yodlee/Saltedge based approaches) but I’d think this would be at least mid to late 2017 before Monzo’s API is stable enough to leverage.
Currently there is no live integration. I’ve summarised at the end of the thread linked above.
Turns out chatbots are in Monzo’s plans
Screenshot from Monzo’s Crowdcube investment deck.
that is a real shame
They talked about it on one of the podcasts. They want to have 1 support person per 100,000 customers by automating as much as possible to speed up helping customers. If they do it well it’ll be worth it.
Still, that’s an impressive stat -
We currently field about 5000 queries per week, 99% of which come via in-app text chat. In early trials, our AI has been able to identify 80% of questions correctly.
If the technology continues to evolve at the current rate, then the experience could be a lot better than what we’re seeing now…
I think chatbots can be fine as long as they’re 1. smart, 2. easy to get past to a person.
Auto-suggesting an action (or relevant knowledge-base article) from the first message is acceptable as long as you don’t get trapped in some sort of gibberish-based conversation of little use.
If the intent can be guessed 80% of the time then it seems it will likely be acceptable.
I read your blog post about the machine learning going on behind article suggestions in the support section. Quick question: since Intercom is developing just that (and you’re already using them), is there any point in duplicating their core capability yourself? Should you not just focus on being a financial hub for everyone and delegate AI support to them?
I guess it would depend on whether what Intercom is developing is any good. It seems the path Monzo are taking is to only rely on third-parties where they have to.