Video and Slides
Outline: How Not to Build a Chatbot
Muzzamel Mazidee, Director of Partnerships UIB
We try to understand, from the consumer’s perspective, why chatbots fail to deliver a good experience and what thought processes need to be considered when venturing into such deployments.
Presentation Review
You can ask Mel questions in the comments section of this weblog, or contact him directly, his info is in the presentation.
UIB is an omnichannel messaging service platform with APIs for ISVs, SIs, chatbot builders, and developers. In Asia, there is a diverse range of popular messaging platforms: WhatsApp, Facebook Messenger, LINE, Apple Business Chat, Google’s Business Messages, Telegram, etc. I’ve known UIB since Tony Ruckert founded the company.
One of TADSummit’s policies is “no BS,” and this presentation is a great example of just that. While many presentations hype chatbots as the greatest customer service innovation ever, it simply does not jive with our everyday experience. Mel’s presentation lifts the lid on why that’s the case and helps us understand the steps towards success with chatbots.
Mel says what many of us think, “chatbots are stupid!” We’ve all experienced a bot trying to force us to a particular resolution when that is not what we want. And when we say that, it asks us another question, and still forces us to that resolution. Or worse, sends us to a link to read/watch a video that is irrelevant to resolving our issue, and just wastes 10 or 20 minutes of our time.
It’s also important to realize the bots we experience with Siri, Alexa, and OK Google are exceptional. They’re state of the art, while most chatbots used by enterprises are less sophisticated.
Mel highlights the importance of using the right communication channels in Asia, i.e., the bot has to be on the customers’/users’ preferred platform. Whether that’s WhatsApp, LINE, FB, etc.
Bot applications include FAQ, IoT control, automation, sales, and HR processes. In the early days of bots, internal applications focused on HR, a sort of HR FAQs. It was a safe way to gain experience with bots and not impact the customer. However, some HR bot implementations were rarely used, as the user volume and user benefits were slight.
The drivers for bots are lowering costs, increasing revenues (lead qualification), improving accuracy/efficiency, and delighting customers. I know the last point given what we’ve said so far may seem incongruous. As an example, Amazon delivered a package to me with nothing in it. I contacted them, told the bot what had happened, and in less than a minute of interaction a new shipment was on its way and this time did have batteries in the package. Bots can delight, but many do not.
Mel reviews a couple of use cases including du’s introduction of a WhatsApp-powered virtual assistant to interact with customers. It successfully reduced workload in its call centres and stores and found that more than 50% of customer inquiries were being successfully resolved by the bot. WhatsApp has been the dominant messaging channel in the UAE for years.
When Mel runs through what bot implementations get wrong, this is spot-on and excellent insight:
- Users’ needs. If the bot cannot help the customer get to want they want faster / easier than what you’re doing today. Don’t do it, as it will frustrate the customer. If you’re going to send your customers to watch a video for 10 minutes, make sure this will resolve their problem. After 10 minutes and an unresolved problem, the customer is now furious at your brand/business for wasting their time.
- KPIs, tracking, and ROI. It takes time to fine-tune the bot to the specific application — test, test, test. Without metrics, the project cannot develop towards delivering business results and customer satisfaction.
- Users’ favorite channels and preferred languages. I cannot overstate the importance of this. Asia is the most diverse region for channels and languages.
- It’s “dumb.” The bot can’t handle multiple intents/languages.
- It’s not used. Some of the initial HR and FAQ bots suffered from this as the user volumes were low and/or user benefits slight. If it doesn’t benefit the customer, they will not use it.
I love this quote he uses at the end, “One reason a chatbot takes more effort than either self-service or a human agent is that compared to us humans, it’s stupid.”
Thank you Mel for one of the best bot implementation guides I’ve ever seen.
Thanks Mel for the excellent insights on how not to build a chatbot 🙂 I have some questions:
1) What KPIs do you recommend tracking for chatbot projects?
2) What are some of the excellent or unique chatbots your Reseller Partners have implemented?
3) Some brands are concerned about the impact of using social media for customer communications. Facebook’s reputation on customer privacy is frequently discussed. What do you recommend here?
4) One of the benefits you highlighted is improved accuracy/efficiency. Would you please give some examples?
5) If everyone uses WhatsApp, why should chatbots use more than just one channel?
6) What do you estimate is the typical ROI for a chatbot project? And how long do ‘typical’ chatbot projects take to implement (a range is OK)?
Hi Alan! Just answering your questions below
1) What KPIs do you recommend tracking for chatbot projects?
It depends on how certain business units involved will measure it, but here are a few examples;
a) Customer issue resolution (both success rates/time)
b) Customer retention (repeated engagement/success)
c) Accuracy of the questions trained vs unanswered questions
d) Channel usage (which channel frequently used by customers)
e) Usage/engagement (how many chats across x amount of days/weeks/months)
f) Lead conversion (how many customers qualified on certain sales oriented process)
There can be a few more but I believe these are vital to show the effectiveness of the deployment
2) What are some of the excellent or unique chatbots your Reseller Partners have implemented?
Ah! So we have multiple reseller partners who have developed excellent bots so far for the likes of Prime Bank Bangladesh, NGO like Care SG, and even in the toys and collectibles side with Ichiban Kuji (Bandai)
3) Some brands are concerned about the impact of using social media for customer communications. Facebook’s reputation on customer privacy is frequently discussed. What do you recommend here?
This has been buzzing around but to clear the air, these social media companies are sticking with their policy of protecting the user’s privacy, therefore end-to-end encryption, not tracking your location/chat logs & content . In the instance when talking with a registered business on these channels, companies would need to know what kind of information they need to analyze when the customers are reaching out to them to enquire about their services to deliver a better experience.
4) One of the benefits you highlighted is improved accuracy/efficiency. Would you please give some examples?
Most people when they start rolling out a bot, they can only assume what a customer would ask, but can you imagine hundreds and thousands of people using the bot and asking a lot of questions but for the same purpose? It’s important that we’re able to track all of these inputs to determine the intent of the user and train it to ensure the next time a user comes in and chat with the bot, the frequency of the previous users will attribute to the accuracy of the bot’s ability to understand the intent and deliver the response quickly rather than saying “I’m sorry, I do not understand what you’re saying, would you like to talk to a human?”
5) If everyone uses WhatsApp, why should chatbots use more than just one channel?
If you look geographically, WhatsApp may be only dominant in certain regions but that leaves a lot more countries to explore which has the likes of iMessage, Facebook Messenger, Viver, LINE, WeChat in which you should cater to.
Also another scenario is that, some people prefer to use a certain messaging channel for personal use and have another messaging channel for business use. It’s important that we give them the flexibility to choose how they want to communicate with us.
6) What do you estimate is the typical ROI for a chatbot project? And how long do ‘typical’ chatbot projects take to implement (a range is OK)?
Depends on how you want to define it, some companies would look at the significant savings to the customer support/call centers they’re able to achieve across an x amount of months/years through automation. And when I say savings in this aspect, it could either mean reducing the talents in the field or shifting them to a different task that could maximize the operations.
With automation also gives the opportunity to convert sales efficiently, depending on how the service is dealt with. Can you imagine if a bot can handle a quick sales process of a thousand customers compared to one salesperson handling a thousand customers at the same time? I mean this is just an example but I think you catch my drift.
As for the timeline for implementation, some can be within a week’s time if they know what they want and have done a study, some would take between 2-8 weeks. It really varies on the requirements and business case, especially on how big of a brain do you want the chatbot to have.