Why? Because you get payback in a month? Because you probably don’t know your customers as well as they do? Because you need to understand digital conversations? With bots, the old slogan of “start small, scale fast” that normally relates to innovation is truly applicable and, in my opinion, the safest – and probably normally the only - way to ultimately succeed.
Why? Because there simply are many unknowns, and learning is part of the journey. Just ask yourself this:
- Does your organization have experience in designing successful digital customer conversations?
- Do you know what scenarios your customers actually are prepared to address with a bot?
- Do you have experience in the cost of integrating a bot with your backend systems?
These are questions that all have fundamental impact on a business case, on scope and on decisions to be made moving forward. Despite that, many organizations tend to base their decisions based on conference room guesswork. Albeit insightful, in principle still guesswork.
I have talked to customers who spent months analyzing which bot package to use, while other customers instead learnt from a pilot. The difference? On of the two I am thinking of still did not have anything running after those three months while the other had had trained the bot using customer data and had a 94% accurate response rate on the questions answered.
Collecting data and learning should be part of the plan to build a bot, and the plan itself have room for adjustments from the findings. These are classical agile development principles, that make a huge difference, and easily are applied to bot building.
So why an FAQ?
I have found that it is hard to predict what your customers will ask the bot. Yes, you can draw general conclusions, as that gentlemen above the age of 50 tend to use a bot as if it was Google, speaking to it in one-word sentences, or that younger people tend to chat with a bot as they would with any friend on any messaging platform. However, neither of those examples help with the actual questions and utterances your bot will be exposed to.
The point is to involve customers early and to get exposed to real customer issues, not conference room guesswork. The conference room activities tend to involve costly external resources that could have been used on resolving issues people using the bot actually have – if they were known. Get an FAQ bot up and running and start collecting experience! Your FAQ bot can be operational in a few days. It will provide you with valuable insights in a few days – for free. That is exactly what an FAQ bot does – apart from the obvious of answering the questions it gets. You can start learning what your customers actually discuss with a bot and make your first improvements within the first week of a project. Learning from real customer interactions, doing it frequently and learning again as new findings are made, is a known fundamental pillar in successful innovation work. Working with bots, the reason why becomes very tangible: With as many unknowns as we see in bot conversation development, the process is very similar to innovation and successful innovation methods relevant and applicable.
Apply and improve
Understanding how to improve how the bot interacts with people, learning from mistakes in a contained environment and apply changes quickly, is what transforms a bot from your average FAQ machine to a great experience. As with any successful software development, understanding the technology, mixing tech and business skills is important to create solutions effectively. But with bots, where you can apply a fix within minutes and release new functionality weekly, the growth of benefits is so much faster when done right. Therefore, basing decisions on facts and experience in your own environment will make a larger difference than in many other projects. The fast release cycles usually surprise the business, which is used to a slower development phase and rare interactions with development. With bots, the business can drive much of the improvement without even involving IT. Simply act on the feedback the bot is gathering!
The tangible benefits
Getting feedback from an FAQ bot is simple – just log the questions. In a bot context, this has several uses:
- You get a measurement of the quality of the FAQ. Over time you will see the bot resolving a continuously increasing part of the incoming requests.
- You get a log of everything you had no idea people would ask. Herein lies the real value.
Of course, there are savings in automating the questions you already know the bot would get asked. But – and this is a big “but” – in the beginning the bot will lack the skills to get many questions right. So there has to be focus on training the bot to learn, because even if you use logs from customer service and get the staff to document their most frequent questions, you may end up with not having answers to about half of the questions people are asking. People have fantasy. They ask questions in different ways, or don’t even ask what they really want until after a while. So, it does take work before the obvious value in automating frequent questions is realized, but usually this is a matter of spending a few hours per week training the bot.
This still means that with a very limited effort you reduce workload on customer service within weeks. But that is not the most important value in my opinion. Yes, it does save work, but the interesting result, are all the questions nobody predicted. These questions or requests from customers could be more complex than what can be solved with a simple question/answer pair. To be able to answer a customer, more development might be required, whether it is e.g. because you need to ask more questions to understand what the right answer is or because you need to connect to other systems to get to the information asked for. This is not necessarily complex, but definitely a lot more complex that adding a question/answer pair to your data and retraining your model. So, what the bot gives you is the frequency with which those questions are asked, which in a sense is a direct figure answering what the return on the investment is on every feature requested by a customer. And the sooner you get that FAQ bot out there, the sooner you will learn. Setting up a pilot costs less than a day of hired consulting help. You could get started in a week. Are you set on getting far ahead, or will you be watching while others do?